repo
stringlengths
2
99
file
stringlengths
13
225
code
stringlengths
0
18.3M
file_length
int64
0
18.3M
avg_line_length
float64
0
1.36M
max_line_length
int64
0
4.26M
extension_type
stringclasses
1 value
chainer
chainer-master/chainer/training/extensions/warmup_shift.py
from __future__ import division from chainer.training import extension class WarmupShift(extension.Extension): """Trainer extension to gradually initialize an optimizer attribute. This extension changes an optimizer attribute evenly at the beginning of one training. For example, suppose that this extension is called at every iteration, and warmup_start = x , init = y, warmup_iter = t. Then this extension will set the corresponding attribute to from ``x`` to ``y`` evenly in first ``t`` iterations. This extension is also called before the training loop starts by default. Args: attr (str): Name of the optimizer attribute to adjust. warmup_start (float): the value of the attr at the beginning of one training. init (float): the value of the attr after warm up iterations. warmup_iter (int): the number of the iterations in which the attr changes from ``warmup_start`` to ``init``. optimizer (~chainer.Optimizer): Target optimizer object. If it is None, the main optimizer of the trainer is used. """ def __init__(self, attr, warmup_start, warmup_iter, init, optimizer=None): self._attr = attr self._warmup_start = warmup_start self._warmup_iter = warmup_iter self._init = init self._optimizer = optimizer self._t = 0 def initialize(self, trainer): optimizer = self._optimizer or trainer.updater.get_optimizer('main') if self._warmup_start is None: self._warmup_start = getattr(optimizer, self._attr) else: setattr(optimizer, self._attr, self._warmup_start) def __call__(self, trainer): self._t += 1 if self._t <= self._warmup_iter: optimizer = self._optimizer or \ trainer.updater.get_optimizer('main') value = (self._t * self._init + (self._warmup_iter - self._t) * self._warmup_start) / self._warmup_iter setattr(optimizer, self._attr, value) def serialize(self, serializer): self._t = serializer('_t', self._t)
2,174
35.864407
77
py
chainer
chainer-master/chainer/training/extensions/print_report.py
import os import sys from chainer.training import extension from chainer.training.extensions import log_report as log_report_module from chainer.training.extensions import util class PrintReport(extension.Extension): """Trainer extension to print the accumulated results. This extension uses the log accumulated by a :class:`LogReport` extension to print specified entries of the log in a human-readable format. Args: entries (list of str): List of keys of observations to print. log_report (str or LogReport): Log report to accumulate the observations. This is either the name of a LogReport extensions registered to the trainer, or a LogReport instance to use internally. out: Stream to print the bar. Standard output is used by default. """ def __init__(self, entries, log_report='LogReport', out=sys.stdout): self._entries = entries self._log_report = log_report self._out = out self._log_len = 0 # number of observations already printed # format information entry_widths = [max(10, len(s)) for s in entries] header = ' '.join(('{:%d}' % w for w in entry_widths)).format( *entries) + '\n' self._header = header # printed at the first call templates = [] for entry, w in zip(entries, entry_widths): templates.append((entry, '{:<%dg} ' % w, ' ' * (w + 2))) self._templates = templates def __call__(self, trainer): out = self._out if self._header: out.write(self._header) self._header = None log_report = self._log_report if isinstance(log_report, str): log_report = trainer.get_extension(log_report) elif isinstance(log_report, log_report_module.LogReport): log_report(trainer) # update the log report else: raise TypeError('log report has a wrong type %s' % type(log_report)) log = log_report.log log_len = self._log_len while len(log) > log_len: # delete the printed contents from the current cursor if os.name == 'nt': util.erase_console(0, 0) else: out.write('\033[J') self._print(log[log_len]) log_len += 1 self._log_len = log_len def serialize(self, serializer): log_report = self._log_report if isinstance(log_report, log_report_module.LogReport): log_report.serialize(serializer['_log_report']) def _print(self, observation): out = self._out for entry, template, empty in self._templates: if entry in observation: out.write(template.format(observation[entry])) else: out.write(empty) out.write('\n') if hasattr(out, 'flush'): out.flush()
2,967
32.727273
77
py
chainer
chainer-master/chainer/training/extensions/fail_on_nonnumber.py
from chainer.training import extension class FailOnNonNumber(extension.Extension): """Trainer extension to raise RuntimeError if parameters contain NaN or Inf. Although parameters including non-number such as NaN and Inf are unnecessary in most cases, :class:`~chainer.training.Trainer` will continue to compute even if the parameters in a given optimizer diverge. This extension is aimed to reduce unnecessary computations by throwing ``RuntimeError`` if the parameters contain NaN or Inf. """ def __call__(self, trainer): optimizers = trainer.updater.get_all_optimizers() for name, optimizer in optimizers.items(): target = optimizer.target xp = target.xp for param in target.params(): if not xp.isfinite(param.array).all(): raise RuntimeError( 'Kill the process since parameters in optimizer' ' \'{}\' diverge. R.I.P.'.format(name))
1,008
41.041667
80
py
chainer
chainer-master/chainer/training/extensions/computational_graph.py
import os import subprocess from chainer import computational_graph from chainer import configuration from chainer.training import extension from chainer.utils import argument from chainer import variable def is_return_code_zero(args): """Return `True` if the return code of the given command is zero. All the messages sent to stdout or stderr are suppressed. Args: args (list of str): A command to execute. """ with open(os.devnull, 'wb') as FNULL: try: subprocess.check_call(args, stdout=FNULL, stderr=FNULL) except subprocess.CalledProcessError: # The given command returned an error return False except OSError: # The given command was not found return False return True def is_graphviz_available(): """Tell whether graphviz is available or not.""" return is_return_code_zero(['dot', '-V']) _var_style = {'shape': 'octagon', 'fillcolor': '#E0E0E0', 'style': 'filled'} _func_style = {'shape': 'record', 'fillcolor': '#6495ED', 'style': 'filled'} class DumpGraph(extension.Extension): """__init__(\ root_name, filename='cg.dot', variable_style=None, function_style=None) Trainer extension to dump a computational graph. This extension dumps a computational graph. The graph is output in DOT language. If graphviz is available, this also renders and saves the image of the computational graph. It only dumps a graph at the first invocation. .. note:: The computational graph is not kept by default. This extension changes this behavior until the first invocation. **It is strongly recommended that you use it with the default trigger setting.** The detailed behavior of this extension is as follows. 1. In its initializer, it turns on the ``chainer.config.keep_graph_on_report`` flag. 2. At the first iteration, it dumps the graph using the graph held by the reported variable. 3. After dumping the graph, it turns off the flag (if it was originally turned off) so that any variable reported afterward does not hold a computational graph. When the ``keep_graph_on_report`` flag is turned on, the computational graph created by the updater is kept during the invocation of extensions. It will cause an unnecessarily large memory consumption when an extension also uses a large amount of memory, e.g. :class:`~chainer.training.extensions.Evaluator`. With the default setting, the ``DumpGraph`` extension is called at the first iteration. Since :class:`~chainer.training.extensions.Evaluator` is not called at the first iteration in most cases, it does not cause any memory problem. Args: root_name (str): Name of the root of the computational graph. The root variable is retrieved by this name from the observation dictionary of the trainer. filename (str): Output file name. For historical reasons ``out_name`` is also accepted as an alias of this argument. variable_style (dict): Dot node style for variables. Each variable is rendered by an octagon by default. function_style (dict): Dot node style for functions. Each function is rendered by a rectangular by default. .. seealso:: See :func:`~chainer.computational_graph.build_computational_graph` for the ``variable_style`` and ``function_style`` arguments. """ default_name = 'dump_graph' def __init__(self, root_name, filename=None, variable_style=None, function_style=None, **kwargs): out_name, = argument.parse_kwargs(kwargs, ('out_name', 'cg.dot')) if filename is None: filename = out_name del out_name # avoid accidental use self._root_name = root_name self._filename = filename if variable_style is None: variable_style = _var_style self._variable_style = variable_style if function_style is None: function_style = _func_style self._function_style = function_style self._original_flag = None self._flag_called = False def initialize(self, trainer): if not self._flag_called: self._original_flag = configuration.config.keep_graph_on_report configuration.config.keep_graph_on_report = True def trigger(self, trainer): if self._flag_called: return False self._flag_called = True return True def __call__(self, trainer): try: var = trainer.observation[self._root_name] if not isinstance(var, variable.Variable): raise TypeError('root value is not a Variable') cg = computational_graph.build_computational_graph( [var], variable_style=self._variable_style, function_style=self._function_style ).dump() filename = os.path.join(trainer.out, self._filename) with open(filename, 'w') as f: f.write(cg) if is_graphviz_available(): img_fn = os.path.splitext(self._filename)[0] + '.png' image_filename = os.path.join(trainer.out, img_fn) subprocess.check_call( ['dot', '-Tpng', filename, '-o', image_filename]) finally: configuration.config.keep_graph_on_report = self._original_flag def serialize(self, serializer): self._original_flag = serializer('_original_flag', self._original_flag) self._flag_called = serializer('_flag_called', self._flag_called)
5,787
36.584416
79
py
chainer
chainer-master/chainer/training/extensions/variable_statistics_plot.py
from __future__ import division import os import warnings import numpy import six import chainer from chainer import backend from chainer.backends import cuda from chainer.training import extension from chainer.training import trigger as trigger_module from chainer.utils import argument _available = None def _try_import_matplotlib(): global matplotlib, _available global _plot_color, _plot_color_trans, _plot_common_kwargs try: import matplotlib _available = True except ImportError: _available = False if _available: if hasattr(matplotlib.colors, 'to_rgba'): _to_rgba = matplotlib.colors.to_rgba else: # For matplotlib 1.x _to_rgba = matplotlib.colors.ColorConverter().to_rgba _plot_color = _to_rgba('#1f77b4') # C0 color _plot_color_trans = _plot_color[:3] + (0.2,) # apply alpha _plot_common_kwargs = { 'alpha': 0.2, 'linewidth': 0, 'color': _plot_color_trans} def _check_available(): if _available is None: _try_import_matplotlib() if not _available: warnings.warn('matplotlib is not installed on your environment, ' 'so nothing will be plotted at this time. ' 'Please install matplotlib to plot figures.\n\n' ' $ pip install matplotlib\n') def _unpack_variables(x, memo=None): if memo is None: memo = () if isinstance(x, chainer.Variable): memo += (x,) elif isinstance(x, chainer.Link): memo += tuple(x.params(include_uninit=True)) elif isinstance(x, (list, tuple)): for xi in x: memo += _unpack_variables(xi) return memo class Reservoir(object): """Reservoir sample with a fixed sized buffer.""" def __init__(self, size, data_shape, dtype=numpy.float32): self.size = size self.data = numpy.zeros((size,) + data_shape, dtype=dtype) self.idxs = numpy.zeros((size,), dtype=numpy.int32) self.counter = 0 def add(self, x, idx=None): if self.counter < self.size: self.data[self.counter] = x self.idxs[self.counter] = idx or self.counter elif self.counter >= self.size and \ numpy.random.random() < self.size / float(self.counter + 1): i = numpy.random.randint(self.size) self.data[i] = x self.idxs[i] = idx or self.counter self.counter += 1 def get_data(self): idxs = self.idxs[:min(self.counter, self.size)] sorted_args = numpy.argsort(idxs) return idxs[sorted_args], self.data[sorted_args] class Statistician(object): """Helper to compute basic NumPy-like statistics.""" def __init__(self, collect_mean, collect_std, percentile_sigmas): self.collect_mean = collect_mean self.collect_std = collect_std self.percentile_sigmas = percentile_sigmas def __call__(self, x, axis=0, dtype=None, xp=None): if axis is None: axis = tuple(range(x.ndim)) elif not isinstance(axis, (tuple, list)): axis = axis, return self.collect(x, axis) def collect(self, x, axis): out = dict() if self.collect_mean: out['mean'] = x.mean(axis=axis) if self.collect_std: out['std'] = x.std(axis=axis) if self.percentile_sigmas: xp = backend.get_array_module(x) p = xp.percentile(x, self.percentile_sigmas, axis=axis) out['percentile'] = p return out class VariableStatisticsPlot(extension.Extension): """__init__(\ targets, max_sample_size=1000, report_data=True, report_grad=True, \ plot_mean=True, plot_std=True, \ percentile_sigmas=(0, 0.13, 2.28, 15.87, 50, 84.13, 97.72, 99.87, 100), \ trigger=(1, 'epoch'), filename='statistics.png', figsize=None, marker=None, \ grid=True) Trainer extension to plot statistics for :class:`~chainer.Variable`\\s. This extension collects statistics for a single :class:`Variable`, a list of :class:`Variable`\\s or similarly a single or a list of :class:`Link`\\s containing one or more :class:`Variable`\\s. In case multiple :class:`Variable`\\s are found, the means are computed. The collected statistics are plotted and saved as an image in the directory specified by the :class:`Trainer`. Statistics include mean, standard deviation and percentiles. This extension uses reservoir sampling to preserve memory, using a fixed size running sample. This means that collected items in the sample are discarded uniformly at random when the number of items becomes larger than the maximum sample size, but each item is expected to occur in the sample with equal probability. Args: targets (:class:`Variable`, :class:`Link` or list of either): Parameters for which statistics are collected. max_sample_size (int): Maximum number of running samples. report_data (bool): If ``True``, data (e.g. weights) statistics are plotted. If ``False``, they are neither computed nor plotted. report_grad (bool): If ``True``, gradient statistics are plotted. If ``False``, they are neither computed nor plotted. plot_mean (bool): If ``True``, means are plotted. If ``False``, they are neither computed nor plotted. plot_std (bool): If ``True``, standard deviations are plotted. If ``False``, they are neither computed nor plotted. percentile_sigmas (float or tuple of floats): Percentiles to plot in the range :math:`[0, 100]`. trigger: Trigger that decides when to save the plots as an image. This is distinct from the trigger of this extension itself. If it is a tuple in the form ``<int>, 'epoch'`` or ``<int>, 'iteration'``, it is passed to :class:`IntervalTrigger`. filename (str): Name of the output image file under the output directory. For historical reasons ``file_name`` is also accepted as an alias of this argument. figsize (tuple of int): Matlotlib ``figsize`` argument that specifies the size of the output image. marker (str): Matplotlib ``marker`` argument that specified the marker style of the plots. grid (bool): Matplotlib ``grid`` argument that specifies whether grids are rendered in in the plots or not. """ def __init__(self, targets, max_sample_size=1000, report_data=True, report_grad=True, plot_mean=True, plot_std=True, percentile_sigmas=( 0, 0.13, 2.28, 15.87, 50, 84.13, 97.72, 99.87, 100), trigger=(1, 'epoch'), filename=None, figsize=None, marker=None, grid=True, **kwargs): file_name, = argument.parse_kwargs( kwargs, ('file_name', 'statistics.png') ) if filename is None: filename = file_name del file_name # avoid accidental use self._vars = _unpack_variables(targets) if not self._vars: raise ValueError( 'Need at least one variables for which to collect statistics.' '\nActual: 0 <= 0') if not any((plot_mean, plot_std, bool(percentile_sigmas))): raise ValueError('Nothing to plot') self._keys = [] if report_data: self._keys.append('data') if report_grad: self._keys.append('grad') self._report_data = report_data self._report_grad = report_grad self._statistician = Statistician( collect_mean=plot_mean, collect_std=plot_std, percentile_sigmas=percentile_sigmas) self._plot_mean = plot_mean self._plot_std = plot_std self._plot_percentile = bool(percentile_sigmas) self._trigger = trigger_module.get_trigger(trigger) self._filename = filename self._figsize = figsize self._marker = marker self._grid = grid if not self._plot_percentile: n_percentile = 0 else: if not isinstance(percentile_sigmas, (list, tuple)): n_percentile = 1 # scalar, single percentile else: n_percentile = len(percentile_sigmas) self._data_shape = ( len(self._keys), int(plot_mean) + int(plot_std) + n_percentile) self._samples = Reservoir(max_sample_size, data_shape=self._data_shape) @staticmethod def available(): _check_available() return _available def __call__(self, trainer): if self.available(): # Dynamically import pyplot to call matplotlib.use() # after importing chainer.training.extensions import matplotlib.pyplot as plt else: return xp = backend.get_array_module(self._vars[0].data) stats = xp.zeros(self._data_shape, dtype=xp.float32) for i, k in enumerate(self._keys): xs = [] for var in self._vars: x = getattr(var, k, None) if x is not None: xs.append(x.ravel()) if xs: stat_dict = self._statistician( xp.concatenate(xs, axis=0), axis=0, xp=xp) stat_list = [] if self._plot_mean: stat_list.append(xp.atleast_1d(stat_dict['mean'])) if self._plot_std: stat_list.append(xp.atleast_1d(stat_dict['std'])) if self._plot_percentile: stat_list.append(xp.atleast_1d(stat_dict['percentile'])) stats[i] = xp.concatenate(stat_list, axis=0) if xp == cuda.cupy: stats = cuda.to_cpu(stats) self._samples.add(stats, idx=trainer.updater.iteration) if self._trigger(trainer): file_path = os.path.join(trainer.out, self._filename) self.save_plot_using_module(file_path, plt) def save_plot_using_module(self, file_path, plt): nrows = int(self._plot_mean or self._plot_std) \ + int(self._plot_percentile) ncols = len(self._keys) fig, axes = plt.subplots( nrows, ncols, figsize=self._figsize, sharex=True) if not isinstance(axes, numpy.ndarray): # single subplot axes = numpy.asarray([axes]) if nrows == 1: axes = axes[None, :] elif ncols == 1: axes = axes[:, None] assert axes.ndim == 2 idxs, data = self._samples.get_data() # Offset to access percentile data from `data` offset = int(self._plot_mean) + int(self._plot_std) n_percentile = data.shape[-1] - offset n_percentile_mid_floor = n_percentile // 2 n_percentile_odd = n_percentile % 2 == 1 for col in six.moves.range(ncols): row = 0 ax = axes[row, col] ax.set_title(self._keys[col]) # `data` or `grad` if self._plot_mean or self._plot_std: if self._plot_mean and self._plot_std: ax.errorbar( idxs, data[:, col, 0], data[:, col, 1], color=_plot_color, ecolor=_plot_color_trans, label='mean, std', marker=self._marker) else: if self._plot_mean: label = 'mean' elif self._plot_std: label = 'std' ax.plot( idxs, data[:, col, 0], color=_plot_color, label=label, marker=self._marker) row += 1 if self._plot_percentile: ax = axes[row, col] for i in six.moves.range(n_percentile_mid_floor + 1): if n_percentile_odd and i == n_percentile_mid_floor: # Enters at most once per sub-plot, in case there is # only a single percentile to plot or when this # percentile is the mid percentile and the number of # percentiles are odd ax.plot( idxs, data[:, col, offset + i], color=_plot_color, label='percentile', marker=self._marker) else: if i == n_percentile_mid_floor: # Last percentiles and the number of all # percentiles are even label = 'percentile' else: label = '_nolegend_' ax.fill_between( idxs, data[:, col, offset + i], data[:, col, -i - 1], label=label, **_plot_common_kwargs) ax.set_xlabel('iteration') for ax in axes.ravel(): ax.legend() if self._grid: ax.grid() ax.set_axisbelow(True) fig.savefig(file_path) plt.close()
13,511
35.617886
79
py
chainer
chainer-master/chainer/training/extensions/evaluator.py
import copy import datetime import warnings import six from chainer import backend from chainer import configuration from chainer.dataset import convert from chainer.dataset import iterator as iterator_module from chainer import function from chainer import iterators from chainer import link from chainer import reporter as reporter_module from chainer.training import extension from chainer.training.extensions import util from chainer.utils import argument class Evaluator(extension.Extension): """__init__(self, iterator, target, converter=convert.concat_examples, \ device=None, eval_hook=None, eval_func=None, *, progress_bar=False) Trainer extension to evaluate models on a validation set. This extension evaluates the current models by a given evaluation function. It creates a :class:`~chainer.Reporter` object to store values observed in the evaluation function on each iteration. The report for all iterations are aggregated to :class:`~chainer.DictSummary`. The collected mean values are further reported to the reporter object of the trainer, where the name of each observation is prefixed by the evaluator name. See :class:`~chainer.Reporter` for details in naming rules of the reports. Evaluator has a structure to customize similar to that of :class:`~chainer.training.updaters.StandardUpdater`. The main differences are: - There are no optimizers in an evaluator. Instead, it holds links to evaluate. - An evaluation loop function is used instead of an update function. - Preparation routine can be customized, which is called before each evaluation. It can be used, e.g., to initialize the state of stateful recurrent networks. There are two ways to modify the evaluation behavior besides setting a custom evaluation function. One is by setting a custom evaluation loop via the ``eval_func`` argument. The other is by inheriting this class and overriding the :meth:`evaluate` method. In latter case, users have to create and handle a reporter object manually. Users also have to copy the iterators before using them, in order to reuse them at the next time of evaluation. In both cases, the functions are called in testing mode (i.e., ``chainer.config.train`` is set to ``False``). This extension is called at the end of each epoch by default. Args: iterator: Dataset iterator for the validation dataset. It can also be a dictionary of iterators. If this is just an iterator, the iterator is registered by the name ``'main'``. target: Link object or a dictionary of links to evaluate. If this is just a link object, the link is registered by the name ``'main'``. converter: Converter function to build input arrays. :func:`~chainer.dataset.concat_examples` is used by default. device: Device to which the validation data is sent. Negative value indicates the host memory (CPU). eval_hook: Function to prepare for each evaluation process. It is called at the beginning of the evaluation. The evaluator extension object is passed at each call. eval_func: Evaluation function called at each iteration. The target link to evaluate as a callable is used by default. progress_bar: Boolean flag to show a progress bar while training, which is similar to :class:`~chainer.training.extensions.ProgressBar`. (default: ``False``) .. warning:: The argument ``progress_bar`` is experimental. The interface can change in the future. Attributes: converter: Converter function. device: Device to which the validation data is sent. eval_hook: Function to prepare for each evaluation process. eval_func: Evaluation function called at each iteration. """ trigger = 1, 'epoch' default_name = 'validation' priority = extension.PRIORITY_WRITER name = None def __init__(self, iterator, target, converter=convert.concat_examples, device=None, eval_hook=None, eval_func=None, **kwargs): progress_bar, = argument.parse_kwargs(kwargs, ('progress_bar', False)) if device is not None: device = backend.get_device(device) if isinstance(iterator, iterator_module.Iterator): iterator = {'main': iterator} self._iterators = iterator if isinstance(target, link.Link): target = {'main': target} self._targets = target self.converter = converter self.device = device self.eval_hook = eval_hook self.eval_func = eval_func self._progress_bar = progress_bar for key, iter in six.iteritems(iterator): if (isinstance(iter, (iterators.SerialIterator, iterators.MultiprocessIterator, iterators.MultithreadIterator)) and getattr(iter, 'repeat', False)): msg = 'The `repeat` property of the iterator {} ' 'is set to `True`. Typically, the evaluator sweeps ' 'over iterators until they stop, ' 'but as the property being `True`, this iterator ' 'might not stop and evaluation could go into ' 'an infinite loop. ' 'We recommend to check the configuration ' 'of iterators'.format(key) warnings.warn(msg) def get_iterator(self, name): """Returns the iterator of the given name.""" return self._iterators[name] def get_all_iterators(self): """Returns a dictionary of all iterators.""" return dict(self._iterators) def get_target(self, name): """Returns the target link of the given name.""" return self._targets[name] def get_all_targets(self): """Returns a dictionary of all target links.""" return dict(self._targets) def __call__(self, trainer=None): """Executes the evaluator extension. Unlike usual extensions, this extension can be executed without passing a trainer object. This extension reports the performance on validation dataset using the :func:`~chainer.report` function. Thus, users can use this extension independently from any trainer by manually configuring a :class:`~chainer.Reporter` object. Args: trainer (~chainer.training.Trainer): Trainer object that invokes this extension. It can be omitted in case of calling this extension manually. Returns: dict: Result dictionary that contains mean statistics of values reported by the evaluation function. """ # set up a reporter reporter = reporter_module.Reporter() if self.name is not None: prefix = self.name + '/' else: prefix = '' for name, target in six.iteritems(self._targets): reporter.add_observer(prefix + name, target) reporter.add_observers(prefix + name, target.namedlinks(skipself=True)) with reporter: with configuration.using_config('train', False): result = self.evaluate() reporter_module.report(result) return result def evaluate(self): """Evaluates the model and returns a result dictionary. This method runs the evaluation loop over the validation dataset. It accumulates the reported values to :class:`~chainer.DictSummary` and returns a dictionary whose values are means computed by the summary. Note that this function assumes that the main iterator raises ``StopIteration`` or code in the evaluation loop raises an exception. So, if this assumption is not held, the function could be caught in an infinite loop. Users can override this method to customize the evaluation routine. .. note:: This method encloses :attr:`eval_func` calls with :func:`function.no_backprop_mode` context, so all calculations using :class:`~chainer.FunctionNode`\\s inside :attr:`eval_func` do not make computational graphs. It is for reducing the memory consumption. Returns: dict: Result dictionary. This dictionary is further reported via :func:`~chainer.report` without specifying any observer. """ iterator = self._iterators['main'] eval_func = self.eval_func or self._targets['main'] if self.eval_hook: self.eval_hook(self) if hasattr(iterator, 'reset'): iterator.reset() it = iterator else: warnings.warn( 'This iterator does not have the reset method. Evaluator ' 'copies the iterator instead of resetting. This behavior is ' 'deprecated. Please implement the reset method.', DeprecationWarning) it = copy.copy(iterator) summary = reporter_module.DictSummary() if self._progress_bar: pbar = _IteratorProgressBar(iterator=it) for batch in it: observation = {} with reporter_module.report_scope(observation): in_arrays = convert._call_converter( self.converter, batch, self.device) with function.no_backprop_mode(): if isinstance(in_arrays, tuple): eval_func(*in_arrays) elif isinstance(in_arrays, dict): eval_func(**in_arrays) else: eval_func(in_arrays) summary.add(observation) if self._progress_bar: pbar.update() if self._progress_bar: pbar.close() return summary.compute_mean() def finalize(self): """Finalizes the evaluator object. This method calls the `finalize` method of each iterator that this evaluator has. It is called at the end of training loops. """ for iterator in six.itervalues(self._iterators): iterator.finalize() class _IteratorProgressBar(util.ProgressBar): def __init__(self, iterator, bar_length=None, out=None): if not (hasattr(iterator, 'current_position') and hasattr(iterator, 'epoch_detail')): raise TypeError('Iterator must have the following attributes ' 'to enable a progress bar: ' 'current_position, epoch_detail') self._iterator = iterator super(_IteratorProgressBar, self).__init__( bar_length=bar_length, out=out) def get_lines(self): iteration = self._iterator.current_position epoch_detail = self._iterator.epoch_detail epoch_size = getattr(self._iterator, '_epoch_size', None) lines = [] rate = epoch_detail marks = '#' * int(rate * self._bar_length) lines.append('validation [{}{}] {:6.2%}\n'.format( marks, '.' * (self._bar_length - len(marks)), rate)) if epoch_size: lines.append('{:10} / {} iterations\n' .format(iteration, epoch_size)) else: lines.append('{:10} iterations\n'.format(iteration)) speed_t, speed_e = self.update_speed(iteration, epoch_detail) estimated_time = (1.0 - epoch_detail) / speed_e lines.append('{:10.5g} iters/sec. Estimated time to finish: {}.\n' .format(speed_t, datetime.timedelta(seconds=estimated_time))) return lines
11,956
38.075163
79
py
chainer
chainer-master/chainer/training/extensions/polynomial_shift.py
from __future__ import division import numpy from chainer.training import extension class PolynomialShift(extension.Extension): """Trainer extension to polynomially shift an optimizer attribute. This extension polynomially decreases the specified attribute of the optimizer. The typical use case is a polynomial decay of the learning rate at each iteration. For example, suppose that this extension is invoke at every iteration. Then this extension will set the corresponding attribute to ``init_value * (1 - i / max_iter) ^ rate`` at the ``i``-th iteration, where the ``max_iter`` is the number of iterations to be running. This extension is also called before the training loop starts by default. Args: attr (str): Name of the attribute to shift. rate (float): Exponent of polynomial shift. max_count (int): Number of this extension to be invoked. init (float): Initial value of the attribute. If it is ``None``, the extension extracts the attribute at the first call and uses it as the initial value. target (float): Target value of the attribute. If the attribute reaches this value, the shift stops. optimizer (~chainer.Optimizer): Target optimizer to adjust the attribute. If it is ``None``, the main optimizer of the updater is used. """ invoke_before_training = True def __init__(self, attr, rate, max_count, init=None, target=None, optimizer=None): self._attr = attr self._rate = rate self._init = init self._target = target self._optimizer = optimizer self._t = 0 self._max_count = max_count self._last_value = None def initialize(self, trainer): optimizer = self._get_optimizer(trainer) # ensure that _init is set if self._init is None: self._init = getattr(optimizer, self._attr) if self._last_value is not None: # resuming from a snapshot self._update_value(optimizer, self._last_value) else: self._update_value(optimizer, self._init) def __call__(self, trainer): self._t += 1 optimizer = self._get_optimizer(trainer) decay = max(1 - self._t / self._max_count, 0) value = self._init * decay ** self._rate if self._target is not None: if self._rate > 0: # almost same as value = min(value, self._target), but this # line supports negative values, too if self._target / value > 1: value = self._target else: # ditto if self._target / value < 1: value = self._target self._update_value(optimizer, value) def serialize(self, serializer): self._t = serializer('_t', self._t) self._last_value = serializer('_last_value', self._last_value) if isinstance(self._last_value, numpy.ndarray): self._last_value = self._last_value.item() def _get_optimizer(self, trainer): return self._optimizer or trainer.updater.get_optimizer('main') def _update_value(self, optimizer, value): setattr(optimizer, self._attr, value) self._last_value = value
3,354
34.315789
79
py
chainer
chainer-master/chainer/training/extensions/micro_average.py
from chainer import reporter from chainer.training import extension from chainer.training import util class MicroAverage(extension.Extension): """Calculates micro-average ratio. Give :math:`N` batches and values :math:`\\{n_1, \\dots, n_N\\}` and :math:`\\{d_1, \\dots, d_N\\}`, this extension calculates micro-average of these ratio defined as: .. math:: \\frac{\\sum_i^N n_i}{\\sum_i^N d_i}. A user usually uses the number of examples which a system correctly predict as :math:`n_i` and the number of total examples in :math:`i`-th batch as :math:`d_i`. This value is called macro-average of precision. Note that macro-average is defined as: .. math:: \\frac{1}{N}\\sum_i^N (n_i / d_i), It is same to the micro-average when each mini-batch has the same :math:`d_i`. You need to report numerator value (the number of correct examples) and denominator value (the number of examples) in your model. >>> class MyModel(chainer.Link): ... def __call__(self, x, y): ... loss = F.softmax_cross_entropy(x, y) ... correct = (x.data.argmax(axis=1) == y.data).sum() ... total = len(y.data) ... reporter.report({'correct': correct, 'total': total}, self) ... return loss And then, make an extension with corresponding reporting keys and register it. >>> ext = extensions.MicroAverage( ... 'main/correct', 'main/total', 'main/accuracy') Args: numerator_key (str): Key string of obserbation storing a numerator value. denominator_key (str): Key string of obserbation storing a denominator value. result_key (str): Key string of obserbation to store a result. trigger: Trigger that decides when to calcurate average. This is distinct from the trigger of this extension itself. If it is a tuple in the form ``<int>, 'epoch'`` or ``<int>, 'iteration'``, it is passed to :class:`IntervalTrigger`. """ priority = extension.PRIORITY_EDITOR def __init__( self, numerator_key, denominator_key, result_key, trigger=(1, 'epoch')): self._trigger = util.get_trigger(trigger) self._numerator_key = numerator_key self._denominator_key = denominator_key self._result_key = result_key self._numerator = 0 self._denominator = 0 def __call__(self, trainer): observation = trainer.observation if not (self._numerator_key in observation and self._denominator_key in observation): return self._numerator += observation[self._numerator_key] self._denominator += observation[self._denominator_key] if self._trigger(trainer): result = float(self._numerator) / self._denominator self._numerator = 0 self._denominator = 0 reporter.report({self._result_key: result}) def serialize(self, serializer): self._numerator = serializer('_numerator', self._numerator) self._denominator = serializer('_denominator', self._denominator)
3,192
33.706522
78
py
chainer
chainer-master/chainer/training/extensions/util.py
import collections import os import sys import time if os.name == 'nt': import ctypes _STD_OUTPUT_HANDLE = -11 class _COORD(ctypes.Structure): _fields_ = [('X', ctypes.c_short), ('Y', ctypes.c_short)] class _SMALL_RECT(ctypes.Structure): _fields_ = [('Left', ctypes.c_short), ('Top', ctypes.c_short), ('Right', ctypes.c_short), ('Bottom', ctypes.c_short)] class _CONSOLE_SCREEN_BUFFER_INFO(ctypes.Structure): _fields_ = [('dwSize', _COORD), ('dwCursorPosition', _COORD), ('wAttributes', ctypes.c_ushort), ('srWindow', _SMALL_RECT), ('dwMaximumWindowSize', _COORD)] def set_console_cursor_position(x, y): """Set relative cursor position from current position to (x,y)""" whnd = ctypes.windll.kernel32.GetStdHandle(_STD_OUTPUT_HANDLE) csbi = _CONSOLE_SCREEN_BUFFER_INFO() ctypes.windll.kernel32.GetConsoleScreenBufferInfo(whnd, ctypes.byref(csbi)) cur_pos = csbi.dwCursorPosition pos = _COORD(cur_pos.X + x, cur_pos.Y + y) ctypes.windll.kernel32.SetConsoleCursorPosition(whnd, pos) def erase_console(x, y, mode=0): """Erase screen. Mode=0: From (x,y) position down to the bottom of the screen. Mode=1: From (x,y) position down to the beginning of line. Mode=2: Hole screen """ whnd = ctypes.windll.kernel32.GetStdHandle(_STD_OUTPUT_HANDLE) csbi = _CONSOLE_SCREEN_BUFFER_INFO() ctypes.windll.kernel32.GetConsoleScreenBufferInfo(whnd, ctypes.byref(csbi)) cur_pos = csbi.dwCursorPosition wr = ctypes.c_ulong() if mode == 0: num = csbi.srWindow.Right * (csbi.srWindow.Bottom - cur_pos.Y) - cur_pos.X ctypes.windll.kernel32.FillConsoleOutputCharacterA( whnd, ord(' '), num, cur_pos, ctypes.byref(wr)) elif mode == 1: num = cur_pos.X ctypes.windll.kernel32.FillConsoleOutputCharacterA( whnd, ord(' '), num, _COORD(0, cur_pos.Y), ctypes.byref(wr)) elif mode == 2: os.system('cls') class ProgressBar(object): def __init__(self, bar_length=None, out=None): self._bar_length = 50 if bar_length is None else bar_length self._out = sys.stdout if out is None else out self._recent_timing = collections.deque([], maxlen=100) def update_speed(self, iteration, epoch_detail): now = time.time() self._recent_timing.append((iteration, epoch_detail, now)) old_t, old_e, old_sec = self._recent_timing[0] span = now - old_sec if span != 0: speed_t = (iteration - old_t) / span speed_e = (epoch_detail - old_e) / span else: speed_t = float('inf') speed_e = float('inf') return speed_t, speed_e def get_lines(self): raise NotImplementedError def update(self): self.erase_console() lines = self.get_lines() for line in lines: self._out.write(line) self.move_cursor_up(len(lines)) self.flush() def close(self): self.erase_console() self.flush() def erase_console(self): if os.name == 'nt': erase_console(0, 0) else: self._out.write('\033[J') def move_cursor_up(self, n): # move the cursor to the head of the progress bar if os.name == 'nt': set_console_cursor_position(0, - n) else: self._out.write('\033[{:d}A'.format(n)) def flush(self): if hasattr(self._out, 'flush'): self._out.flush()
3,869
32.947368
77
py
chainer
chainer-master/chainer/training/extensions/variable_unchain.py
import six from chainer import configuration from chainer.training import extension from chainer import variable class unchain_variables(extension.Extension): """Trainer extension to unchain all comptational graphs. This extenstion unchains all comptational graphs after all extensions are run to release memory and to avoid memory leak. This extension can be used as a last resort when there is an extension that use a variable graph and cannot release the graph in itself. It observes the previous ``chainer.config.keep_graph_on_report`` flag. The extension is triggered when the flag is turned on. """ priority = 0 def __init__(self): self._prev_flag = None def initialize(self, _): self._prev_flag = configuration.config.keep_graph_on_report def trigger(self, _): flag = self._prev_flag self._prev_flag = configuration.config.keep_graph_on_report return flag def __call__(self, trainer): for var in six.itervalues(trainer.observation): if isinstance(var, variable.Variable): var.unchain_backward()
1,139
30.666667
79
py
chainer
chainer-master/chainer/training/extensions/value_observation.py
from chainer.training import extension def observe_value(observation_key, target_func): """Returns a trainer extension to continuously record a value. Args: observation_key (str): Key of observation to record. target_func (function): Function that returns the value to record. It must take one argument: :class:~chainer.training.Trainer object. Returns: The extension function. This extension is triggered each epoch by default. To change this, use the ``trigger`` argument with the :meth:`Trainer.extend() <chainer.training.Trainer.extend>` method. """ @extension.make_extension( trigger=(1, 'epoch'), priority=extension.PRIORITY_WRITER) def _observe_value(trainer): trainer.observation[observation_key] = target_func(trainer) return _observe_value def observe_lr(optimizer_name='main', observation_key='lr'): """Returns a trainer extension to record the learning rate. Args: optimizer_name (str): Name of optimizer whose learning rate is recorded. observation_key (str): Key of observation to record. Returns: The extension function. This extension is triggered each epoch by default. To change this, use the ``trigger`` argument with the :meth:`Trainer.extend() <chainer.training.Trainer.extend>` method. """ return observe_value( observation_key, lambda trainer: trainer.updater.get_optimizer(optimizer_name).lr)
1,504
32.444444
79
py
chainer
chainer-master/chainer/training/extensions/parameter_statistics.py
import numpy import six import chainer from chainer import backend from chainer import reporter from chainer.training import extension from chainer.training import trigger as trigger_module _default_statistics = { 'mean': lambda x: backend.get_array_module(x).mean(x), 'std': lambda x: backend.get_array_module(x).std(x), 'min': lambda x: backend.get_array_module(x).min(x), 'max': lambda x: backend.get_array_module(x).max(x), 'zeros': lambda x: backend.get_array_module(x).count_nonzero(x == 0), 'percentile': lambda x: backend.get_array_module(x).percentile( x, (0.13, 2.28, 15.87, 50, 84.13, 97.72, 99.87)) } class ParameterStatistics(extension.Extension): """Trainer extension to report parameter statistics. Statistics are collected and reported for a given :class:`~chainer.Link` or an iterable of :class:`~chainer.Link`\\ s. If a link contains child links, the statistics are reported separately for each child. Any function that takes a one-dimensional :class:`numpy.ndarray` or a :class:`cupy.ndarray` and outputs a single or multiple real numbers can be registered to handle the collection of statistics, e.g. :meth:`numpy.ndarray.mean`. The keys of reported statistics follow the convention of link name followed by parameter name, attribute name and function name, e.g. ``VGG16Layers/conv1_1/W/data/mean``. They are prepended with an optional prefix and appended with integer indices if the statistics generating function return multiple values. Args: links (~chainer.Link or iterable of ~chainer.Link): Link(s) containing the parameters to observe. The link is expected to have a ``name`` attribute which is used as a part of the report key. statistics (dict or 'default'): Dictionary with function name to function mappings. The name is a string and is used as a part of the report key. The function is responsible for generating the statistics. If the special value ``'default'`` is specified, the default statistics functions will be used. report_params (bool): If ``True``, report statistics for parameter values such as weights and biases. report_grads (bool): If ``True``, report statistics for parameter gradients. prefix (str): Optional prefix to prepend to the report keys. trigger: Trigger that decides when to aggregate the results and report the values. skip_nan_params (bool): If ``True``, statistics are not computed for parameters including NaNs and a single NaN value is immediately reported instead. Otherwise, this extension will simply try to compute the statistics without performing any checks for NaNs. .. note:: The default statistic functions are as follows: * ``'mean'`` (``xp.mean(x)``) * ``'std'`` (``xp.std(x)``) * ``'min'`` (``xp.min(x)``) * ``'max'`` (``xp.max(x)``) * ``'zeros'`` (``xp.count_nonzero(x == 0)``) * ``'percentile'`` (``xp.percentile(x, \ (0.13, 2.28, 15.87, 50, 84.13, 97.72, 99.87))``) """ default_name = 'parameter_statistics' priority = extension.PRIORITY_WRITER # prefix ends with a '/' and param_name is preceded by a '/' report_key_template = ('{prefix}{link_name}{param_name}/{attr_name}/' '{function_name}') default_statistics = _default_statistics def __init__(self, links, statistics='default', report_params=True, report_grads=True, prefix=None, trigger=(1, 'epoch'), skip_nan_params=False): if not isinstance(links, (list, tuple)): links = links, self._links = links if statistics is None: statistics = {} elif statistics == 'default': statistics = self.default_statistics self._statistics = dict(statistics) attrs = [] if report_params: attrs.append('data') if report_grads: attrs.append('grad') self._attrs = attrs self._prefix = prefix self._trigger = trigger_module.get_trigger(trigger) self._summary = reporter.DictSummary() self._skip_nan_params = skip_nan_params def __call__(self, trainer): """Execute the statistics extension. Collect statistics for the current state of parameters. Note that this method will merely update its statistic summary, unless the internal trigger is fired. If the trigger is fired, the summary will also be reported and then reset for the next accumulation. Args: trainer (~chainer.training.Trainer): Associated trainer that invoked this extension. """ statistics = {} for link in self._links: link_name = getattr(link, 'name', 'None') for param_name, param in link.namedparams(): for attr_name in self._attrs: for function_name, function in \ six.iteritems(self._statistics): # Get parameters as a flattened one-dimensional array # since the statistics function should make no # assumption about the axes params = getattr(param, attr_name).ravel() if (self._skip_nan_params and ( backend.get_array_module(params).isnan(params) .any())): value = numpy.nan else: value = function(params) key = self.report_key_template.format( prefix=self._prefix + '/' if self._prefix else '', link_name=link_name, param_name=param_name, attr_name=attr_name, function_name=function_name ) if (isinstance(value, chainer.get_array_types()) and value.size > 1): # Append integer indices to the keys if the # statistic function return multiple values statistics.update({'{}/{}'.format(key, i): v for i, v in enumerate(value)}) else: statistics[key] = value self._summary.add(statistics) if self._trigger(trainer): reporter.report(self._summary.compute_mean()) self._summary = reporter.DictSummary() # Clear summary def register_statistics(self, name, function): """Register a function to compute a certain statistic. The registered function will be called each time the extension runs and the results will be included in the report. Args: name (str): Name of the statistic. function: Function to generate the statistic. Any function that takes a one-dimensional :class:`numpy.ndarray` or a :class:`cupy.ndarray` and outputs a single or multiple real numbers is allowed. """ self._statistics[name] = function
7,507
41.179775
79
py
chainer
chainer-master/chainer/training/extensions/__init__.py
# import classes and functions from chainer.training.extensions._snapshot import snapshot # NOQA from chainer.training.extensions._snapshot import snapshot_object # NOQA from chainer.training.extensions.computational_graph import DumpGraph # NOQA from chainer.training.extensions.evaluator import Evaluator # NOQA from chainer.training.extensions.exponential_shift import ExponentialShift # NOQA from chainer.training.extensions.fail_on_nonnumber import FailOnNonNumber # NOQA from chainer.training.extensions.inverse_shift import InverseShift # NOQA from chainer.training.extensions.linear_shift import LinearShift # NOQA from chainer.training.extensions.log_report import LogReport # NOQA from chainer.training.extensions.micro_average import MicroAverage # NOQA from chainer.training.extensions.multistep_shift import MultistepShift # NOQA from chainer.training.extensions.parameter_statistics import ParameterStatistics # NOQA from chainer.training.extensions.plot_report import PlotReport # NOQA from chainer.training.extensions.polynomial_shift import PolynomialShift # NOQA from chainer.training.extensions.print_report import PrintReport # NOQA from chainer.training.extensions.progress_bar import ProgressBar # NOQA from chainer.training.extensions.step_shift import StepShift # NOQA from chainer.training.extensions.value_observation import observe_lr # NOQA from chainer.training.extensions.value_observation import observe_value # NOQA from chainer.training.extensions.variable_statistics_plot import VariableStatisticsPlot # NOQA from chainer.training.extensions.variable_unchain import unchain_variables # NOQA from chainer.training.extensions.warmup_shift import WarmupShift # NOQA # Alias from chainer.training.extensions.computational_graph import DumpGraph as dump_graph # NOQA
1,820
66.444444
95
py
chainer
chainer-master/chainer/training/extensions/step_shift.py
from __future__ import division import numpy from chainer.training import extension class StepShift(extension.Extension): """Trainer extension to shift an optimizer attribute in "steps". This extension multiplies the specified attribute of the optimizer in "steps". The typical use case is to scale the attribute at every ``k``\\ th iteration. For example, suppose that this extension is invoked at every iteration, then given ``k``, a multiplier ``gamma`` and an initial value ``init``, the optimizer attribute is set to ``init * gamma ^ (floor(i / k))``, where ``i`` represents the index of the current iteration. This extension is also called before the training loop starts by default. Args: attr (str): Name of the optimizer attribute to adjust. gamma (float): The multiplier. step (int): The interval for the multiplication, i.e., ``k``. init (float): Initial value of the attribute. If it is ``None``, the extension extracts the attribute at the first call and uses it as the initial value. target (float): Target value of the attribute. If the attribute reaches this value, the shift stops. optimizer (~chainer.Optimizer): Target optimizer object. If it is None, the main optimizer of the trainer is used. """ def __init__(self, attr, gamma, step, init=None, target=None, optimizer=None): self._attr = attr self._gamma = gamma self._step = step self._init = init self._target = target self._optimizer = optimizer self._t = 0 self._last_value = None def initialize(self, trainer): optimizer = self._get_optimizer(trainer) # ensure that _init is set if self._init is None: self._init = getattr(optimizer, self._attr) if self._last_value is not None: value = self._last_value else: value = self._init self._update_value(optimizer, value) def __call__(self, trainer): self._t += 1 optimizer = self._get_optimizer(trainer) value = self._init * self._gamma ** numpy.floor(self._t / self._step) if self._target is not None: if self._gamma > 1: # almost same as value = min(value, self._target), but this # line supports negative values, too if value / self._target > 1: value = self._target else: # ditto if value / self._target < 1: value = self._target self._update_value(optimizer, value) def serialize(self, serializer): self._t = serializer('_t', self._t) self._last_value = serializer('_last_value', self._last_value) if isinstance(self._last_value, numpy.ndarray): self._last_value = self._last_value.item() def _get_optimizer(self, trainer): return self._optimizer or trainer.updater.get_optimizer('main') def _update_value(self, optimizer, value): setattr(optimizer, self._attr, value) self._last_value = value
3,214
35.534091
79
py
chainer
chainer-master/chainer/training/extensions/log_report.py
import json import os import shutil import warnings import six from chainer import reporter from chainer import serializer as serializer_module from chainer.training import extension from chainer.training import trigger as trigger_module from chainer import utils from chainer.utils import argument class LogReport(extension.Extension): """__init__(\ keys=None, trigger=(1, 'epoch'), postprocess=None, filename='log') Trainer extension to output the accumulated results to a log file. This extension accumulates the observations of the trainer to :class:`~chainer.DictSummary` at a regular interval specified by a supplied trigger, and writes them into a log file in JSON format. There are two triggers to handle this extension. One is the trigger to invoke this extension, which is used to handle the timing of accumulating the results. It is set to ``1, 'iteration'`` by default. The other is the trigger to determine when to emit the result. When this trigger returns True, this extension appends the summary of accumulated values to the list of past summaries, and writes the list to the log file. Then, this extension makes a new fresh summary object which is used until the next time that the trigger fires. It also adds some entries to each result dictionary. - ``'epoch'`` and ``'iteration'`` are the epoch and iteration counts at the output, respectively. - ``'elapsed_time'`` is the elapsed time in seconds since the training begins. The value is taken from :attr:`Trainer.elapsed_time`. Args: keys (iterable of strs): Keys of values to accumulate. If this is None, all the values are accumulated and output to the log file. trigger: Trigger that decides when to aggregate the result and output the values. This is distinct from the trigger of this extension itself. If it is a tuple in the form ``<int>, 'epoch'`` or ``<int>, 'iteration'``, it is passed to :class:`IntervalTrigger`. postprocess: Callback to postprocess the result dictionaries. Each result dictionary is passed to this callback on the output. This callback can modify the result dictionaries, which are used to output to the log file. filename (str): Name of the log file under the output directory. It can be a format string: the last result dictionary is passed for the formatting. For example, users can use '{iteration}' to separate the log files for different iterations. If the log name is None, it does not output the log to any file. For historical reasons ``log_name`` is also accepted as an alias of this argument. """ def __init__(self, keys=None, trigger=(1, 'epoch'), postprocess=None, filename=None, **kwargs): self._keys = keys self._trigger = trigger_module.get_trigger(trigger) self._postprocess = postprocess self._log = [] log_name, = argument.parse_kwargs( kwargs, ('log_name', 'log'), ) if filename is None: filename = log_name del log_name # avoid accidental use self._log_name = filename self._init_summary() def __call__(self, trainer): # accumulate the observations keys = self._keys observation = trainer.observation summary = self._summary if keys is None: summary.add(observation) else: summary.add({k: observation[k] for k in keys if k in observation}) if trainer.is_before_training or self._trigger(trainer): # output the result stats = self._summary.compute_mean() stats_cpu = {} for name, value in six.iteritems(stats): stats_cpu[name] = float(value) # copy to CPU updater = trainer.updater stats_cpu['epoch'] = updater.epoch stats_cpu['iteration'] = updater.iteration stats_cpu['elapsed_time'] = trainer.elapsed_time if self._postprocess is not None: self._postprocess(stats_cpu) self._log.append(stats_cpu) # write to the log file if self._log_name is not None: log_name = self._log_name.format(**stats_cpu) with utils.tempdir(prefix=log_name, dir=trainer.out) as tempd: path = os.path.join(tempd, 'log.json') with open(path, 'w') as f: json.dump(self._log, f, indent=4) new_path = os.path.join(trainer.out, log_name) shutil.move(path, new_path) # reset the summary for the next output self._init_summary() @property def log(self): """The current list of observation dictionaries.""" return self._log def serialize(self, serializer): if hasattr(self._trigger, 'serialize'): self._trigger.serialize(serializer['_trigger']) try: self._summary.serialize(serializer['_summary']) except KeyError: warnings.warn('The statistics are not saved.') # Note that this serialization may lose some information of small # numerical differences. if isinstance(serializer, serializer_module.Serializer): log = json.dumps(self._log) serializer('_log', log) else: log = serializer('_log', '') self._log = json.loads(log) def _init_summary(self): self._summary = reporter.DictSummary()
5,706
37.560811
79
py
chainer
chainer-master/chainer/training/extensions/plot_report.py
import json from os import path import warnings import numpy import six from chainer import reporter from chainer import serializer as serializer_module from chainer.training import extension from chainer.training import trigger as trigger_module from chainer.utils import argument _available = None def _try_import_matplotlib(): global matplotlib, _available try: import matplotlib # NOQA _available = True except (ImportError, TypeError): _available = False def _check_available(): if _available is None: _try_import_matplotlib() if not _available: warnings.warn('matplotlib is not installed on your environment, ' 'so nothing will be plotted at this time. ' 'Please install matplotlib to plot figures.\n\n' ' $ pip install matplotlib\n') class PlotReport(extension.Extension): """__init__(\ y_keys, x_key='iteration', trigger=(1, 'epoch'), postprocess=None, \ filename='plot.png', marker='x', grid=True) Trainer extension to output plots. This extension accumulates the observations of the trainer to :class:`~chainer.DictSummary` at a regular interval specified by a supplied trigger, and plot a graph with using them. There are two triggers to handle this extension. One is the trigger to invoke this extension, which is used to handle the timing of accumulating the results. It is set to ``1, 'iteration'`` by default. The other is the trigger to determine when to emit the result. When this trigger returns True, this extension appends the summary of accumulated values to the list of past summaries, and writes the list to the log file. Then, this extension makes a new fresh summary object which is used until the next time that the trigger fires. It also adds ``'epoch'`` and ``'iteration'`` entries to each result dictionary, which are the epoch and iteration counts at the output. .. warning:: If your environment needs to specify a backend of matplotlib explicitly, please call ``matplotlib.use`` before calling ``trainer.run``. For example: .. code-block:: python import matplotlib matplotlib.use('Agg') trainer.extend( extensions.PlotReport(['main/loss', 'validation/main/loss'], 'epoch', filename='loss.png')) trainer.run() Then, once one of instances of this extension is called, ``matplotlib.use`` will have no effect. For the details, please see here: https://matplotlib.org/faq/usage_faq.html#what-is-a-backend Args: y_keys (iterable of strs): Keys of values regarded as y. If this is ``None``, nothing is output to the graph. x_key (str): Keys of values regarded as x. The default value is 'iteration'. trigger: Trigger that decides when to aggregate the result and output the values. This is distinct from the trigger of this extension itself. If it is a tuple in the form ``<int>, 'epoch'`` or ``<int>, 'iteration'``, it is passed to :class:`IntervalTrigger`. postprocess: Callback to postprocess the result dictionaries. Figure object, Axes object, and all plot data are passed to this callback in this order. This callback can modify the figure. filename (str): Name of the figure file under the output directory. It can be a format string. For historical reasons ``file_name`` is also accepted as an alias of this argument. marker (str): The marker used to plot the graph. Default is ``'x'``. If ``None`` is given, it draws with no markers. grid (bool): If ``True``, set the axis grid on. The default value is ``True``. """ def __init__(self, y_keys, x_key='iteration', trigger=(1, 'epoch'), postprocess=None, filename=None, marker='x', grid=True, **kwargs): file_name, = argument.parse_kwargs(kwargs, ('file_name', 'plot.png')) if filename is None: filename = file_name del file_name # avoid accidental use _check_available() self._x_key = x_key if isinstance(y_keys, str): y_keys = (y_keys,) self._y_keys = y_keys self._trigger = trigger_module.get_trigger(trigger) self._file_name = filename self._marker = marker self._grid = grid self._postprocess = postprocess self._init_summary() self._data = {k: [] for k in y_keys} @staticmethod def available(): _check_available() return _available def __call__(self, trainer): if self.available(): # Dynamically import pyplot to call matplotlib.use() # after importing chainer.training.extensions import matplotlib.pyplot as plt else: return keys = self._y_keys observation = trainer.observation summary = self._summary if keys is None: summary.add(observation) else: summary.add({k: observation[k] for k in keys if k in observation}) if trainer.is_before_training or self._trigger(trainer): stats = self._summary.compute_mean() stats_cpu = {} for name, value in six.iteritems(stats): stats_cpu[name] = float(value) # copy to CPU updater = trainer.updater stats_cpu['epoch'] = updater.epoch stats_cpu['iteration'] = updater.iteration x = stats_cpu[self._x_key] data = self._data for k in keys: if k in stats_cpu: data[k].append((x, stats_cpu[k])) f = plt.figure() a = f.add_subplot(111) a.set_xlabel(self._x_key) if self._grid: a.grid() for k in keys: xy = data[k] if len(xy) == 0: continue xy = numpy.array(xy) a.plot(xy[:, 0], xy[:, 1], marker=self._marker, label=k) if a.has_data(): if self._postprocess is not None: self._postprocess(f, a, summary) l = a.legend(bbox_to_anchor=(1.05, 1), loc=2, borderaxespad=0.) f.savefig(path.join(trainer.out, self._file_name), bbox_extra_artists=(l,), bbox_inches='tight') plt.close() self._init_summary() def serialize(self, serializer): if isinstance(serializer, serializer_module.Serializer): serializer('_plot_{}'.format(self._file_name), json.dumps(self._data)) else: self._data = json.loads( serializer('_plot_{}'.format(self._file_name), '')) def _init_summary(self): self._summary = reporter.DictSummary()
7,114
33.877451
79
py
chainer
chainer-master/chainer/training/triggers/interval_trigger.py
import warnings class IntervalTrigger(object): """Trigger based on a fixed interval. This trigger accepts iterations divided by a given interval. There are two ways to specify the interval: per iterations and epochs. `Iteration` means the number of updates, while `epoch` means the number of sweeps over the training dataset. Fractional values are allowed if the interval is a number of epochs; the trigger uses the `iteration` and `epoch_detail` attributes defined by the updater. For the description of triggers, see :func:`~chainer.training.get_trigger`. Args: period (int or float): Length of the interval. Must be an integer if unit is ``'iteration'``. unit (str): Unit of the length specified by ``period``. It must be either ``'iteration'`` or ``'epoch'``. """ def __init__(self, period, unit): if unit not in ('epoch', 'iteration'): raise ValueError( 'Trigger unit must be either \'epoch\' or \'iteration\'.') self.period = period self.unit = unit self._previous_iteration = 0 self._previous_epoch_detail = 0. # count is kept for backward compatibility self.count = 0 def __call__(self, trainer): """Decides whether the extension should be called on this iteration. Args: trainer (Trainer): Trainer object that this trigger is associated with. The updater associated with this trainer is used to determine if the trigger should fire. Returns: bool: True if the corresponding extension should be invoked in this iteration. """ updater = trainer.updater if self.unit == 'epoch': epoch_detail = updater.epoch_detail previous_epoch_detail = self._previous_epoch_detail # if previous_epoch_detail is invalid value, # use the value of updater. if previous_epoch_detail < 0: previous_epoch_detail = updater.previous_epoch_detail # count is kept for backward compatibility self.count = epoch_detail // self.period fire = previous_epoch_detail // self.period != \ epoch_detail // self.period else: iteration = updater.iteration previous_iteration = self._previous_iteration # if previous_iteration is invalid value, # guess it from current iteration. if previous_iteration < 0: previous_iteration = iteration - 1 fire = previous_iteration // self.period != \ iteration // self.period # save current values self._previous_iteration = updater.iteration if hasattr(updater, 'epoch_detail'): self._previous_epoch_detail = updater.epoch_detail return fire def serialize(self, serializer): try: self._previous_iteration = serializer( 'previous_iteration', self._previous_iteration) except KeyError: warnings.warn( 'The previous value of iteration is not saved. ' 'IntervalTrigger guesses it using current iteration. ' 'If this trigger is not called at every iteration, ' 'it may not work correctly.') # set a negative value for invalid self._previous_iteration = -1 try: self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) except KeyError: warnings.warn( 'The previous value of epoch_detail is not saved. ' 'IntervalTrigger uses the value of ' 'trainer.updater.previous_epoch_detail. ' 'If this trigger is not called at every iteration, ' 'it may not work correctly.') # set a negative value for invalid self._previous_epoch_detail = -1. def get_training_length(self): return (self.period, self.unit) def __str__(self): """Returns a string describing the class and interval Returns: str: IntervalTrigger(<period>, '<unit>') """ return '{}({}, \'{}\')'.format( self.__class__.__name__, self.period, self.unit )
4,434
34.766129
79
py
chainer
chainer-master/chainer/training/triggers/minmax_value_trigger.py
from chainer import reporter from chainer.training import util class BestValueTrigger(object): """Trigger invoked when specific value becomes best. Args: key (str): Key of value. compare (callable): Compare function which takes current best value and new value and returns whether new value is better than current best. trigger: Trigger that decides the comparison interval between current best value and new value. This must be a tuple in the form of ``<int>, 'epoch'`` or ``<int>, 'iteration'`` which is passed to :class:`~chainer.training.triggers.IntervalTrigger`. """ def __init__(self, key, compare, trigger=(1, 'epoch')): self._key = key self._best_value = None self._interval_trigger = util.get_trigger(trigger) self._init_summary() self._compare = compare def __call__(self, trainer): """Decides whether the extension should be called on this iteration. Args: trainer (~chainer.training.Trainer): Trainer object that this trigger is associated with. The ``observation`` of this trainer is used to determine if the trigger should fire. Returns: bool: ``True`` if the corresponding extension should be invoked in this iteration. """ observation = trainer.observation summary = self._summary key = self._key if key in observation: summary.add({key: observation[key]}) if not self._interval_trigger(trainer): return False stats = summary.compute_mean() value = float(stats[key]) # copy to CPU self._init_summary() if self._best_value is None or self._compare(self._best_value, value): self._best_value = value return True return False def _init_summary(self): self._summary = reporter.DictSummary() def serialize(self, serializer): self._interval_trigger.serialize(serializer['interval_trigger']) self._summary.serialize(serializer['summary']) self._best_value = serializer('best_value', self._best_value) class MaxValueTrigger(BestValueTrigger): """Trigger invoked when specific value becomes maximum. For example you can use this trigger to take snapshot on the epoch the validation accuracy is maximum. Args: key (str): Key of value. The trigger fires when the value associated with this key becomes maximum. trigger: Trigger that decides the comparison interval between current best value and new value. This must be a tuple in the form of ``<int>, 'epoch'`` or ``<int>, 'iteration'`` which is passed to :class:`~chainer.training.triggers.IntervalTrigger`. """ def __init__(self, key, trigger=(1, 'epoch')): super(MaxValueTrigger, self).__init__( key, lambda max_value, new_value: new_value > max_value, trigger) class MinValueTrigger(BestValueTrigger): """Trigger invoked when specific value becomes minimum. For example you can use this trigger to take snapshot on the epoch the validation loss is minimum. Args: key (str): Key of value. The trigger fires when the value associated with this key becomes minimum. trigger: Trigger that decides the comparison interval between current best value and new value. This must be a tuple in the form of ``<int>, 'epoch'`` or ``<int>, 'iteration'`` which is passed to :class:`~chainer.training.triggers.IntervalTrigger`. """ def __init__(self, key, trigger=(1, 'epoch')): super(MinValueTrigger, self).__init__( key, lambda min_value, new_value: new_value < min_value, trigger)
3,888
34.036036
79
py
chainer
chainer-master/chainer/training/triggers/manual_schedule_trigger.py
import warnings class ManualScheduleTrigger(object): """Trigger invoked at specified point(s) of iterations or epochs. This trigger accepts iterations or epochs indicated by given point(s). There are two ways to specify the point(s): iteration and epoch. ``iteration`` means the number of updates, while ``epoch`` means the number of sweeps over the training dataset. Fractional values are allowed if the point is a number of epochs; the trigger uses the ``iteration`` and ``epoch_detail`` attributes defined by the updater. Args: points (int, float, or list of int or float): time of the trigger. Must be an integer or list of integer if unit is ``'iteration'``. unit (str): Unit of the time specified by ``points``. It must be either ``'iteration'`` or ``'epoch'``. Attributes: finished (bool): Flag that indicates whether or not this trigger will fire in the future. This flag is used to determine if the extension should be initialized after resume. """ def __init__(self, points, unit): if unit not in ('epoch', 'iteration'): raise ValueError( 'Trigger unit must be either \'epoch\' or \'iteration\'.') self.points = (points if isinstance(points, list) else [points]) self.unit = unit self.finished = False self._previous_iteration = 0 self._previous_epoch_detail = 0. def __call__(self, trainer): """Decides whether the extension should be called on this iteration. Args: trainer (Trainer): Trainer object that this trigger is associated with. The updater associated with this trainer is used to determine if the trigger should fire. Returns: bool: True if the corresponding extension should be invoked in this iteration. """ updater = trainer.updater if self.unit == 'epoch': epoch_detail = updater.epoch_detail previous_epoch_detail = self._previous_epoch_detail # if previous_epoch_detail is invalid value, # use the value of updater. if previous_epoch_detail < 0: previous_epoch_detail = updater.previous_epoch_detail fire = any( previous_epoch_detail < p <= epoch_detail for p in self.points) if hasattr(self, '_finished_is_tmp'): del self._finished_is_tmp if epoch_detail >= max(self.points): self.finished = True if fire and epoch_detail >= max(self.points): self.finished = True else: iteration = updater.iteration previous_iteration = self._previous_iteration # if previous_iteration is invalid value, # guess it from current iteration. if previous_iteration < 0: previous_iteration = iteration - 1 fire = any( previous_iteration < p <= iteration for p in self.points) if hasattr(self, '_finished_is_tmp'): del self._finished_is_tmp if iteration >= max(self.points): self.finished = True if fire and iteration >= max(self.points): self.finished = True # save current values self._previous_iteration = updater.iteration if hasattr(updater, 'epoch_detail'): self._previous_epoch_detail = updater.epoch_detail return fire def serialize(self, serializer): try: self._previous_iteration = serializer( 'previous_iteration', self._previous_iteration) except KeyError: warnings.warn( 'The previous value of iteration is not saved. ' 'ManualScheduleTrigger guesses it using current iteration. ' 'If this trigger is not called at every iteration, ' 'it may not work correctly.') # set a negative value for invalid self._previous_iteration = -1 try: self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) except KeyError: warnings.warn( 'The previous value of epoch_detail is not saved. ' 'ManualScheduleTrigger uses the value of ' 'trainer.updater.previous_epoch_detail. ' 'If this trigger is not called at every iteration, ' 'it may not work correctly.') # set a negative value for invalid self._previous_epoch_detail = -1. try: self.finished = serializer('finished', self.finished) except KeyError: warnings.warn( 'The flag of finished is not saved. ' 'ManualScheduleTrigger set the flag to `False` to force ' 'initialization and reset in next `__call__`.') # set False to force initialization. self.finished = False self._finished_is_tmp = True
5,231
37.470588
79
py
chainer
chainer-master/chainer/training/triggers/once_trigger.py
import warnings class OnceTrigger(object): """Trigger based on the starting point of the iteration. This trigger accepts only once at starting point of the iteration. There are two ways to specify the starting point: only starting point in whole iteration or called again when training resumed. Args: call_on_resume (bool): Whether the extension is called again or not when restored from a snapshot. It is set to ``False`` by default. Attributes: finished (bool): Flag that indicates whether or not this trigger will fire in the future. This flag is used to determine if the extension should be initialized after resume. """ def __init__(self, call_on_resume=False): self._flag_first = True self._flag_resumed = call_on_resume @property def finished(self): return not (self._flag_first or self._flag_resumed) def __call__(self, trainer): fire = not self.finished self._flag_resumed = False self._flag_first = False return fire def serialize(self, serializer): try: self._flag_first = serializer('_flag_first', self._flag_first) except KeyError: warnings.warn( 'The flag is not saved.' 'OnceTrigger guess it is not first when resumed. ' 'If this trigger is resumed before first called, ' 'it may not work correctly.') self._flag_first = False
1,517
31.297872
77
py
chainer
chainer-master/chainer/training/triggers/__init__.py
# import classes and functions from chainer.training.triggers.early_stopping_trigger import EarlyStoppingTrigger # NOQA from chainer.training.triggers.interval_trigger import IntervalTrigger # NOQA from chainer.training.triggers.manual_schedule_trigger import ManualScheduleTrigger # NOQA from chainer.training.triggers.minmax_value_trigger import BestValueTrigger # NOQA from chainer.training.triggers.minmax_value_trigger import MaxValueTrigger # NOQA from chainer.training.triggers.minmax_value_trigger import MinValueTrigger # NOQA from chainer.training.triggers.once_trigger import OnceTrigger # NOQA from chainer.training.triggers.time_trigger import TimeTrigger # NOQA
684
67.5
91
py
chainer
chainer-master/chainer/training/triggers/early_stopping_trigger.py
import operator import warnings from chainer import reporter from chainer.training import util from chainer.utils import argument class EarlyStoppingTrigger(object): """__init__(\ self, check_trigger=(1, 'epoch'), monitor='main/loss', \ patience=3, mode='auto', verbose=False, \ max_trigger=(100, 'epoch')) Trigger for Early Stopping It can be used as a stop trigger of :class:`~chainer.training.Trainer` to realize *early stopping* technique. This trigger works as follows. Within each *check interval* defined by the ``check_trigger`` argument, it monitors and accumulates the reported value at each iteration. At the end of each interval, it computes the mean of the accumulated values and compares it to the previous ones to maintain the *best* value. When it finds that the best value is not updated for some periods (defined by ``patience``), this trigger fires. Args: monitor (str) : The metric you want to monitor check_trigger: Trigger that decides the comparison interval between current best value and new value. This must be a tuple in the form of ``<int>, 'epoch'`` or ``<int>, 'iteration'`` which is passed to :class:`~chainer.training.triggers.IntervalTrigger`. patience (int) : Counts to let the trigger be patient. The trigger will not fire until the condition is met for successive ``patience`` checks. mode (str) : ``'max'``, ``'min'``, or ``'auto'``. It is used to determine how to compare the monitored values. verbose (bool) : Enable verbose output. If verbose is true, you can get more information max_trigger: Upper bound of the number of training loops .. note:: ``patients`` is also available as an alias of ``patience`` for historical reason. """ def __init__(self, check_trigger=(1, 'epoch'), monitor='main/loss', patience=None, mode='auto', verbose=False, max_trigger=(100, 'epoch'), **kwargs): # `patients` as an alias of `patience` patients, = argument.parse_kwargs(kwargs, ('patients', None)) if patients is None: if patience is None: patience = 3 else: pass else: if patience is None: patience = patients else: raise TypeError( 'Both \'patience\' and \'patients\' arguments are ' 'specified. \'patients\' is an alias of the former. ' 'Specify only \'patience\'.') self.count = 0 self.patience = patience self.monitor = monitor self.verbose = verbose self.already_warning = False self._max_trigger = util.get_trigger(max_trigger) self._interval_trigger = util.get_trigger(check_trigger) self._init_summary() if mode == 'max': self._compare = operator.gt elif mode == 'min': self._compare = operator.lt else: if 'accuracy' in monitor: self._compare = operator.gt else: self._compare = operator.lt if self._compare == operator.gt: if verbose: print('early stopping: operator is greater') self.best = float('-inf') else: if verbose: print('early stopping: operator is less') self.best = float('inf') def __call__(self, trainer): """Decides whether the training loop should be stopped. Args: trainer (~chainer.training.Trainer): Trainer object that this trigger is associated with. The ``observation`` of this trainer is used to determine if the trigger should fire. Returns: bool: ``True`` if the training loop should be stopped. """ observation = trainer.observation summary = self._summary if self.monitor in observation: summary.add({self.monitor: observation[self.monitor]}) if self._max_trigger(trainer): return True if not self._interval_trigger(trainer): return False if self.monitor not in observation.keys(): warnings.warn('{} is not in observation'.format(self.monitor)) return False stat = self._summary.compute_mean() current_val = stat[self.monitor] self._init_summary() if self._compare(current_val, self.best): self.best = current_val self.count = 0 else: self.count += 1 if self._stop_condition(): if self.verbose: print('Epoch {}: early stopping'.format(trainer.updater.epoch)) return True return False def _stop_condition(self): return self.count >= self.patience def _init_summary(self): self._summary = reporter.DictSummary() def get_training_length(self): return self._max_trigger.get_training_length()
5,200
32.127389
79
py
chainer
chainer-master/chainer/training/triggers/time_trigger.py
class TimeTrigger(object): """Trigger based on a fixed time interval. This trigger accepts iterations with a given interval time. Args: period (float): Interval time. It is given in seconds. """ def __init__(self, period): self._period = period self._next_time = self._period def __call__(self, trainer): if self._next_time < trainer.elapsed_time: self._next_time += self._period return True else: return False def serialize(self, serializer): self._next_time = serializer('next_time', self._next_time)
622
23.92
66
py
chainer
chainer-master/chainer/backends/intel64.py
from __future__ import absolute_import import numpy import chainer from chainer import _backend from chainer.backends import _cpu from chainer.configuration import config _ideep_version = None _error = None try: import ideep4py as ideep # NOQA from ideep4py import mdarray # type: ignore # NOQA _ideep_version = 2 if hasattr(ideep, '__version__') else 1 except ImportError as e: _error = e _ideep_version = None class mdarray(object): # type: ignore pass # for type testing class Intel64Device(_backend.Device): """Device for Intel64 (Intel Architecture) backend with iDeep""" xp = numpy name = '@intel64' supported_array_types = (numpy.ndarray, mdarray) __hash__ = _backend.Device.__hash__ def __init__(self): check_ideep_available() super(Intel64Device, self).__init__() @staticmethod def from_array(array): if isinstance(array, mdarray): return Intel64Device() return None def __eq__(self, other): return isinstance(other, Intel64Device) def __repr__(self): return '<{}>'.format(self.__class__.__name__) def send_array(self, array): if isinstance(array, ideep.mdarray): return array if not isinstance(array, numpy.ndarray): array = _cpu._to_cpu(array) # to numpy.ndarray if (isinstance(array, numpy.ndarray) and array.ndim in (1, 2, 4) and 0 not in array.shape): # TODO(kmaehashi): Remove ndim validation once iDeep has fixed. # Currently iDeep only supports (1, 2, 4)-dim arrays. # Note that array returned from `ideep.array` may not be an # iDeep mdarray, e.g., when the dtype is not float32. array = ideep.array(array, itype=ideep.wgt_array) return array def is_array_supported(self, array): return isinstance(array, (numpy.ndarray, mdarray)) # ------------------------------------------------------------------------------ # ideep configuration # ------------------------------------------------------------------------------ _SHOULD_USE_IDEEP = { '==always': {'always': True, 'auto': False, 'never': False}, '>=auto': {'always': True, 'auto': True, 'never': False}, } def is_ideep_available(): """Returns if iDeep is available. Returns: bool: ``True`` if the supported version of iDeep is installed. """ return _ideep_version is not None and _ideep_version == 2 def check_ideep_available(): """Checks if iDeep is available. When iDeep is correctly set up, nothing happens. Otherwise it raises ``RuntimeError``. """ if _ideep_version is None: # If the error is missing shared object, append a message to # redirect to the ideep website. msg = str(_error) if 'cannot open shared object file' in msg: msg += ('\n\nEnsure iDeep requirements are satisfied: ' 'https://github.com/intel/ideep') raise RuntimeError( 'iDeep is not available.\n' 'Reason: {}: {}'.format(type(_error).__name__, msg)) elif _ideep_version != 2: raise RuntimeError( 'iDeep is not available.\n' 'Reason: Unsupported iDeep version ({})'.format(_ideep_version)) def should_use_ideep(level): """Determines if we should use iDeep. This function checks ``chainer.config.use_ideep`` and availability of ``ideep4py`` package. Args: level (str): iDeep use level. It must be either ``'==always'`` or ``'>=auto'``. ``'==always'`` indicates that the ``use_ideep`` config must be ``'always'`` to use iDeep. Returns: bool: ``True`` if the caller should use iDeep. """ if not is_ideep_available(): return False # TODO(niboshi): # Add lowest_version argument and compare with ideep version. # Currently ideep does not provide a way to retrieve its version. if level not in _SHOULD_USE_IDEEP: raise ValueError('invalid iDeep use level: %s ' '(must be either of "==always" or ">=auto")' % repr(level)) flags = _SHOULD_USE_IDEEP[level] use_ideep = config.use_ideep if use_ideep not in flags: raise ValueError('invalid use_ideep configuration: %s ' '(must be either of "always", "auto", or "never")' % repr(use_ideep)) return flags[use_ideep] def inputs_all_ready(inputs, supported_ndim=(2, 4)): """Checks if input arrays are supported for an iDeep primitive. Before calling an iDeep primitive (e.g., ``ideep4py.linear.Forward``), you need to make sure that all input arrays are ready for the primitive by calling this function. Information to be checked includes array types, dimesions and data types. The function checks ``inputs`` info and ``supported_ndim``. Inputs to be tested can be any of ``Variable``, ``numpy.ndarray`` or ``ideep4py.mdarray``. However, all inputs to iDeep primitives must be ``ideep4py.mdarray``. Callers of iDeep primitives are responsible of converting all inputs to ``ideep4py.mdarray``. Args: inputs (sequence of arrays or variables): Inputs to be checked. supported_ndim (tuple of ints): Supported ndim values for the iDeep primitive. Returns: bool: ``True`` if all conditions meet. """ def _is_supported_array_type(a): return isinstance(a, ideep.mdarray) or ideep.check_type([a]) if not is_ideep_available(): return False inputs = [x.data if isinstance(x, chainer.variable.Variable) else x for x in inputs] return (ideep.check_ndim(inputs, supported_ndim) and all([_is_supported_array_type(a) for a in inputs]))
5,920
30.833333
80
py
chainer
chainer-master/chainer/backends/cuda.py
"""Device, context and memory management on CuPy. .. note:: The package ``chainer.cuda`` has been renamed to :mod:`chainer.backends.cuda` as of v4.0.0, but the previous module path ``chainer.cuda`` is also available. Chainer uses `CuPy <https://cupy.chainer.org/>`_ (with very thin wrapper) to exploit the speed of GPU computation. Following modules and classes defined in CuPy are imported to :mod:`chainer.backends.cuda` module for convenience (refer to this table when reading chainer's source codes). ===================================== ================================= imported name original name ===================================== ================================= ``chainer.backends.cuda.cupy`` :mod:`cupy` ``chainer.backends.cuda.cupyx`` :mod:`cupyx` ``chainer.backends.cuda.ndarray`` :class:`cupy.ndarray` ``chainer.backends.cuda.cupy.cuda`` :mod:`cupy.cuda` ``chainer.backends.cuda.Device`` :class:`cupy.cuda.Device` ``chainer.backends.cuda.Event`` :class:`cupy.cuda.Event` ``chainer.backends.cuda.Stream`` :class:`cupy.cuda.Stream` ===================================== ================================= Chainer replaces the default allocator of CuPy by its memory pool implementation. It enables us to reuse the device memory over multiple forward/backward computations, and temporary arrays for consecutive elementwise operations. """ import binascii import functools import itertools import os import threading import time import typing as tp # NOQA import warnings import numpy import six import chainer from chainer import _backend from chainer.backends import _cpu from chainer.backends import intel64 from chainer.configuration import config from chainer import types # NOQA import chainerx available = False # type: bool cudnn_enabled = False # type: bool _cupy_major = 0 # type: bool try: import cupy from cupy import cuda # NOQA from cupy.cuda import cublas # NOQA import cupyx # NOQA import cupyx.scipy.linalg # NOQA import cupyx.scipy.special # NOQA from cupy import ndarray # type: ignore # NOQA from cupy.cuda import Device # type: ignore # NOQA from cupy.cuda import Event # type: ignore # NOQA from cupy.cuda import Stream # type: ignore # NOQA # Alias for ignoring the warning in setup.cfg try: from cupy._util import PerformanceWarning as _PerformanceWarning # NOQA except ImportError: from cupy.util import PerformanceWarning as _PerformanceWarning # NOQA available = True except Exception as e: _resolution_error = e class ndarray(object): # type: ignore # for type testing @property def shape(self) -> types.Shape: pass @property def device(self) -> 'Device': pass def get(self, stream: tp.Optional['Stream'] = None) -> numpy.ndarray: pass def set( self, arr: numpy.ndarray, stream: tp.Optional['Stream'] = None ) -> None: pass class Device(object): # type: ignore # for type testing def __init__(self, device: tp.Optional[int] = None) -> None: pass def __enter__(self) -> 'Device': pass def __exit__(self, *args: tp.Any) -> None: pass class Event(object): # type: ignore # for type testing pass class Stream(object): # type: ignore # for type testing pass # for `xp is chainer.backends.cuda.cupy` to always work cupy = object() # Dummy class for ignoring cupy.util.PerformanceWarning in setup.cfg. class _PerformanceWarning(Warning): pass if available: _cupy_major = int(cupy.__version__.split('.')[0]) _cudnn_disabled_by_user = int(os.environ.get('CHAINER_CUDNN', '1')) == 0 try: if 7 < _cupy_major: import cupy.cuda.cudnn import cupy.cudnn cudnn = cupy.cudnn # type: tp.Optional[types.ModuleType] libcudnn = cupy.cuda.cudnn # type: tp.Any # NOQA cudnn_enabled = not _cudnn_disabled_by_user except Exception as e: _resolution_error = e # for `chainer.backends.cuda.libcudnn` to always work libcudnn = object() def check_cuda_available(): """Checks if CUDA is available. When CUDA is correctly set up, nothing happens. Otherwise it raises ``RuntimeError``. """ if not available: msg = ('CUDA environment is not correctly set up\n' '(see https://github.com/chainer/chainer#installation).') msg += str(_resolution_error) raise RuntimeError(msg) if (not cudnn_enabled and not _cudnn_disabled_by_user and not getattr(check_cuda_available, '_already_warned', False)): warnings.warn( 'cuDNN is not enabled.\n' 'Please reinstall CuPy after you install cudnn\n' '(see https://docs-cupy.chainer.org/en/stable/install.html' '#install-cudnn).') check_cuda_available._already_warned = True class DummyDeviceType(object): """Dummy device class that does nothing with cupy.cuda.Device interface. This class is used to represent CPU device. """ id = -1 def __init__(self): pass def __int__(self): return -1 def __enter__(self): return self def __exit__(self, *args): pass def use(self): pass def synchronize(self): pass def __eq__(self, other): return isinstance(other, DummyDeviceType) def __ne__(self, other): return not (self == other) DummyDevice = DummyDeviceType() # ------------------------------------------------------------------------------ # Global states # ------------------------------------------------------------------------------ if available: # This is for backward compatibility memory_pool = cupy.get_default_memory_pool() pinned_memory_pool = cupy.get_default_pinned_memory_pool() _integer_types = six.integer_types + (numpy.integer,) # ------------------------------------------------------------------------------ # Device # ------------------------------------------------------------------------------ class GpuDevice(_backend.Device): """Device for GPU (CuPy) backend""" xp = cupy supported_array_types = (ndarray,) __hash__ = _backend.Device.__hash__ def __init__(self, device): check_cuda_available() assert isinstance(device, Device) super(GpuDevice, self).__init__() self.device = device @staticmethod def from_device_id(device_id): """Returns a :class:`~chainer.backend.GpuDevice` corresponding \ to the CUDA device ID. """ check_cuda_available() if not (isinstance(device_id, _integer_types) and device_id >= 0): raise ValueError('Invalid CUDA device ID: {}'.format(device_id)) return GpuDevice(Device(device_id)) @staticmethod def from_array(array): if isinstance(array, ndarray) and array.device is not None: return GpuDevice(array.device) return None @property def name(self): return '@cupy:{}'.format(self.device.id) def __eq__(self, other): return isinstance(other, GpuDevice) and other.device == self.device def __repr__(self): return '<{} (cupy):{}>'.format( self.__class__.__name__, self.device.id) def create_context(self): # Creates a new cuda.Device instance because a single cuda.Device # instance cannot be used across threads. return Device(self.device.id) def send_array(self, array): return _array_to_gpu(array, self.device, None) def use(self): self.device.use() def is_array_supported(self, array): return isinstance(array, ndarray) and self.device == array.device # ------------------------------------------------------------------------------ # Global states # ------------------------------------------------------------------------------ def get_device_from_id(device_id: tp.Optional[int]) -> Device: """Gets the device from an ID integer. Args: device_id (int or None): The ID of the device which this function returns. """ if device_id is not None: if device_id >= 0: check_cuda_available() return Device(int(device_id)) return DummyDevice def get_device_from_array(*arrays: ndarray) -> Device: """Gets the device from a list of CuPy array or a single CuPy array. .. deprecated:: v6.0.0 This API is deprecated. Please use :func:`chainer.backend.get_device_from_array` instead. The device on which the given CuPy array reside is returned. .. note:: This method only recognizes :class:`cupy.ndarray`\\ s in arguments. Especially note that, unlike :func:`get_array_module`, this method does not recognize :class:`~chainer.Variable` objects. If you need to get device from the :class:`~chainer.Variable` instance ``v``, you need to use ``get_device_from_array(v.array)``. Args: arrays (:class:`cupy.ndarray` or list of :class:`cupy.ndarray`): A CuPy array which this function returns the device corresponding to. If a list of :class:`cupy.ndarray`\\ s are given, it returns the first device object of an array in the list. """ for array in arrays: if isinstance(array, ndarray) and array.device is not None: return array.device return DummyDevice def get_device(*args): """Gets the device from a device object, an ID integer or an array object. .. note:: This API is deprecated since v3.0.0. Please use :func:`~chainer.backends.cuda.get_device_from_id` or :func:`~chainer.backends.cuda.get_device_from_array` instead. This is a convenient utility to select a correct device if the type of ``arg`` is unknown (i.e., one can use this function on arrays that may be on CPU or GPU). The returned device object supports the context management protocol of Python for the *with* statement. Args: args: Values to specify a GPU device. The first device object, integer or :class:`cupy.ndarray` object is used to select a device. If it is a device object, it is returned. If it is an integer, the corresponding device is returned. If it is a CuPy array, the device on which this array reside is returned. If any arguments are neither integers nor CuPy arrays, a dummy device object representing CPU is returned. Returns: Device object specified by given ``args``. .. seealso:: See :class:`cupy.cuda.Device` for the device selection not by arrays. """ warnings.warn('get_device is deprecated. Please use get_device_from_id or' ' get_device_from_array instead.', DeprecationWarning) return _get_cuda_device(*args) def _get_cuda_device(*args): # Returns cuda.Device or DummyDevice. for arg in args: if type(arg) is not bool and isinstance(arg, _integer_types): check_cuda_available() return Device(arg) if isinstance(arg, ndarray): if arg.device is None: continue return arg.device if available and isinstance(arg, Device): return arg # NOTE: This function returns DummyDevice for both NumPy and ChainerX return DummyDevice def _get_device_or_current( device: tp.Optional[types.CudaDeviceSpec] ) -> Device: # Returns cuda.Device. # - If cuda.Device instance, it's returned intact. # - If None, the current device is returned. # - If non-negative integer, cuda.Device is returned. # - Otherwise: error. if device is None: return cuda.Device() if isinstance(device, Device): return device if not (isinstance(device, _integer_types) and device >= 0): raise ValueError('Invalid CUDA device specifier: {}'.format(device)) return cuda.Device(int(device)) # ------------------------------------------------------------------------------ # cupy.ndarray allocation and copy # ------------------------------------------------------------------------------ def to_gpu(array, device=None, stream=None): """Copies the given CPU array to the specified device. Args: array (*array*, None, list or tuple): Array or arrays to be sent to GPU. device: CUDA device specifier. If ``None`` or :data:`cuda.DummyDevice`, the arrays will be copied to the current CUDA device. stream (~cupy.cuda.Stream): *(deprecated since v3.0.0)* CUDA stream. If not ``None``, the copy runs asynchronously. Returns: cupy.ndarray, list or tuple: Array or arrays on GPU. If some of the arrays are already on GPU, then this function just returns those arrays without performing any copy. If input arrays include `None`, it is returned as `None` as is. """ if stream is not None: warnings.warn( 'The stream option is deprecated in chainer.backends.cuda.to_gpu. ' 'Please remove it.', DeprecationWarning) check_cuda_available() if device is DummyDevice: device = cuda.Device() else: device = _get_device_or_current(device) return _backend._convert_arrays( array, lambda arr: _array_to_gpu(arr, device, stream)) def _array_to_gpu(array, device, stream): if array is None: return None if isinstance(array, chainerx.ndarray): # TODO(niboshi): Update this logic once both CuPy and ChainerX support # the array interface. if array.device.backend.name == 'cuda': # Convert to cupy.ndarray on the same device as source array array = chainerx._to_cupy(array) else: array = chainerx.to_numpy(array) elif isinstance(array, (numpy.number, numpy.bool_)): array = numpy.asarray(array) elif isinstance(array, intel64.mdarray): array = numpy.asarray(array) if isinstance(array, ndarray): if array.device == device: return array is_numpy = False elif isinstance(array, numpy.ndarray): is_numpy = True else: raise TypeError( 'The array sent to gpu must be an array or a NumPy scalar.' '\nActual type: {0}.'.format(type(array))) if stream is not None: with device: with stream: if is_numpy: return cupy.asarray(array) # Need to make a copy when an array is copied to another device return cupy.array(array, copy=True) with device: if is_numpy: return cupy.asarray(array) # Need to make a copy when an array is copied to another device return cupy.array(array, copy=True) def to_cpu(array, stream=None): """Copies the given GPU array to host CPU. Args: array (*array*, None, list or tuple): Array or arrays to be sent to CPU. stream (cupy.cuda.Stream): CUDA stream. Returns: numpy.ndarray, list or tuple: Array on CPU. If some of the arrays are already on CPU, then this function just returns those arrays without performing any copy. If input arrays include `None`, it is returned as `None` as is. """ return _backend._convert_arrays( array, lambda arr: _array_to_cpu(arr, stream)) def _array_to_cpu(array, stream): if array is None: return None if isinstance(array, ndarray): check_cuda_available() with get_device_from_array(array): return array.get(stream) return _cpu._array_to_cpu(array) def copy(array, out=None, out_device=None, stream=None): """Copies a :class:`cupy.ndarray` object using the default stream. This function can copy the device array to the destination array on another device. Args: array (cupy.ndarray): Array to be copied. out (cupy.ndarray): Destination array. If it is not ``None``, then ``out_device`` argument is ignored. out_device: Destination device specifier. Actual device object is obtained by passing this value to :func:`get_device`. stream (cupy.cuda.Stream): CUDA stream. Returns: cupy.ndarray: Copied array. If ``out`` is not specified, then the array is allocated on the device specified by ``out_device`` argument. """ # TODO(niboshi): Update docstring not to mention deprecated `get_device` check_cuda_available() assert stream is None # TODO(beam2d): FIX IT if out is None: if out_device is None: out_device = array with chainer.get_device(out_device): out = cupy.empty_like(array) with get_device_from_array(array): cupy.copyto(out, array) return out # ------------------------------------------------------------------------------ # Function result memoization # ------------------------------------------------------------------------------ def memoize(for_each_device=False): """Makes a function memoizing the result for each argument and device. This is a similar version of :func:`cupy.memoize`. The difference is that this function can be used in the global scope even if CUDA is not available. In such case, this function does nothing. .. note:: This decorator acts as a dummy if CUDA is not available. It cannot be used for general purpose memoization even if ``for_each_device`` is set to False. """ if available: return cupy.memoize(for_each_device) def dummy_decorator(f): @functools.wraps(f) def ret(*args, **kwargs): return f(*args, **kwargs) return ret return dummy_decorator def clear_memo(): """Clears the memoized results for all functions decorated by memoize. This function works like :func:`cupy.clear_memo` as a counterpart for :func:`chainer.backends.cuda.memoize`. It can be used even if CUDA is not available. In such a case, this function does nothing. """ if available: cupy.clear_memo() # ------------------------------------------------------------------------------ # Kernel definition utility # ------------------------------------------------------------------------------ @memoize() def elementwise(in_params, out_params, operation, name, **kwargs): """Creates an elementwise kernel function. This function uses :func:`~chainer.backends.cuda.memoize` to cache the kernel object, i.e. the resulting kernel object is cached for each argument combination and CUDA device. The arguments are the same as those for :class:`cupy.ElementwiseKernel`, except that the ``name`` argument is mandatory. """ check_cuda_available() return cupy.ElementwiseKernel( in_params, out_params, operation, name, **kwargs) @memoize() def reduce(in_params, out_params, map_expr, reduce_expr, post_map_expr, identity, name, **kwargs): """Creates a global reduction kernel function. This function uses :func:`~chainer.backends.cuda.memoize` to cache the resulting kernel object, i.e. the resulting kernel object is cached for each argument combination and CUDA device. The arguments are the same as those for :class:`cupy.ReductionKernel`, except that the ``name`` argument is mandatory. """ check_cuda_available() return cupy.ReductionKernel( in_params, out_params, map_expr, reduce_expr, post_map_expr, identity, name, **kwargs) @memoize() def raw(code, name, *args, **kwargs): """Creates a raw kernel function. This function uses :func:`~chainer.backends.cuda.memoize` to cache the resulting kernel object, i.e. the resulting kernel object is cached for each argument combination and CUDA device. The arguments are the same as those for :class:`cupy.RawKernel`. """ check_cuda_available() return cupy.RawKernel(code, name, *args, **kwargs) # ------------------------------------------------------------------------------ # numpy/cupy compatible coding # ------------------------------------------------------------------------------ def get_array_module(*args): """Gets an appropriate one from :mod:`numpy` or :mod:`cupy`. This is almost equivalent to :func:`cupy.get_array_module`. The differences are that this function can be used even if CUDA is not available and that it will return their data arrays' array module for :class:`~chainer.Variable` arguments. .. deprecated:: v5.0.0 This API is deprecated. Please use :func:`~chainer.backend.get_array_module` instead. Args: args: Values to determine whether NumPy or CuPy should be used. Returns: module: :mod:`cupy` or :mod:`numpy` is returned based on the types of the arguments. """ return chainer.backend.get_array_module(*args) def get_max_workspace_size(): """Gets the workspace size for cuDNN. Check "cuDNN Library User Guide" for detail. Returns: int: The workspace size for cuDNN. """ # To avoid error on no cuDNN environment if cudnn_enabled: return cudnn.get_max_workspace_size() return 0 def set_max_workspace_size(size): """Sets the workspace size for cuDNN. Check "cuDNN Library User Guide" for detail. Args: size: The workspace size for cuDNN. """ # To avoid error on no cuDNN environment if cudnn_enabled: cudnn.set_max_workspace_size(size) def fuse(*args, **kwargs): """Function fusing decorator. It calls :func:`cupy.fuse` when CuPy is available to make fused function and does nothing otherwise. .. seealso:: :func:`cupy.fuse` """ if available: return cupy.fuse(*args, **kwargs) elif len(args) == 1 and len(kwargs) == 0 and callable(args[0]): return args[0] else: return lambda f: f # ------------------------------------------------------------------------------ # cuDNN # ------------------------------------------------------------------------------ _SHOULD_USE_CUDNN = { '==always': {'always': True, 'auto': False, 'never': False}, '>=auto': {'always': True, 'auto': True, 'never': False}, } _cudnn_version = cuda.cudnn.getVersion() if cudnn_enabled else -1 def should_use_cudnn(level, lowest_version=0): """Determines if we should use cuDNN. This function checks ``chainer.config.use_cudnn``, ``chainer.backends.cuda.cudnn_enabled``, and the cuDNN version. Note that ``cudnn_enabled`` flag is fixed at loading of :mod:`chainer` module. Args: level (str): cuDNN use level. It must be either ``'==always'`` or ``'>=auto'``. ``'==always'`` indicates that the ``use_cudnn`` config must be ``'always'`` to use cuDNN. lowest_version (int): Required lowest cuDNN version. It must be non-negative. Returns: bool: ``True`` if the caller should use cuDNN. """ if _cudnn_version < lowest_version: return False if level not in _SHOULD_USE_CUDNN: raise ValueError('invalid cuDNN use level: %s ' '(must be either of "==always" or ">=auto")' % repr(level)) flags = _SHOULD_USE_CUDNN[level] use_cudnn = config.use_cudnn if use_cudnn not in flags: raise ValueError('invalid use_cudnn configuration: %s ' '(must be either of "always", "auto", or "never")' % repr(use_cudnn)) return flags[use_cudnn] _tensor_core_flag = {'always': True, 'auto': None, 'never': False} def should_use_cudnn_tensor_core(dtype): """Determines if Tensor Core should be used. Args: dtype (numpy.dtype): data type of input tensor. Returns: bool: ``True`` if Tensor Core should be used. """ use_cudnn_tensor_core = config.use_cudnn_tensor_core if use_cudnn_tensor_core not in _tensor_core_flag: raise ValueError('invalid use_cudnn_tensor_core configuration: %s ' '(must be either of "always", "auto", or "never")' % repr(use_cudnn_tensor_core)) use_tensor_core = _tensor_core_flag[use_cudnn_tensor_core] if use_tensor_core is None: use_tensor_core = cudnn.is_tensor_core_available(dtype) return use_tensor_core # ------------------------------------------------------------------------------ # cupy.cudnn utility # ------------------------------------------------------------------------------ def get_cudnn_dropout_states(): if not cudnn_enabled: raise RuntimeError('cuDNN is not enabled.') thread_id = threading.current_thread().ident return get_cudnn_dropout_states_core(thread_id) _dropout_states_count = itertools.count() @memoize(for_each_device=True) def get_cudnn_dropout_states_core(thread_id): states_id = next(_dropout_states_count) seed = os.getenv('CHAINER_SEED') if seed is None: try: seed_str = binascii.hexlify(os.urandom(8)) seed = numpy.uint64(int(seed_str, 16)) except NotImplementedError: seed = numpy.uint64(time.clock() * 1000000) else: seed = numpy.uint64(seed) seed += numpy.uint64(states_id) return cudnn.DropoutStates(None, seed) def _get_cudnn_tensor_layout_x(x_layout): if x_layout == chainer.memory_layouts.CUDNN_CHANNEL_FIRST_X: return cuda.cudnn.CUDNN_TENSOR_NCHW assert x_layout == chainer.memory_layouts.CUDNN_CHANNEL_LAST_X return cuda.cudnn.CUDNN_TENSOR_NHWC def _get_cudnn_tensor_layout_w(w_layout): if w_layout == chainer.memory_layouts.CUDNN_CHANNEL_FIRST_W: return cuda.cudnn.CUDNN_TENSOR_NCHW assert w_layout == chainer.memory_layouts.CUDNN_CHANNEL_LAST_W return cuda.cudnn.CUDNN_TENSOR_NHWC
26,329
30.915152
80
py
chainer
chainer-master/chainer/backends/_chainerx.py
import numpy import chainer from chainer import _backend from chainer.backends import _cpu from chainer.backends import cuda from chainer.backends import intel64 import chainerx class ChainerxDevice(_backend.Device): """Device for ChainerX backend""" xp = chainerx supported_array_types = (chainerx.ndarray,) __hash__ = _backend.Device.__hash__ def __init__(self, device: 'chainerx.Device') -> None: assert isinstance(device, chainerx.Device) super(ChainerxDevice, self).__init__() self.device = device # type: chainerx.Device @staticmethod def from_array(array): if isinstance(array, chainerx.ndarray) and array.device is not None: return ChainerxDevice(array.device) return None @staticmethod def from_fallback_device(device): """Returns a :class:`~chainer.backend.ChainerxDevice` corresponding \ to the fallback device. .. seealso:: :data:`~chainer.backend.ChainerxDevice.fallback_device` """ assert isinstance(device, _backend.Device) if isinstance(device, _cpu.CpuDevice): return ChainerxDevice(chainerx.get_device('native', 0)) if isinstance(device, cuda.GpuDevice): return ChainerxDevice( chainerx.get_device('cuda', device.device.id)) raise RuntimeError( 'Only CPU or GPU devices are allowed. ' 'Actual: {}'.format(device)) @property def name(self): return self.device.name @property def fallback_device(self): """Fallback device. A fallback device is either a :class:`~chainer.backend.CpuDevice` or a :class:`~chainer.backend.GpuDevice` which shares the same physical device with the original ChainerX device. For example, the fallback device of ``native:0`` ChainerX device is :class:`~chainer.backend.CpuDevice`. The fallback device of ``cuda:1`` ChainerX device is :class:`~chainer.backend.GpuDevice` with device ID 1. """ backend_name = self.device.backend.name if backend_name == 'native': return _cpu.CpuDevice() if backend_name == 'cuda': return cuda.GpuDevice.from_device_id(self.device.index) raise RuntimeError( 'Only \'native\' or \'cuda\' devices have corresponding fallback ' 'devices. Actual: {}'.format(backend_name)) def __eq__(self, other): return ( isinstance(other, ChainerxDevice) and other.device == self.device) def __repr__(self): return '<{} {}>'.format( self.__class__.__name__, self.device.name) def create_context(self): # Returns a context that sets the default device. return chainerx.using_device(self.device) def send_array(self, array): device = self.device if isinstance(array, chainerx.ndarray): if array.device is device: return array return array.to_device(device) return _array_to_chainerx(array, device) def use(self): chainerx.set_default_device(self.device) def is_array_supported(self, array): return ( isinstance(array, chainerx.ndarray) and self.device == array.device) def to_chx(array): """Converts an array or arrays to ChainerX. Destination ChainerX devices are chosen according to the types of input arrays. """ return _backend._convert_arrays(array, _array_to_chainerx) def from_chx(array): """Converts an array or arrays from ChainerX to NumPy or CuPy ones. Destination array types are chosen such that no copies occur. """ return _backend._convert_arrays(array, _array_from_chainerx) def _get_chainerx_device(device_spec): # Returns chainerx.Device if isinstance(device_spec, chainerx.Device): return device_spec return chainerx.get_device(device_spec) def _array_to_chainerx(array, device=None): # If device is None, appropriate device is chosen according to the input # arrays. assert device is None or isinstance(device, chainerx.Device) if array is None: return None if array.dtype not in chainerx.all_dtypes: raise TypeError( 'Dtype {} is not supported in ChainerX.'.format(array.dtype.name)) if isinstance(array, chainerx.ndarray): if device is None: return array if device is array.device: return array return array.to_device(device) if isinstance(array, numpy.ndarray): if device is None: device = chainerx.get_device('native', 0) return chainerx.array(array, device=device, copy=False) if isinstance(array, cuda.ndarray): if device is None: device = chainerx.get_device('cuda', array.device.id) elif device.backend.name != 'cuda': # cupy to non-cuda backend # TODO(niboshi): Remove conversion to numpy when both CuPy and # ChainerX support the array interface. array = _cpu._to_cpu(array) return chainerx.array(array, device=device, copy=False) elif device.index != array.device.id: # cupy to cuda backend but different device array = cuda.to_gpu(array, device=device.index) # cupy to cuda backend with the same device return chainerx._core._fromrawpointer( array.data.mem.ptr, array.shape, array.dtype, array.strides, device, array.data.ptr - array.data.mem.ptr, array) if isinstance(array, intel64.mdarray): return _array_to_chainerx(numpy.array(array), device) if numpy.isscalar(array): return chainerx.asarray(array) raise TypeError( 'Array cannot be converted into chainerx.ndarray' '\nActual type: {0}.'.format(type(array))) def _array_from_chainerx(array): if array is None: return None if not isinstance(array, chainerx.ndarray): if isinstance(array, chainer.get_array_types()): return array raise TypeError( 'Tried to convert to a non-ChainerX array from an invalid type: ' '{}'.format(type(array))) backend_name = array.device.backend.name if backend_name == 'native': return _cpu._to_cpu(array) if backend_name == 'cuda': return cuda.to_gpu(array, array.device.index) raise ValueError( 'Only ChainerX arrays with native or cuda backends can be converted ' 'to non-ChainerX arrays.\nActual: {0}.'.format(backend_name))
6,701
32.343284
78
py
chainer
chainer-master/chainer/backends/__init__.py
from chainer.backends import cuda # NOQA from chainer.backends import intel64 # NOQA # TODO(niboshi): Refactor registration of backend modules for functions like # chainer.get_device().
190
26.285714
76
py
chainer
chainer-master/chainer/backends/_cpu.py
import numpy from chainer import _backend # TODO(kmaehashi): `from chainer.backends import cuda` causes circular imports. # Surprisingly, `import chianer.backends` works as a workaround to avoid, but # we should fix circular dependencies themselves around `chainer.backends.*`. import chainer.backends import chainerx class CpuDevice(_backend.Device): """Device for CPU (NumPy) backend""" name = '@numpy' xp = numpy supported_array_types = (numpy.ndarray,) __hash__ = _backend.Device.__hash__ @staticmethod def from_array(array): if isinstance(array, numpy.ndarray): return CpuDevice() return None def __eq__(self, other): return isinstance(other, CpuDevice) def __repr__(self): return '<{} (numpy)>'.format(self.__class__.__name__) def send_array(self, array): return _array_to_cpu(array) def is_array_supported(self, array): return isinstance(array, numpy.ndarray) def _to_cpu(array): """Converts an array or arrays to NumPy.""" return _backend._convert_arrays(array, _array_to_cpu) def _array_to_cpu(array): if array is None: return None if isinstance(array, numpy.ndarray): return array if isinstance(array, chainer.backends.intel64.mdarray): return numpy.asarray(array) if isinstance(array, chainerx.ndarray): return chainerx.to_numpy(array, copy=False) if isinstance(array, chainer.backends.cuda.ndarray): with chainer.backends.cuda.get_device_from_array(array): return array.get() if numpy.isscalar(array): return numpy.asarray(array) raise TypeError( 'Array cannot be converted into an numpy.ndarray' '\nActual type: {0}.'.format(type(array)))
1,783
27.774194
79
py
chainer
chainer-master/chainer/initializers/uniform.py
import numpy from chainer import backend from chainer import initializer from chainer.utils import argument # Original code forked from MIT licensed keras project # https://github.com/fchollet/keras/blob/master/keras/initializations.py class Uniform(initializer.Initializer): """Initializes array with a scaled uniform distribution. Each element of the array is initialized by the value drawn independently from uniform distribution :math:`[-scale, scale]`. Attributes: scale (float): A constant that determines the scale of the uniform distribution. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=0.05, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(Uniform, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) if self.rng is None: device = backend.get_device_from_array(array) array[...] = device.xp.random.uniform( low=-self.scale, high=self.scale, size=array.shape) else: backend.copyto(array, self.rng.uniform( low=-self.scale, high=self.scale, size=array.shape).astype(array.dtype, copy=False)) class LeCunUniform(initializer.Initializer): """Initializes array with a scaled uniform distribution. Each element of the array is initialized by the value drawn independently from uniform distribution :math:`[-s, s]` where :math:`s = scale \\times \\sqrt{\\frac{3}{fan_{in}}}`. Here :math:`fan_{in}` is the number of input units. Reference: LeCun 98, Efficient Backprop http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf Attributes: scale (float): A constant that determines the scale of the uniform distribution. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(LeCunUniform, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) s = self.scale * numpy.sqrt(3. / fan_in) Uniform(s, rng=self.rng)(array) class GlorotUniform(initializer.Initializer): """Initializes array with a scaled uniform distribution. Each element of the array is initialized by the value drawn independently from uniform distribution :math:`[-s, s]` where :math:`s = scale \\times \\sqrt{\\frac{6}{fan_{in} + fan_{out}}}`. Here, :math:`fan_{in}` and :math:`fan_{out}` are the number of input and output units, respectively. Attributes: scale (float): A constant that determines the scale of the uniform distribution. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(GlorotUniform, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) s = self.scale * numpy.sqrt(6. / (fan_in + fan_out)) Uniform(s, rng=self.rng)(array) class HeUniform(initializer.Initializer): """Initializes array with scaled uniform distribution. Each element of the array is initialized by the value drawn independently from uniform distribution :math:`[-s, s]` where :math:`s = scale \\times \\sqrt{\\frac{6}{fan_{in}}}`. Here, :math:`fan_{in}` is the number of input units. Attributes: scale (float): A constant that determines the scale of the uniform distribution. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(HeUniform, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) s = self.scale * numpy.sqrt(6. / fan_in) Uniform(s, rng=self.rng)(array)
5,202
33.230263
76
py
chainer
chainer-master/chainer/initializers/normal.py
import numpy from chainer import backend from chainer.backends import cuda from chainer import initializer from chainer.utils import argument # Original code forked from MIT licensed keras project # https://github.com/fchollet/keras/blob/master/keras/initializations.py class Normal(initializer.Initializer): """Initializes array with a normal distribution. Each element of the array is initialized by the value drawn independently from Gaussian distribution whose mean is 0, and standard deviation is ``scale``. Args: scale (float): Standard deviation of Gaussian distribution. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=0.05, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(Normal, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) if self.rng is None: device = backend.get_device_from_array(array) args = {'loc': 0.0, 'scale': self.scale, 'size': array.shape} if device.xp is cuda.cupy: # Only CuPy supports dtype option if self.dtype == numpy.float32 or self.dtype == numpy.float16: # float16 is not supported in cuRAND args['dtype'] = numpy.float32 array[...] = device.xp.random.normal(**args) else: backend.copyto(array, self.rng.normal( loc=0.0, scale=self.scale, size=array.shape).astype(array.dtype, copy=False)) class LeCunNormal(initializer.Initializer): """Initializes array with scaled Gaussian distribution. Each element of the array is initialized by the value drawn independently from Gaussian distribution whose mean is 0, and standard deviation is :math:`scale \\times \\sqrt{\\frac{1}{fan_{in}}}`, where :math:`fan_{in}` is the number of input units. Reference: LeCun 98, Efficient Backprop http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf Args: scale (float): A constant that determines the scale of the standard deviation. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(LeCunNormal, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) s = self.scale * numpy.sqrt(1. / fan_in) Normal(s, rng=self.rng)(array) class GlorotNormal(initializer.Initializer): """Initializes array with scaled Gaussian distribution. Each element of the array is initialized by the value drawn independently from Gaussian distribution whose mean is 0, and standard deviation is :math:`scale \\times \\sqrt{\\frac{2}{fan_{in} + fan_{out}}}`, where :math:`fan_{in}` and :math:`fan_{out}` are the number of input and output units, respectively. Reference: Glorot & Bengio, AISTATS 2010 Args: scale (float): A constant that determines the scale of the standard deviation. dtype: Data type specifier. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, **kwargs): self.scale = scale rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(GlorotNormal, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) s = self.scale * numpy.sqrt(2. / (fan_in + fan_out)) Normal(s, rng=self.rng)(array) class HeNormal(initializer.Initializer): """Initializes array with scaled Gaussian distribution. Each element of the array is initialized by the value drawn independently from Gaussian distribution whose mean is 0, and standard deviation is :math:`scale \\times \\sqrt{\\frac{2}{fan}}`. If ``fan_option == 'fan_in'``, :math:`fan` is the number of input units. If ``fan_option == 'fan_out'``, :math:`fan` is the number of output units. Reference: He et al., https://arxiv.org/abs/1502.01852 Args: scale (float): A constant that determines the scale of the standard deviation. dtype: Data type specifier. fan_option ({'fan_in', 'fan_out'}): Decides how to compute the standard deviation. The default value is ``'fan_in'``. rng (xp.random.RandomState): Pseudo-random number generator. """ def __init__(self, scale=1.0, dtype=None, fan_option='fan_in', **kwargs): self.scale = scale self.fan_option = fan_option rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng super(HeNormal, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) fan_in, fan_out = initializer.get_fans(array.shape) if self.fan_option == 'fan_in': s = self.scale * numpy.sqrt(2. / fan_in) elif self.fan_option == 'fan_out': s = self.scale * numpy.sqrt(2. / fan_out) else: raise ValueError( 'fan_option should be either \'fan_in\' or \'fan_out\'.') Normal(s, rng=self.rng)(array)
6,197
34.016949
78
py
chainer
chainer-master/chainer/initializers/constant.py
import numpy import chainer from chainer import backend from chainer import initializer from chainer import types # NOQA class Identity(initializer.Initializer): """Initializes array with the identity matrix. It initializes the given array with the constant multiple of the identity matrix. Note that arrays to be passed must be 2D squared matrices. Attributes: scale (scalar): A constant to be multiplied to identity matrices. """ def __init__(self, scale=1.0, dtype=None): self.scale = scale super(Identity, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype shape = array.shape if len(shape) != 2 or shape[0] != shape[1]: raise ValueError('Identity matrix initialization can only be used ' 'for 2D squared matrices.') device = backend.get_device_from_array(array) array[...] = device.xp.identity(shape[0]) * self.scale class _Constant(initializer.Initializer): fill_value = None # type: types.ScalarValue def __init__(self, dtype=None): if not (isinstance(self.fill_value, chainer.get_array_types()) or numpy.isscalar(self.fill_value)): raise ValueError( 'fill_value must be either scalar, numpy.ndarray, ' 'cupy.ndarray or chainerx.ndarray.') super(_Constant, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype # Calling copy to ensures that the fill_value array # is moved to the device where array resides if isinstance(self.fill_value, chainer.get_array_types()): backend.copyto( array, self.fill_value.astype(array.dtype, copy=False)) else: device = backend.get_device_from_array(array) array[...] = device.xp.asarray(self.fill_value) class Constant(_Constant): """Initializes array with constant value. Attributes: ~Constant.fill_value (scalar or :ref:`ndarray`): A constant to be assigned to the initialized array. Broadcast is allowed on this assignment. dtype: Data type specifier. """ def __init__(self, fill_value, dtype=None): self.fill_value = fill_value super(Constant, self).__init__(dtype) class Zero(_Constant): """Initializes array to all-zero. Attributes: ~Zero.dtype: Data type specifier. """ fill_value = 0.0 class One(_Constant): """Initializes array to all-one. Attributes: ~One.dtype: Data type specifier. """ fill_value = 1.0 class NaN(_Constant): """Initializes array to all-NaN. Attributes: ~NaN.dtype: Data type specifier. """ fill_value = numpy.nan
2,918
25.779817
79
py
chainer
chainer-master/chainer/initializers/sampling.py
import numpy from chainer.backends import cuda from chainer import initializer # Original code from Berkeley FCN # https://github.com/shelhamer/fcn.berkeleyvision.org/blob/master/surgery.py def _get_linear_filter(size, ndim, upsampling=True): """Make a 2D and 3D linear kernel suitable for up/downsampling""" factor = (size + 1) // 2 if size % 2 == 1: center = factor - 1. else: center = factor - 0.5 slices = (slice(size),) * ndim og = numpy.ogrid[slices] filt = 1. for og_i in og: filt = filt * (1. - abs(og_i - center) / factor) if not upsampling: filt /= filt.sum() return filt class _SamplingFilter(initializer.Initializer): def __init__(self, upsampling=True, interpolation='linear', dtype=None): self._upsampling = upsampling if interpolation == 'linear': self._get_filter_func = _get_linear_filter else: raise ValueError( 'Unsupported interpolation method: {}'.format(interpolation)) super(_SamplingFilter, self).__init__(dtype) def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype xp = cuda.get_array_module(array) in_c, out_c = array.shape[:2] assert in_c == out_c or out_c == 1 ksize = None for k in array.shape[2:]: if ksize is None: ksize = k else: if ksize != k: raise ValueError( 'ksize must be all same: {} != {}'.format(ksize, k)) filt = self._get_filter_func( ksize, ndim=array.ndim - 2, upsampling=self._upsampling) filt = xp.asarray(filt) array[...] = 0. if out_c == 1: array[xp.arange(in_c), 0, ...] = filt else: array[xp.arange(in_c), xp.arange(out_c), ...] = filt class UpsamplingDeconvFilter(_SamplingFilter): """Initializes array with upsampling filter. The array is initialized with a standard image upsampling weight. This initializer is often used as initial weight for :func:`~chainer.links.DeconvolutionND`. :func:`~chainer.links.DeconvolutionND` is expected that its `stride` is equal to `(ksize + 1) // 2`. Reference: Long et al., https://arxiv.org/abs/1411.4038 Attributes: interpolation (str): Upsampling interpolation method. Default is 'linear'. """ def __init__(self, interpolation='linear', dtype=None): if interpolation != 'linear': raise ValueError( 'Unsupported interpolation method: {}'.format(interpolation)) super(UpsamplingDeconvFilter, self).__init__( upsampling=True, interpolation=interpolation, dtype=dtype) class DownsamplingConvFilter(_SamplingFilter): """Initializes array with downsampling filter. The array is initialized with a standard image downsampling weight. This initializer is often used as initial weight for :func:`~chainer.links.ConvolutionND`. :func:`~chainer.links.ConvolutionND` is expected that its `stride` is equal to `(ksize + 1) // 2`. Reference: Long et al., https://arxiv.org/abs/1411.4038 Attributes: interpolation (str): Downsampling interpolation method. Default is 'linear'. """ def __init__(self, interpolation='linear', dtype=None): if interpolation != 'linear': raise ValueError( 'Unsupported interpolation method: {}'.format(interpolation)) super(DownsamplingConvFilter, self).__init__( upsampling=False, interpolation=interpolation, dtype=dtype)
3,704
30.939655
77
py
chainer
chainer-master/chainer/initializers/orthogonal.py
import numpy from chainer import backend from chainer import initializer from chainer import utils from chainer.utils import argument _orthogonal_constraints = { # (assert emb., assert proj.) 'auto': (False, False), 'projection': (False, True), 'embedding': (True, False), 'basis': (True, True), } # Original code forked from MIT licensed keras project # https://github.com/fchollet/keras/blob/master/keras/initializations.py class Orthogonal(initializer.Initializer): """Initializes array with an orthogonal system. This initializer first makes a matrix of the same shape as the array to be initialized whose elements are drawn independently from standard Gaussian distribution. Next, it applies QR decomposition to (the transpose of) the matrix. To make the decomposition (almost surely) unique, we require the diagonal of the triangular matrix R to be non-negative (see e.g. Edelman & Rao, https://web.eecs.umich.edu/~rajnrao/Acta05rmt.pdf). Then, it initializes the array with the (semi-)orthogonal matrix Q. Finally, the array is multiplied by the constant ``scale``. If the ``ndim`` of the input array is more than 2, we consider the array to be a matrix by concatenating all axes except the first one. The number of vectors consisting of the orthogonal system (i.e. first element of the shape of the array) must be equal to or smaller than the dimension of each vector (i.e. second element of the shape of the array). Attributes: scale (float): A constant to be multiplied by. dtype: Data type specifier. mode (str): Assertion on the initialized shape. ``'auto'`` (default), ``'projection'`` (before v7), ``'embedding'``, or ``'basis'``. rng (xp.random.RandomState): Pseudo-random number generator. Reference: Saxe et al., https://arxiv.org/abs/1312.6120 """ def __init__(self, scale=1.1, dtype=None, mode='auto', **kwargs): self.scale = scale self.mode = mode rng = None if kwargs: rng, = argument.parse_kwargs(kwargs, ('rng', rng)) self.rng = rng try: self._checks = _orthogonal_constraints[mode] except KeyError: raise ValueError( 'Invalid mode: {}. Choose from {}.'.format( repr(mode), ', '.join(repr(m) for m in _orthogonal_constraints))) super(Orthogonal, self).__init__(dtype) # TODO(Kenta Oono) # How do we treat overcomplete base-system case? def __call__(self, array): if self.dtype is not None: assert array.dtype == self.dtype,\ '{} != {}'.format(array.dtype, self.dtype) if not array.shape: # 0-dim case if self.rng is None: a = numpy.random.randint(2) else: a = self.rng.randint(2) a = int(a) array[...] = self.scale * (2 * a - 1) elif not array.size: raise ValueError('Array to be initialized must be non-empty.') else: # numpy.prod returns float value when the argument is empty. out_dim = len(array) in_dim = utils.size_of_shape(array.shape[1:]) if (in_dim > out_dim and self._checks[0]) or ( in_dim < out_dim and self._checks[1]): raise ValueError( 'Cannot make orthogonal {}. ' 'shape = {}, interpreted as ' '{}-dim input and {}-dim output.'.format( self.mode, array.shape, in_dim, out_dim)) transpose = in_dim > out_dim if self.rng is None: a = numpy.random.normal(size=(out_dim, in_dim)) else: a_tmp = self.rng.normal(size=(out_dim, in_dim)) a = numpy.empty(a_tmp.shape, dtype=a_tmp.dtype) backend.copyto(a, a_tmp) if transpose: a = a.T # cupy.linalg.qr requires cusolver in CUDA 8+ q, r = numpy.linalg.qr(a) q *= numpy.copysign(self.scale, numpy.diag(r)) if transpose: q = q.T backend.copyto(array, q.reshape(array.shape).astype( array.dtype, copy=False))
4,361
38.297297
78
py
chainer
chainer-master/chainer/initializers/__init__.py
import typing as tp # NOQA import numpy import chainer from chainer import backend from chainer.backends import _chainerx # NOQA # import class and function from chainer.initializers.constant import Constant from chainer.initializers.constant import Identity # NOQA from chainer.initializers.constant import NaN # NOQA from chainer.initializers.constant import One # NOQA from chainer.initializers.constant import Zero # NOQA from chainer.initializers.normal import GlorotNormal # NOQA from chainer.initializers.normal import HeNormal # NOQA from chainer.initializers.normal import LeCunNormal from chainer.initializers.normal import Normal # NOQA from chainer.initializers.orthogonal import Orthogonal # NOQA from chainer.initializers.sampling import DownsamplingConvFilter # NOQA from chainer.initializers.sampling import UpsamplingDeconvFilter # NOQA from chainer.initializers.uniform import GlorotUniform # NOQA from chainer.initializers.uniform import HeUniform # NOQA from chainer.initializers.uniform import LeCunUniform # NOQA from chainer.initializers.uniform import Uniform # NOQA from chainer import types # NOQA def generate_array( initializer: types.AbstractInitializer, shape: types.ShapeSpec, xp: types.Xp, dtype: tp.Optional[types.DTypeSpec] = None, device: tp.Optional[types.DeviceSpec] = None ) -> types.NdArray: """Return initialized array. The algorithms used to make the new values depend on the concrete derived classes. If the initializer has the ``dtype`` attribute, it is used to construct the array. Otherwise, ``chainer.config.dtype`` is used instead. See :ref:`configuration` for the dtype config. Args: initializer: A callable object that takes :ref:`ndarray` and edits its value. shape (int or tuple of int): Shape of the initialized array. xp (module): :mod:`cupy`, :mod:`numpy`, or :mod:`chainerx`. dtype: Dtype specifier. If omitted, ``initializer.dtype`` is used. device: Target device specifier. If omitted, the current device is used for :mod:`cupy`, and the default device is used for :mod:`chainerx`. Returns: :ref:`ndarray`: An initialized array. """ dtype_attr = getattr(initializer, 'dtype', None) if dtype is not None and dtype_attr is not None \ and numpy.dtype(dtype) != numpy.dtype(dtype_attr): raise ValueError( 'dtype mismatch: {} != {}'.format(dtype, dtype_attr)) if dtype is None: dtype = dtype_attr dtype = chainer.get_dtype(dtype) if device is None: backend_device = backend._guess_device_from_array_module(xp) else: backend_device = chainer.get_device(device) if xp != backend_device.xp: raise ValueError('xp and device arguments are inconsistent.') with chainer.using_device(backend_device): array = xp.empty(shape, dtype=dtype) initializer(array) return array def _get_initializer( initializer: tp.Optional[types.InitializerSpec] ) -> types.AbstractInitializer: if initializer is None: return LeCunNormal() if (isinstance(initializer, chainer.get_array_types()) or numpy.isscalar(initializer)): return Constant(initializer) if not callable(initializer): raise TypeError('invalid type of initializer: %s' % type(initializer)) return initializer def _check_is_initializer_like(initializer): if not (initializer is None or isinstance(initializer, chainer.Initializer) or callable(initializer) or isinstance(initializer, chainer.get_array_types()) or numpy.isscalar(initializer)): raise TypeError( 'Initializer is of wrong type: {}. Allowed types are Initializer, ' 'ndarray and scalar.'.format(type(initializer)))
3,923
37.851485
79
py
chainer
chainer-master/chainer/link_hooks/spectral_normalization.py
import chainer from chainer import backend from chainer import configuration import chainer.functions as F from chainer import link_hook import chainer.links as L from chainer import variable import chainerx def l2normalize(xp, v, eps): """Normalize a vector by its L2 norm. Args: xp (numpy or cupy): v (numpy.ndarray or cupy.ndarray) eps (float): Epsilon value for numerical stability. Returns: :class:`numpy.ndarray` or :class:`cupy.ndarray` """ # TODO(crcrpar): Remove this when chainerx.linalg.norm becomes available. if xp is chainerx: # NOTE(crcrpar): `chainerx.power` is not available as of 2019/03/27. # See https://github.com/chainer/chainer/pull/6522 norm = chainerx.sqrt(chainerx.sum(v * v)) else: norm = xp.linalg.norm(v) return v / (norm + eps) def update_approximate_vectors( weight_matrix, u, n_power_iteration, eps): """Update the first left and right singular vectors. This function updates the first left singular vector `u` and the first right singular vector `v`. Args: weight_matrix (~chainer.Variable): 2D weight. u (numpy.ndarray, cupy.ndarray, or None): Vector that approximates the first left singular vector and has the shape of (out_size,). n_power_iteration (int): Number of iterations to approximate the first right and left singular vectors. Returns: :class:`numpy.ndarray` or `cupy.ndarray`: Approximate first left singular vector. :class:`numpy.ndarray` or `cupy.ndarray`: Approximate first right singular vector. """ weight_matrix = weight_matrix.array xp = backend.get_array_module(weight_matrix) for _ in range(n_power_iteration): v = l2normalize(xp, xp.dot(u, weight_matrix), eps) u = l2normalize(xp, xp.dot(weight_matrix, v), eps) return u, v def calculate_max_singular_value(weight_matrix, u, v): """Calculate max singular value by power iteration method. Args: weight_matrix (~chainer.Variable) u (numpy.ndarray or cupy.ndarray) v (numpy.ndarray or cupy.ndarray) Returns: ~chainer.Variable: Max singular value via power iteration method. """ sigma = F.matmul(F.matmul(u, weight_matrix), v) return sigma class SpectralNormalization(link_hook.LinkHook): """Spectral Normalization link hook implementation. This hook normalizes a weight using max singular value and this value is computed via power iteration method. Currently, this hook is supposed to be added to :class:`chainer.links.Linear`, :class:`chainer.links.EmbedID`, :class:`chainer.links.Convolution2D`, :class:`chainer.links.ConvolutionND`, :class:`chainer.links.Deconvolution2D`, and :class:`chainer.links.DeconvolutionND`. However, you can use this to other links like RNNs by specifying ``weight_name``. It is highly recommended to add this hook before optimizer setup because this hook add a scaling parameter ``gamma`` if ``use_gamma`` is True. Otherwise, the registered ``gamma`` will not be updated. .. math:: \\bar{\\mathbf{W}} &=& \\dfrac{\\mathbf{W}}{\\sigma(\\mathbf{W})} \\\\ \\text{, where} \\ \\sigma(\\mathbf{W}) &:=& \\max_{\\mathbf{h}: \\mathbf{h} \\ne 0} \\dfrac{\\|\\mathbf{W} \\mathbf{h}\\|_2}{\\|\\mathbf{h}\\|_2} = \\max_{\\|\\mathbf{h}\\|_2 \\le 1} \\|\\mathbf{W}\\mathbf{h}\\|_2 See: T. Miyato et. al., `Spectral Normalization for Generative Adversarial Networks <https://arxiv.org/abs/1802.05957>`_ Args: n_power_iteration (int): Number of power iteration. The default value is 1. eps (float): Numerical stability in norm calculation. The default value is 1e-6 for the compatibility with mixed precision training. The value used in the author's implementation is 1e-12. use_gamma (bool): If ``True``, weight scaling parameter gamma which is initialized by initial weight's max singular value is introduced. factor (float, None): Scaling parameter to divide maximum singular value. The default value is 1.0. weight_name (str): Link's weight name to apply this hook. The default value is ``'W'``. name (str or None): Name of this hook. The default value is ``'SpectralNormalization'``. Attributes: vector_name (str): Name of the approximate first left singular vector registered in the target link. the target link. axis (int): Axis of weight represents the number of output feature maps or output units (``out_channels`` and ``out_size``, respectively). .. admonition:: Example There are almost the same but 2 ways to apply spectral normalization (SN) hook to links. 1. Initialize link and SN separately. This makes it easy to handle buffer and parameter of links registered by SN hook. >>> l = L.Convolution2D(3, 5, 3) >>> hook = chainer.link_hooks.SpectralNormalization() >>> _ = l.add_hook(hook) >>> # Check the shape of the first left singular vector. >>> getattr(l, hook.vector_name).shape (5,) >>> # Delete SN hook from this link. >>> l.delete_hook(hook.name) 2. Initialize both link and SN hook at one time. This makes it easy to define your original :class:`~chainer.Chain`. >>> # SN hook handles lazy initialization! >>> layer = L.Convolution2D( ... 5, 3, stride=1, pad=1).add_hook( ... chainer.link_hooks.SpectralNormalization()) """ name = 'SpectralNormalization' def __init__(self, n_power_iteration=1, eps=1e-6, use_gamma=False, factor=None, weight_name='W', name=None): assert n_power_iteration > 0 self.n_power_iteration = n_power_iteration self.eps = eps self.use_gamma = use_gamma self.factor = factor self.weight_name = weight_name self.vector_name = weight_name + '_u' self._initialized = False self.axis = 0 if name is not None: self.name = name def __enter__(self): raise NotImplementedError( 'This hook is not supposed to be used as context manager.') def __exit__(self): raise NotImplementedError def added(self, link): # Define axis and register ``u`` if the weight is initialized. if not hasattr(link, self.weight_name): raise ValueError( 'Weight \'{}\' does not exist!'.format(self.weight_name)) if isinstance(link, (L.Deconvolution2D, L.DeconvolutionND)): self.axis = 1 if getattr(link, self.weight_name).array is not None: self._prepare_parameters(link) def deleted(self, link): # Remove approximate vector ``u`` and parameter ``gamma` if exists. delattr(link, self.vector_name) if self.use_gamma: del link.gamma def forward_preprocess(self, cb_args): # This method normalizes target link's weight spectrally # using power iteration method link = cb_args.link input_variable = cb_args.args[0] if not self._initialized: self._prepare_parameters(link, input_variable) weight = getattr(link, self.weight_name) # For link.W or equivalents to be chainer.Parameter # consistently to users, this hook maintains a reference to # the unnormalized weight. self.original_weight = weight # note: `normalized_weight` is ~chainer.Variable normalized_weight = self.normalize_weight(link) setattr(link, self.weight_name, normalized_weight) def forward_postprocess(self, cb_args): # Here, the computational graph is already created, # we can reset link.W or equivalents to be Parameter. link = cb_args.link setattr(link, self.weight_name, self.original_weight) def _prepare_parameters(self, link, input_variable=None): """Prepare one buffer and one parameter. Args: link (:class:`~chainer.Link`): Link to normalize spectrally. input_variable (:class:`~chainer.Variable`): The first minibatch to initialize weight. """ if getattr(link, self.weight_name).array is None: if input_variable is not None: link._initialize_params(input_variable.shape[1]) initialW = getattr(link, self.weight_name) if initialW.shape[self.axis] == 0: raise ValueError( 'Expect {}.shape[{}] > 0'.format(self.weight_name, self.axis) ) u = link.xp.random.normal( size=(initialW.shape[self.axis],)).astype(dtype=initialW.dtype) setattr(link, self.vector_name, u) link.register_persistent(self.vector_name) if self.use_gamma: # Initialize the scaling parameter with the max singular value. weight_matrix = self.reshape_W(initialW.array) # TODO(crcrpar): Remove this when chainerx supports SVD. device = link.device if device.xp is chainerx: fallback_device = device.fallback_device weight_matrix_ = fallback_device.send(weight_matrix) with chainer.using_device(fallback_device): _, s_, _ = fallback_device.xp.linalg.svd(weight_matrix_) s = device.send(s_) else: _, s, _ = link.xp.linalg.svd(weight_matrix) s0 = chainer.utils.force_array(s[0]) with link.init_scope(): link.gamma = variable.Parameter(s0) self._initialized = True def normalize_weight(self, link): """Normalize target weight before every single forward computation.""" weight_name, vector_name = self.weight_name, self.vector_name W = getattr(link, weight_name) u = getattr(link, vector_name) weight_matrix = self.reshape_W(W) if not configuration.config.in_recomputing: with chainer.using_device(link.device): u, v = update_approximate_vectors( weight_matrix, u, self.n_power_iteration, self.eps) else: v = self.v sigma = calculate_max_singular_value(weight_matrix, u, v) if self.factor is not None: sigma /= self.factor if self.use_gamma: W = link.gamma * W / sigma else: W = W / sigma if not configuration.config.in_recomputing: self.v = v with chainer.using_device(link.device): if configuration.config.train: if link.xp is chainerx: # TODO(crcrpar): Remove this when # chainerx supports `copyto`. getattr(link, vector_name)[:] = u else: backend.copyto(getattr(link, vector_name), u) return W def reshape_W(self, W): """Reshape & transpose weight into 2D if necessary.""" if self.axis != 0: axes = [self.axis] + [i for i in range(W.ndim) if i != self.axis] W = W.transpose(axes) if W.ndim == 2: return W return W.reshape(W.shape[0], -1)
11,603
38.604096
79
py
chainer
chainer-master/chainer/link_hooks/timer.py
import collections import os import sys import time import numpy from chainer.backends import cuda from chainer import link_hook # Select the best-resolution timer function try: _get_time = time.perf_counter except AttributeError: if os.name == 'nt': _get_time = time.clock else: _get_time = time.time class TimerHook(link_hook.LinkHook): """Link hook for measuring elapsed time of \ :meth:`Link.forward() <chainer.Link.forward>`. Example: Code example:: from chainer.link_hooks import TimerHook hook = TimerHook() with hook: trainer.run() hook.print_report() Output example:: LinkName ElapsedTime Occurrence Linear 41.42sec 2100 MLP 42.09sec 700 Classifier 42.39sec 700 where *LinkName* is the name of link that calls the hook, and *ElapsedTime* is the elapsed time the link consumed, and *Occurrence* is the number of calls. Warning: Call graph of links are hierarchical. That means reported elapsed times may be overlapping with each other and the sum may exceed the total time. Attributes: call_history: List of measurement results. It consists of pairs of the name of the link that calls this hook and the elapsed time the :meth:`forward` method of link consumes. """ name = 'TimerHook' table = {'sec': 1, 'ms': 10 ** 3, 'us': 10 ** 6, 'ns': 10 ** 9} def __init__(self): self.call_history = [] self._running_stack = [] self._depth = 0 self._total_time = 0 def _preprocess(self): if self.xp is numpy: start = _get_time() self._running_stack.append(start) else: assert self.xp is cuda.cupy start = cuda.Event() stop = cuda.Event() start.record() self._running_stack.append((start, stop)) self._depth += 1 def forward_preprocess(self, args): self.xp = args.link.xp self._preprocess() def _postprocess(self, link): if self.xp is numpy: start = self._running_stack.pop() stop = _get_time() elapsed_time = stop - start else: assert self.xp is cuda.cupy start, stop = self._running_stack.pop() stop.record() stop.synchronize() # Note that `get_elapsed_time` returns result in milliseconds elapsed_time = cuda.cupy.cuda.get_elapsed_time( start, stop) / 1000 self.call_history.append((link.__class__.__name__, elapsed_time)) assert self._depth > 0 self._depth -= 1 if self._depth == 0: self._total_time += elapsed_time def forward_postprocess(self, args): link = args.link assert link.xp == self.xp self._postprocess(link) def total_time(self): """Returns total elapsed time in seconds.""" return self._total_time def summary(self): """Returns a summary of time profiling in links. Returns: A summarized dictionary whose keys are link names and values are dictionaries of `elapsed_time` and `occurrence`. """ summary = collections.OrderedDict() for link_name, elapsed_time in self.call_history: if link_name not in summary: summary[link_name] = {'elapsed_time': 0, 'occurrence': 0} record = summary[link_name] record['elapsed_time'] += elapsed_time record['occurrence'] += 1 return summary def _choose_unit(self, second): """Choose optimal unit.""" factor = 1 for unit in ['sec', 'ms', 'us']: if second * factor >= 1: return factor, unit factor *= 1000.0 return factor, 'ns' def print_report(self, unit='auto', file=sys.stdout): """Prints a summary report of time profiling in links. Args: unit (str): Supplementary units used for computational times. `sec`, `ms`, `us`, `ns`, `auto`(default) and `auto_foreach` are supported. If `auto`, units of times are aligned to the largest, and if `auto_foreach`, units of times are adjusted for each element. """ entries = [['LinkName', 'ElapsedTime', 'Occurrence']] auto_foreach = (unit == 'auto_foreach') if unit == 'auto': max_time = max( record['elapsed_time'] for record in self.summary().values()) factor, unit = self._choose_unit(max_time) elif not auto_foreach: factor = self.table[unit] for link_name, record in self.summary().items(): second = record['elapsed_time'] if auto_foreach: factor, unit = self._choose_unit(second) elapsed_time = '%3.2f%s' % (second * factor, unit) occurrence = str(record['occurrence']) entries.append([link_name, elapsed_time, occurrence]) entry_widths = [] entry_widths.append(max(len(f) for f, _, _ in entries)) entry_widths.append(max(len(e) for _, e, _ in entries)) entry_widths.append(max(len(o) for _, _, o in entries)) template = ' '.join('{:>%d}' % w for w in entry_widths) for link_name, elapsed_time, occurrence in entries: line = template.format(link_name, elapsed_time, occurrence) file.write(line) file.write('\n') file.flush() # TODO(crcrpar): Support backward pre/post process. # See https://github.com/chainer/chainer/issues/5197
5,855
32.849711
79
py
chainer
chainer-master/chainer/link_hooks/__init__.py
from chainer.link_hooks.spectral_normalization import SpectralNormalization # NOQA from chainer.link_hooks.timer import TimerHook # NOQA from chainer.link_hooks.weight_standardization import WeightStandardization # NOQA
223
55
83
py
chainer
chainer-master/chainer/link_hooks/weight_standardization.py
import chainer from chainer.functions.normalization import group_normalization from chainer import link_hook class WeightStandardization(link_hook.LinkHook): """Weight Standardization (WS) link hook implementation. This hook standardizes a weight by *weight statistics*. This link hook implements a WS which computes the mean and variance along axis "output channels", then normalizes by these statistics. WS improves training by reducing the Lipschitz constants of the loss and the gradients like batch normalization (BN) but without relying on large batch sizes during training. Specifically, the performance of WS with group normalization (GN) trained with small-batch is able to match or outperforms that of BN trained with large-batch. WS is originally proposed for 2D convolution layers followed by mainly GN and sometimes BN. Note that this hook is able to handle layers such as N-dimensional convolutional, linear and embedding layers but there is no guarantee that this hook helps training. See: Siyuan Qiao et. al., `Weight Standardization <https://arxiv.org/abs/1903.10520>`_ Args: eps (float): Numerical stability in standard deviation calculation. The default value is 1e-5. weight_name (str): Link's weight name to appky this hook. The default value is ``'W'``. name (str or None): Name of this hook. The default value is ``'WeightStandardization'``. """ name = 'WeightStandardization' def __init__(self, *, eps=1e-5, weight_name='W', name=None): self.eps = eps self.weight_name = weight_name self._initialized = False if name is not None: self.name = name def __enter__(self): raise NotImplementedError( 'This hook is not supposed to be used as context manager.') def __exit__(self): raise NotImplementedError def added(self, link): if not hasattr(link, self.weight_name): raise ValueError( 'Weight \'{}\' does not exist!'.format(self.weight_name)) if getattr(link, self.weight_name).array is not None: self._initialized = True def forward_preprocess(self, cb_args): # This method normalizes target link's weight by statistics link = cb_args.link input_variable = cb_args.args[0] if not self._initialized: if getattr(link, self.weight_name).array is None: if input_variable is None: raise ValueError('Input variable does not exist!') link._initialize_params(input_variable.shape[1]) weight = getattr(link, self.weight_name) with chainer.using_device(link.device): gamma = link.xp.ones( (weight.shape[1],), dtype=weight.dtype) beta = link.xp.zeros( (weight.shape[1],), dtype=weight.dtype) # For link.W or equivalents to be chainer.Parameter # consistently to users, this hook maintains a reference to # the unnormalized weight. self.original_weight = weight # note: `normalized_weight` is ~chainer.Variable normalized_weight = group_normalization.group_normalization( weight, groups=1, gamma=gamma, beta=beta, eps=self.eps) setattr(link, self.weight_name, normalized_weight) def forward_postprocess(self, cb_args): # Here, the computational graph is already created, # we can reset link.W or equivalents to be Parameter. link = cb_args.link setattr(link, self.weight_name, self.original_weight)
3,678
40.806818
79
py
chainer
chainer-master/chainer/graph_optimizations/static_graph_utilities.py
import inspect import chainer def static_code(*dec_args, **dec_kwargs): """Decorator to mark a function for inclusion in the static schedule. This decorator is used to mark a function or method to be included in a static schedule. There are multiple types of static schedules, such as "forward pass schedule", "backward pass schedule", "double backward pass schedule" etc.. The type of schedule that the decorated function's code is added to will depend on the context in which this decorator is used. For example, the decorated code will be added to the "forward pass schedule" if it is called while executing the define-by- run code of a static subgraph. To inform the framework that a particular portion of define-by-run code corresponds to a static subgraph, the code should be placed inside the `__call__()` method of a chain and then apply the `@static_graph` decorator to the `__call__()` method. We will refer to such a chain as a "static chain." This will cause any functions decorated with `static_code` that are called while inside of `__call__()` to be included in the forward pass static schedule in the same order in which they were executed in the define-by-run code. Likewise, for any `FunctionNode` instances that are called inside a static chain, any code that is run while inside the `backward()` method that calls a function using this decorator will be added to the corresponding "backward pass schedule." Usage: This decorator should be applied to any code called from a static chain that needs to run each iteration. This should only include the code that performs the actual forward and/or backward computations and not include code for initializing parameters, checking types, etc.. As long as a chain is marked as static, the framework will automatically wrap any `FunctionNode` instances so that the code inside their `forward()` and `backward()` methods is added to the corresponding forward and backward static schedules, respectively. As a result, any built-in Chainer function and link calls will be automatically included in the static schedule. However, there are two cases where the user will need to use this decorator: 1. Code with side effects that is called from a static chain's define-by- run code must be placed in a function decorated with `@static_code`. 2. Any user-defined links that contain code other chain Chainer function calls that must run every iteration must place such code in a function decorated with `@static_graph`. This decorator can be applied to either a function or a method (usually of a `FunctionNode`). There are no required arguments, and so a user can apply it to "side effect" code to cause an operation to be executed each iteration. The more usual use case is where the core framework code will apply it to the all of (and only) the functions that actually perform the computations needed to compute the forward and backward passes. The simplest usage is when we would like to force a particular user-defined function to run each iteration. For example, such a function might increment a counter, check conditions, and possibly print information to the console. In this use, it is only required to add this decorator to the function definition and then call it during the first iteration from the context of the static chain's `__call__()` method. Passing and returing arrays: If the function needs an array as an input argument that was used elsewhere in the static schedule, it must appear as an item in list of arrays that is supplied in the `inputs` keyword argument. An example would be the typical case where one layer in the network produces output activations `y` which are then used as the input of the next layer. If the corresponding `FunctionNode` instances wrap their computaions using this decorator, this will result in multiple functions that operate on `y`. The following constraints specify how such arrays should be passed into and returned from a function that uses this decorator. If the function will return results in one or more arrays, there are two options: 1. Write the results in-place into preallocated arrays that are supplied in a list in the `outputs` keyword argument. 2. Dynamically allocate the result array(s) inside the function and return them inside a tuple. When two schedule functions "func_A" and "func_B" operate on the same array `x`, `x` must explicitly appear as an input argument and/or output of both functions. For example, it would be an error to have schedule function "func_A" return a dynamically allocated array `x` and then have schedule function "func_B" later read from `x` without it appearing in "func_B"'s `inputs` list. Note that this would work during the first iteration, but during the next iteration when "func_A" is called, it would allocate and return a new array for `x` leading to "func_B" reading from the stale reference to `x` from the previous iteration. This usage is allowed in some special cases by the framework code, but is not allowed for user-defined functions. Performance notes: The function should return any output arrays in-place into pre-allocated arrays (1. above) when possible since this this allows the scheduler to make tradeoffs between computation efficiency and memory usage. For example, this allows the use of a completely static array allocations (no allocations after the first iteration), if desired. If memory reduction is needed, the scheduler may choose to delete the arrays in `inputs` once they are no longer needed in an iteration and then reallocate them again in the next iteration just before the function is called. Note that completely static array allocations are not possible if any of the schedule functions return a tuple of dynamically allocated arrays, as the existing chainer functions do. The following optional arguments apply to the wrapped function or method. Args (of this decorater): func_name (str): An optional descriptive name that will be associated with this function in the static schedule. It is intended for debugging purposes. Args (of the wrapped fuction): inputs (list of ndarray): An optional keyword argument that supplies all arrays needed as input by the function. If the function needs an array that is used by another function in the static schedule, it must appear in this list. outputs (list of ndarray): An optional keyword argument that supplies all output arrays of this function. These arrays must already have been initialized to the correct shapes and dtypes before the function is called. The function must write its results in-place into these arrays. Any output arrays that may be used inside another schedule function must appear in this list. Returns: None or a tuple of ndarray: If the function dynamically allocates its output arrays, they must be returned in a tuple of arrays. """ func_name = None zero_args = False if len(dec_args) == 1 and not dec_kwargs and callable(dec_args[0]): callable_arg = dec_args[0] zero_args = True elif dec_kwargs: if 'func_name' in dec_kwargs: func_name = dec_kwargs['func_name'] def wrap(func): def wrapped_func(*args, **kwargs): # Save arguments, function, and results pointers/references # to the schedule list: # If trace mode is on, add to schedule. schedule_function = chainer.config.schedule_func if schedule_function is not None: assert chainer.config.use_static_graph # Note: 'ret = func(*args, **kwargs)' is called inside # the following method. ret = schedule_function.append_function(func, args, kwargs, func_name=func_name) # Add the schedule function as an attribute of the # FunctionNode instance (or more generally, to any class) # that contains the wrapped function as a method if args: instance = args[0] if inspect.isclass(instance): # note: this is not currently needed. instance.schedule_func = schedule_function else: ret = func(*args, **kwargs) return ret return wrapped_func if zero_args: return wrap(callable_arg) else: return wrap def static_forward_optimizations(func, inputs): """Perform checks needed for creation of a static schedule. Check if `func` supports static graph optimizations. If not, automatically wrap it to be compatible. This function is called from the `FunctionNode` apply() method in place of the original `func.forward(inputs)` call if `chainer.config.schedule_func` is not None. Args: func (instance of FunctionNode): inputs (tuple of ndarray): input arrays to `func` Returns: (tuple of ndarray): The outputs of the function. """ schedule_function = chainer.config.schedule_func if not func._supports_static_optimizations: if schedule_function.verbosity_level >= 2: print('Adding automatic static graph support to ' 'function: ', func) @static_code(func_name=str(func)) def generic_static_forward(func, inputs): """Auto-wrap the supplied function. func (instance of FunctionNode): The function to include in the static schedule. inputs (list of input arrays): The input arguments to `func`. Returns: a tuple of output arrays. """ # Convert back to tuple because func.forward() requires it. in_data = tuple(inputs) ret = func.forward(in_data) return ret # Note: we convert inputs to a list because the API for # static_code requires it. return generic_static_forward(func, inputs=list(inputs)) return func.forward(inputs)
10,620
43.439331
77
py
chainer
chainer-master/chainer/graph_optimizations/__init__.py
from chainer.graph_optimizations.static_graph_utilities import static_code # NOQA
84
27.333333
82
py
chainer
chainer-master/chainer/graph_optimizations/static_graph.py
import sys import weakref import numpy as np import chainer from chainer.backends import cuda import chainer.function_node def _is_xp(x): return isinstance(x, np.ndarray) or isinstance(x, cuda.ndarray) class ScheduleInfo(object): """A callable wrapper for a function in the static schedule. Args: func (FunctionNode): A function in the static schedule. args: Arguments to 'func'. kwargs: Keyword arguments to 'func'. inputs_hooks (list of tuples): A list of hooks that instruct how to update the ndarray references in 'args' so that they refer to the correct master array in 'unique_arrays'. return_hooks (list of tuples): A list of hooks that instruct how to update the ndarray references in 'unique_arrays' so that they refer to the correct arrays that were dynamically allocated and returned by 'func'. These run after 'func' is called. unique_arrays (list of ndarray): The master list of all unique ndarrays that appear in the static schedule. func_name (str): An optional name of the static function. This is the name (if any) that was used as a decorater argument to `@static_code(func_name=name)`. """ def __init__(self, func, args, kwargs, inputs_hooks, outputs_hooks, return_hooks, delete_hooks, unique_arrays, array_infos, func_name=None): self.func = func self.args = args self.kwargs = kwargs self.inputs_hooks = inputs_hooks self.outputs_hooks = outputs_hooks self.return_hooks = return_hooks self.unique_arrays = unique_arrays self.array_infos = array_infos assert len(self.array_infos) == len(self.unique_arrays) self.func_name = func_name self.in_list = None if self.inputs_hooks: self.in_list = self.kwargs['inputs'] if self.outputs_hooks: self.out_list = self.kwargs['outputs'] # Check if 'func' wraps code of a 'FunctionNode': self.function_node = None if self.args: maybe_func = self.args[0] if isinstance(maybe_func, chainer.FunctionNode): self.function_node = maybe_func # List of indices in unique_arrays to delete. self.delete_hooks = delete_hooks def run_pre_hooks(self): """Run hooks to set correct references. This method is called from '__call__()'. Process the list of hooks which will modify the array references in the arguments list of the static function. This method must be called before executing the static function. The hooks specify that each array argument points to a "master" array reference in the unique_arrays list. If the reference in unique_arrays changes, then we must also change the corresponding array reference in the arguments list. The hooks specify the mapping and this method updates the references in args to the corresponding values from unique_arrays. """ for hook in self.inputs_hooks: (ind, unique_ind) = hook self.in_list[ind] = self.unique_arrays[unique_ind] for hook in self.outputs_hooks: (ind, unique_ind) = hook self.out_list[ind] = self.unique_arrays[unique_ind] for ind in self.delete_hooks: self.unique_arrays[ind] = None def run_post_hooks(self, return_arrays): """Run post-hooks. This method should be called after calling the static function `self.func(*self.args)`. This method sets any array references that appear in `self.args` to None. This is safe because the master array reference is still kept in `self.unique_arrays`. Also, process the list of post-hooks which will modify the array references in the unique_arrays list to refer to the new dynamically-allocated arrays that were returned by 'func'. Args: return_arrays (list of ndarray or None): The list of arrays that were returned by the schedule function, if not None. """ for hook in self.inputs_hooks: (ind, unique_ind) = hook self.in_list[ind] = None for hook in self.outputs_hooks: (ind, unique_ind) = hook self.out_list[ind] = None for hook in self.return_hooks: # Update the array reference in unique_arrays to refer to the # array in the results array. (ret_index, unique_list_index) = hook # Note: input/output variables to a FunctionNode that are # retained using retain_inputs() or retain_outputs() are # not currently explicitly used as input arguments to the # auto-wrapped functions, and so their corresponding array # reference could be used inside a function wrapped with # @static_code without the array explicitly appearing in the # 'inputs' argument. It is therefore not safe to change the # reference of such arrays, and so for them, we must be # sure to copy the dynamically-allocated array into the # same array that was used in the define-by-run code and # set 'need_copy' to True in such cases. need_copy = self.array_infos[unique_list_index].retain # todo: possible memory leak when need_copy False is allowed? if need_copy: # This must be used if the model used retain_inputs() or # retain_outputs(). self.unique_arrays[unique_list_index][...] = \ return_arrays[ret_index] else: # This is preferred, when possible, since it should # be faster than a copy to simply update the array # reference. self.unique_arrays[unique_list_index] = \ return_arrays[ret_index] def __call__(self): self.run_pre_hooks() ret = self.func(*self.args, **self.kwargs) self.run_post_hooks(ret) def __repr__(self): out = 'function: ' + str(self.func) + '\n' out += 'name: ' + str(self.func_name) + '\n' out += 'args: ' + str(self.args) + '\n' out += 'kwargs: ' + str(self.args) + '\n' return out class ArrayInfo(object): """Array information needed by the scheduler. This contains information about one array used in the naive static schedule corresponding to the define-by-run code. """ def __init__(self, array): # Weak reference to the array in the define-by-run code. self.weak_ref = weakref.ref(array) self.id = id(array) # The array (normal reference). Do not create in initializer. self.array = None self.shape = array.shape self.dtype = array.dtype # either numpy or cupy self.ndarray_module = chainer.backend.get_array_module(array) if self.ndarray_module is cuda.cupy: # device id, if available. self.device = cuda.get_device_from_array(array) else: # numpy (cpu) self.device = -1 # todo: save array order ('C', 'F' as well? # It specifies the input variable corresponding # to this array as the tuple (pass_depth, in_var_index). self.in_var_index = None # It specifies the output variable corresponding # to this array as the tuple (pass_depth, out_var_index). self.out_var_index = None # todo: set in initializer as keyword arg? self.dynamically_allocated = False # If the array was returned as a dynamically allocated array # in the define-by-run code, this specifies the location # in the schedule as the tuple (pass_depth, sched_func_index) # where sched_func_index is the index of the corresponding # ScheduleInfo object in the StaticScheduleFunction's # self.schedule_info_list self.dynamic_allocation_index = None self.dynamic_allocation_pass_depth = None self.dynamic_deletion_index = None self.dynamic_deletion_pass_depth = None # This is the same as self.dynamic_allocation_index, but for the # case where the array was statically allocated in the # define-by-run code. self.static_allocation_index = None # If the array needs to be retained (was included in # retain_inputs/retain_outputs), # this will be set to True later. self.retain = False def was_deleted(self): return self.weak_ref() is None def get_new_empty_array(self): """Make and return a new empty ndarray. Make and return a new empty ndarray that has the same shape, dtype, and device as the array that was supplied to the initializer. """ # todo: set device id return self.ndarray_module.empty(self.shape, dtype=self.dtype) def __repr__(self): out = 'shape: {}\n'.format(self.shape) if self.was_deleted(): out += 'Weak reference: dead\n' else: out += 'Weak reference: alive\n' if self.retain: out += 'Retained with retain_inputs()/retain_outputs().\n' if self.dynamically_allocated: out += 'Dynamically allocated at\n' out += \ ' pass_depth: {}\n'.format(self.dynamic_allocation_pass_depth) out += ' sched_index: {}\n'.format(self.dynamic_allocation_index) out += 'array id: {}'.format(self.id) return out class StaticScheduleFunction(chainer.function_node.FunctionNode): """A function that executes the static schedule of a Chain. An instance of this class executes the static schedule of computations that are equivalent to executing the define-by-run code of a Chain. This class is used by the `static_graph` decorator to wrap the define-by-run computations of a chain into two static schedules: - The forward schedule corresponds to the computations that are executed by the define-by-run code of the `__call__()` method of a chain. The static schedule corresponding to these computations can be executed by calling the `forward()` method of this class. - The backward schedule corresponds to the computations that are executed by the sequence of calls to `Function.backward()` that occur during when backpropagating the gradients through the same chain. That is, for each `Function.forward()` that was called during the forward propagation, there will be a corresponding call to `Function.backward()` (of the same Function object) in the backward schedule. This backward schedule can be executed by calling the `backward()` method of this class. Note the intended usage of this class: Recall that a "static chain" referes to a chain that is decorated by the `static_graph` decorator. During the first forward pass of a static chain, the define-by-run code is executed. However, for subsequent iterations, that define-by-run code is replaced by an instance of this function and this function will be called instead. Since the static schedules contained by this function perform the same computations, it is safe (and potentially much more efficient) to simply execute the static schedule instead of the define-by-run code. See `static_graph` for details. Args: schedule_manager (ScheduleManager): The schedule manager of this schedule instance. in_vars (tuple of Variable): The flattened tuple of input variables that is supplied to `__call__()` method of the chain that this schedule corresponds to. unique_arrays (list of ndarray): A list of all unique array references deeply used in an StaticScheduleFunction instance. It is 'None' for the StaticScheduleFunction that corresponds to the "forward" schedule, but the contained StaticScheduleFunction for the "backward" schedule should take the unique_arrays of the "forward" schedule. """ def __init__(self, schedule_manager, verbosity_level=0, enable_double_backprop=False): # A pass depth of 0 corresponds to the schedule for the forward pass. # A pass depth of 1 corresponds to the schedule for the backward pass. # A pass depth of 2 corresponds to the schedule for the # double-backward pass, and so on. self.pass_depth = 0 self.schedule_manager = schedule_manager # A list of ScheduleInfo objects, each of which contains one function # in the static schedule. The order of functions in this list is # the order they should be called in the schedule. self.schedule_info_list = [] # A list of all unique ndarrays used in this schedule and any deeply # contained schedules (backward, double-backward schedules). # That is, it is shared among all pass depths. # Note that this typically includes the ndarray attributes of the # parameters of the chain, the input variables to the chain, # and any intermediate arrays (activations, etc) created while # executing the define-by-run code of the chain. self.unique_arrays = [] # A list of UniqueArray objects, where # each object contains information such as what the array corresponds # to (variable, parameter.data, etc), weak or regular reference, # whether # it was dynamically allocated or read-only in the schedule. # It is the same length as unique_arrays. self.unique_array_infos = [] # Maps id(ndarray) to its position in self.unique_arrays # This is shared by this schedule and all deeply-contained schedules. self.array_id_to_unique_index = dict() self.backward_schedule_func = None self.verbosity_level = verbosity_level self.enable_double_backprop = enable_double_backprop self.in_vars = None self.chain = None self.schedule_built = False # A list of all parameters in the model (i.e., that exist when # build_schedule() is called. # This is shared among all deeply-contained schedules of this schedule. self.params_list = [] # This list contains the grad_var corresponding to each variable # in params_list. This is needed so that we can restore any grad_var # that is set to None by outside code. # This is shared among all deeply-contained schedules of this schedule. self.grad_var_list = [] # Maps an array id (of a parameter) to its location. # id(array) -> (index_in_self.params_list, attribute_location) self.array_id_to_param_map = dict() # Maps an array id (of an input variable for forward()) to its # positional index. # id(array) -> (index in inputs argument of forward()) self.array_id_to_input_var_map = dict() # maps a Parameter id to the parameter's index in self.params_list self.param_id_to_index = dict() # A list of tuples that specify the mappings from static schedule # arrays to parameter attributes. # These are pre-hooks that are run before running the schedule. self.param_hooks = [] # These are post hooks that are run after executing the schedule. # They are used to update parameter attributes from dynamically- # allocated arrays in the schedule. self.param_post_hooks = [] # A list of tuples that specify the mappings from static schedule # arrays to 'data' array attributes of the output variables. self.out_var_hooks = [] # A list of tuples that specify the mapping from static schedule # arrays to input variable index in the "inputs" argument of forward() # This is used to update the array references in the static schedule # that refer to the data attribute of input variables. self.in_var_hooks = [] self.dynamically_allocated_unique_index = set() # Maps an index in unique_arrays to the index in the returned # output variables, if the index corresponds to an output # variable. self.unique_ind_to_out_var_ind = dict() def get_unique_index_from_array(self, array): """Return the array index if it exists. Return the index of the array in self.unique_array_infos if the array already exists in self.unique_array_info with a valid reference. Otherwise, return None. """ ar_id = id(array) if ar_id in self.array_id_to_unique_index: # It is possible that this id is stale if a previous # array that had the same id has already been deleted. # So, verify that the existing array with this id is # still alive. unique_ind = self.array_id_to_unique_index[ar_id] info = self.unique_array_infos[unique_ind] assert ar_id == info.id if info.was_deleted(): # id was stale, so remove from the dict. del self.array_id_to_unique_index[ar_id] return None else: return self.array_id_to_unique_index[ar_id] def get_contained_schedule(self): # Make and return the backward schedule (relative to # this schedule). sched = StaticScheduleFunction(self.schedule_manager, self.verbosity_level, self.enable_double_backprop) sched.pass_depth = self.pass_depth + 1 sched.unique_arrays = self.unique_arrays sched.unique_array_infos = self.unique_array_infos sched.array_id_to_unique_index = self.array_id_to_unique_index sched.params_list = self.params_list sched.grad_var_list = self.grad_var_list sched.array_id_to_param_map = self.array_id_to_param_map sched.param_hooks = self.param_hooks sched.param_id_to_index = self.param_id_to_index return sched def is_empty(self): """Return True if this schedule is empty. """ return len(self.schedule_info_list) == 0 def append_function(self, func, args, kwargs, func_name=None): """Append a function to the static schedule. Append a function `func` to the static schedule. `func` can be any function that is decorated with `@static_code` and that was called while executing the static chain's `__call___()` method, which contains the define-by-run code. The code in the `@static_code` decorator will call this method to add the function to the schedule just after it executes in the define-by-run code as follows: `return_arrays = func(*args, **kwargs)` During the next iteration when the static chain switches from define- by-run to the static schedule, a corresponding `ScheduleInfo` object will call `func` as above, except that the scheduler might make modifications to some of the arrays in `kwargs` before and after the function is called to implement various memory optimizations. Args: func (function or method): The function to append to the schedule. This is a function that was decorated with `@static_code`. args: The arguments that were originally supplied to `func` in the define-by-run code of the static chain. kwargs: The keyword arguments that were originally supplied to `func` in the define-by-run code of the static chain. func_name (str): Optional name for `func`, for debugging purposes. return_arrays (tuple of ndarray) or None: The value that is returned by `func`, if any. """ # Check previous function in the schedule, if available. # Check the arrays in the retained inputs/outputs and force them # to remain statically allocated in the schedule. # ids of any retained arrays. retained_ids = set() last_sched_info_ind = len(self.schedule_info_list) - 1 if last_sched_info_ind >= 0: prev_sched_info = self.schedule_info_list[last_sched_info_ind] if prev_sched_info.function_node is not None: # get retained inputs/outputs. retained_in_vars = \ prev_sched_info.function_node.get_retained_inputs() retained_out_vars = \ prev_sched_info.function_node.get_retained_outputs() if (retained_in_vars is not None and retained_out_vars is not None): retained_vars = retained_in_vars + retained_out_vars elif retained_in_vars is not None: retained_vars = retained_in_vars elif retained_out_vars is not None: retained_vars = retained_out_vars else: retained_vars = None if retained_vars is not None: for var in retained_vars: retained_ids.add(id(var.data)) for keep_id in retained_ids: unique_ind = self.array_id_to_unique_index[keep_id] array_info = self.unique_array_infos[unique_ind] array_info.retain = True # Note: the following line is not actually needed. # array_info.array = array_info.weak_ref() delete_hooks = [] for unique_ind, ar_info in enumerate(self.unique_array_infos): # todo: this is O(N^2) and maybe too slow for large graphs. # Optimize it later. if ar_info.was_deleted(): if ar_info.dynamic_deletion_index is None: if self.verbosity_level >= 2: print('Adding delete hook:') delete_hooks.append(unique_ind) ar_info.dynamic_deletion_index = last_sched_info_ind + 1 ar_info.dynamic_deletion_pass_depth = self.pass_depth # Call the `@static_code`-decorated function. ret = func(*args, **kwargs) inputs_hooks = [] if 'inputs' in kwargs: in_list = kwargs['inputs'] assert isinstance(in_list, list) for ind, x in enumerate(in_list): if _is_xp(x): unique_ind = self.get_unique_index_from_array(x) if unique_ind is None: # Note: we append None here because we cannot store any # additional reference to the array. # Otherwise, it would # prevent garbage collection. Note that a # weak reference # will be stored in the ArrayInfo below. self.unique_arrays.append(None) self.unique_array_infos.append(ArrayInfo(x)) unique_ind = len(self.unique_arrays) - 1 self.array_id_to_unique_index[id(x)] = unique_ind inputs_hooks.append((ind, unique_ind)) # Now that the hook has been added, we can delete # array reference from 'args'. in_list[ind] = None outputs_hooks = [] if 'outputs' in kwargs: out_list = kwargs['outputs'] assert isinstance(out_list, list) for ind, x in enumerate(out_list): if _is_xp(x): unique_ind = self.get_unique_index_from_array(x) if unique_ind is None: self.unique_arrays.append(x) # todo: enable the following line instead once the # auto-intializing hooks are added. This will further # reduce memory usage. # self.unique_arrays.append(None) self.unique_array_infos.append(ArrayInfo(x)) unique_ind = len(self.unique_arrays) - 1 self.array_id_to_unique_index[id(x)] = unique_ind outputs_hooks.append((ind, unique_ind)) # Now that the hook has been added, we can delete # array reference from 'args'. out_list[ind] = None # A list of hooks (each is a tuple) that will be used to set # correct array references in 'unique_arrays' after executing # the static schedule function 'func'. These hooks update # the references in 'unique_arrays' to refer to the arrays # that were dynamically allocated in the return value of # 'func'. return_hooks = [] if ret is not None: assert (isinstance(ret, list) or isinstance(ret, tuple)) for ret_index, item in enumerate(ret): if _is_xp(item): # note: id might not be unique if objects have been # garbage collected. item_id = id(item) unique_index = self.get_unique_index_from_array(item) if unique_index is None: # Note: Append None instead of 'item' to prevent an # extra reference from being stored. Otherwise it # would prevent garbage collection. self.unique_arrays.append(None) ar_info = ArrayInfo(item) ar_info.dynamically_allocated = True sched_info_ind = len(self.schedule_info_list) ar_info.dynamic_allocation_index = sched_info_ind ar_info.dynamic_allocation_pass_depth = self.pass_depth self.unique_array_infos.append(ar_info) unique_index = len(self.unique_arrays) - 1 self.array_id_to_unique_index[item_id] = \ unique_index else: # Since all of the return arrays are supposed to # have been dynamically allocated inside 'func', # they had better not already be in unique_arrays. # If so, it is an error. unique_index = self.array_id_to_unique_index[item_id] print('the current id: ', item_id) print('the unique_index: ', unique_index) print('array info: ', self.unique_array_infos[unique_ind]) raise RuntimeError('Found result array from schedule ' 'function already in ' 'unique_arrays!') return_hooks.append((ret_index, unique_index)) self.dynamically_allocated_unique_index.add(unique_index) if self.verbosity_level >= 2: print('Adding function to static schedule: ', func) self.schedule_info_list.append(ScheduleInfo(func, args, kwargs, inputs_hooks, outputs_hooks, return_hooks, delete_hooks, self.unique_arrays, self.unique_array_infos, func_name=func_name)) return ret def __repr__(self): out = 'StaticSchedule:\n' if self.pass_depth == 0: depth = 'forward pass' elif self.pass_depth == 1: depth = 'backward pass' elif self.pass_depth == 2: depth = 'double backward pass' else: depth = str(self.pass_depth) out += 'Pass depth: ' + depth + '\n' out += 'Length of unique_arrays: ' + \ str(len(self.unique_arrays)) + '\n' for x in self.schedule_info_list: out += str(x) return out def debug_print_ref_counts(self): print('reference counts in unique_arrays:') for ind in range(len(self.unique_arrays)): print('index: ', ind) print('reference count: ', sys.getrefcount(self.unique_arrays[ind])) def run_param_pre_hooks(self): """Run parameter reference updater hooks. It also handles the case where the 'grad' attribute was set to 'None' by outside Chainer code. """ for hook in self.param_hooks: (unique_array_index, param_attribute_location) = hook (params_list_index, attribute_location) = param_attribute_location if attribute_location == 'data': # This is the corresponding parameter array, which might # have had its reference changed to a different array or set # to None. self.unique_arrays[unique_array_index] = \ self.params_list[params_list_index].data elif attribute_location == 'grad': # This is the corresponding parameter array, which might # have had its reference changed to a different array or set # to None. self.params_list[params_list_index].grad = \ self.unique_arrays[unique_array_index] def run_param_post_hooks(self): """Update parameter attributes after schedule is executed. If any dynamically-allocated arrays in the schedule correspond to a parameter attribute, it must be updated after the schedule is run. """ if self.verbosity_level >= 2: print('run_param_post_hooks()...') for hook in self.param_post_hooks: (unique_array_index, param_attribute_location) = hook (params_list_index, attribute_location) = param_attribute_location if attribute_location == 'data': self.params_list[params_list_index].data = \ self.unique_arrays[unique_array_index] elif attribute_location == 'grad': self.params_list[params_list_index].grad = \ self.unique_arrays[unique_array_index] def run_in_var_hooks(self, input_var_arrays): """Run hooks to update variable array references. Args: input_var_arrays (tuple of ndarray): The 'data' array attributes of the input variables to this function. """ for hook in self.in_var_hooks: (unique_array_index, in_var_ind) = hook if self.verbosity_level >= 2: print('input var hook:') print('unique_array_index: ', unique_array_index) print('in_var_ind: ', in_var_ind) print('_run_in_var_hooks(): Using this input variable array ' 'for forward pass: ', input_var_arrays[in_var_ind]) self.unique_arrays[unique_array_index] = \ input_var_arrays[in_var_ind] def debug_print_unique_arrays_info(self): for ind, item in enumerate(self.unique_arrays): print('--- unique_arrays ---') print('index: {0}; id: {1}'.format(ind, id(item))) if item is not None: print('shape: ', item.shape) if ind in self.unique_ind_to_out_var_ind: out_var_ind = self.unique_ind_to_out_var_ind[ind] print('output variable at return index: ', out_var_ind) if ind in self.dynamically_allocated_unique_index: print('Dynamically allocated inside schedule.') def run_out_var_hooks(self): """Run hooks to update output variable array references. """ for hook in self.out_var_hooks: (out_var_ind, unique_list_index) = hook out_var = self.out_vars[out_var_ind] out_var.data = self.unique_arrays[unique_list_index] if self.verbosity_level >= 2: print('StaticScheduleFunction: running output variable hook: ' 'out_var_ind, unique_list_index): ', hook) def set_out_variables(self, out_vars): """Set output variables. This should be called after the define-by-run code in the chain's `__call__()` has already run but before running the static schedule. Args: out_vars (list of Variable): The (flattened) list of output variables obtained by performing a define-by-run forward pass (or corresponding backward pass) on the local sub-graph corresponding to the static chain. """ self.out_vars = out_vars # Create output-variable update hooks. for var_ind, var in enumerate(out_vars): if var is not None: key = id(var.data) if key in self.array_id_to_unique_index: unique_list_index = self.array_id_to_unique_index[key] self.out_var_hooks.append((var_ind, unique_list_index)) self.unique_ind_to_out_var_ind[unique_list_index] = var_ind else: raise RuntimeError('Could not find output variable in ' 'unique_arrays.') def build_schedule(self, chain, in_vars): """Build the static schedule. Perform one-time post-processing on the functions and arguments that were previously supplied in 'append_function()' to create the static schedule. This method must be called after the final call of 'append_function()' and before calling 'forward()' for the first time. Args: chain: The static chain that uses this scheudle. in_vars (list of Variable): The input variables to this static schedule. This are the input variables (each having no creator) of the local sub-graph corresponding to the static chain. """ self.chain = chain self.in_vars = in_vars # Iterate through all array info objects and for any arrays that # still have a valid reference, copy into unique_arrays. if self.verbosity_level >= 2: print('Building schedule for pass depth: ', self.pass_depth) for ind, info in enumerate(self.unique_array_infos): if self.verbosity_level >= 2: print('unique array index: ', ind) print('array info: ', info) if not info.was_deleted(): self.unique_arrays[ind] = info.weak_ref() # Verify that all array references are actually unique. unique_ids = set() for ar in self.unique_arrays: if ar is not None: assert id(ar) not in unique_ids unique_ids.add(id(ar)) for param in chain.params(): param_key = id(param) if param_key not in self.param_id_to_index: self.params_list.append(param) grad_var = param.grad_var self.grad_var_list.append(grad_var) param_index = len(self.params_list) - 1 self.param_id_to_index[param_key] = param_index else: # We have seen this parameter before. param_index = self.param_id_to_index[param_key] grad_var = param.grad_var self.grad_var_list[param_index] = grad_var if param.data is not None: key = id(param.data) if key not in self.array_id_to_param_map: self.array_id_to_param_map[key] = (param_index, 'data') if param.grad is not None: key = id(param.grad) if key not in self.array_id_to_param_map: self.array_id_to_param_map[key] = (param_index, 'grad') for var_ind, in_var in enumerate(self.in_vars): assert in_var.data is not None key = id(in_var.data) self.array_id_to_input_var_map[key] = var_ind # Iterate over all arrays used in the schedule and check which ones # correspond to parameter arrays or input variables. When a match # is found, create a corresponding hook function. This hook will # run just before executing the schedule and set the array # references used in the schedule to be consistent with the # input variables and parameters. assert len(self.unique_arrays) > 0 for unique_array_index, ar in enumerate(self.unique_arrays): key = id(ar) # Create pre-run parameter hooks. if key in self.array_id_to_param_map: param_attribute_location = self.array_id_to_param_map[key] param_hook = (unique_array_index, param_attribute_location) self.param_hooks.append(param_hook) # Create pre-run input variable hooks. if key in self.array_id_to_input_var_map: in_var_ind = self.array_id_to_input_var_map[key] in_var_hook = (unique_array_index, in_var_ind) self.in_var_hooks.append(in_var_hook) if self.verbosity_level >= 2: print('build_schedule(): Adding input variable hook: ', in_var_hook) print('For input variable: ', ar) # Create post-run hooks for any arrays that are dynamically # allocated inside the schedule. if unique_array_index in self.dynamically_allocated_unique_index: if key in self.array_id_to_param_map: param_attribute_location = self.array_id_to_param_map[key] param_hook = (unique_array_index, param_attribute_location) self.param_post_hooks.append(param_hook) if self.verbosity_level >= 2: print('self.param_hooks: ', self.param_hooks) self.debug_print_unique_arrays_info() # todo: We can potentially reduce memory usage by freeing memory # of intermediate arrays in self.unique_arrays # once they are no longer needed in the schedule or by # parameters. print('end of build_schedule()') self.schedule_built = True def forward(self, inputs): if self.verbosity_level >= 2: print('Calling StaticScheduleFunction.forward()...') # Note: This method will be invoked every iteration starting # from the second # iteration. That is because the corresponding define-by-run # code runs instead # during the first iteration. if not self.schedule_built: raise RuntimeError('forward() was called before ' 'build_schedule()!') self.run_param_pre_hooks() self.run_in_var_hooks(inputs) if self.verbosity_level >= 2: print('Running static schedule...') # Run each function in the static schedule. for x in self.schedule_info_list: x() if self.verbosity_level >= 2: self.debug_print_unique_arrays_info() self.run_out_var_hooks() self.run_param_post_hooks() ret = [] for y in self.out_vars: if y is None or y.data is None: ret.append(None) else: # todo: add test case for an example where the following # copy is required (evaluation mode, repeated calls of # chain that reuse same schedule). ret.append(y.data.copy()) return tuple(ret) def backward(self, target_input_indexes, grad_outputs): if self.verbosity_level >= 2: print('Calling StaticScheduleFunction.backward()...') # The first time this method is called, the define-by-run code is # executed in order to create a static schedule. self.schedule_manager.end_forward() if self.backward_schedule_func is None: print('Creating new backward schedule...') # Create backward schedule and run define-by-run backward code. self.backward_schedule_func = self.get_contained_schedule() # Make local copies of the variables in grad_outputs. new_grad_outputs = [] for var in grad_outputs: # Replace each input variable with a new variable having # the same data. new_grad_outputs.append(chainer.Variable(var.data)) with chainer.using_config('schedule_func', self.backward_schedule_func): with chainer.using_config('enable_backprop', True): for ind, var in enumerate(new_grad_outputs): # todo: possibly don't need the following: self.out_vars[ind].grad = new_grad_outputs[ind].data inputs = [param for param in self.chain.params()] for var in self.in_vars: inputs.append(var) # Need shorter var to avoid "line too long error" ugh = self.enable_double_backprop chainer.grad(self.out_vars, inputs, grad_outputs=new_grad_outputs, set_grad=True, enable_double_backprop=ugh) # We no longer need the backward graph from self.out_vars, so # unchain them. # todo (vogel): enable this eventually. For now, it # causes some needed variables to be set to None # in some models such as CIFAR example. # for var in self.out_vars: # var.unchain_backward() # Note: var.grad_var is allowed to be None below: backward_out_vars = [var.grad_var for var in self.in_vars] self.backward_schedule_func.set_out_variables(backward_out_vars) for n in range(len(self.in_vars)): self.in_vars[n] = None if self.verbosity_level >= 2: print('building backward schedule.') self.backward_schedule_func.build_schedule(self.chain, new_grad_outputs) return self.backward_schedule_func.apply(grad_outputs) class ScheduleManager(object): """A manager of static schedules for a static chain. This is a container of the static schedules that are used by a static chain. Args: minimize_cache_size (bool): If `True`, attempt to reduce memory usage by clearing the cached schedules whenever the training mode changes (that is, whenever `chainer.config.train` changes value) or whenever the mini-batch size changes. """ def __init__(self, minimize_cache_size=True, verbosity_level=0): # Maps a key string to a list of schedule functions. self.schedules = dict() self.minimize_cache_size = minimize_cache_size self.in_use_count = dict() self.forward_over = False self.prev_train_config = None self.max_in_use_train = 0 self.train_count = 0 self.verbosity_level = verbosity_level def get_schedule(self, in_vars, enable_double_backprop=False): """Get a static schedule. Return a static schedule object (that is, an instance of ``StaticScheduleFunction``) that is compatible with the current configuration and input variables to the supplied chain. If there is no existing schedule available, return an empty schedule object. During the usual "training mode" (that is, when both `chainer.config.enable_backprop` and `chainer.config.train` are `True`), this method will always return a distince static schedule each time it is called within the same iteration. It will also try to reuse existing schedules across iterations. Therefore, any schedule that is returned in a given iteration cannot be returned again until the following iteration. However, if either of these flags is 'False', then this method may return the same schedule instance multiple times within the same iteration, as long as it is compatible with `in_vars`. Note that in order to implement the above behavior, the schedule manager must be informed when the current iteration has finished. This is accomplished by calling `end_forward()` after the iteration has finished. If a backward pass is performed, then `end_forward()` will be automatically called. Otherwise, it will not be called and the user will be responsible for calling it. Args: in_vars (tuple of :class:`~chainer.Variable`): The input variables to the chain. Returns: An instance of ``StaticScheduleFunction``. """ if self.forward_over: self.forward_over = False if self.minimize_cache_size: if chainer.config.train != self.prev_train_config: # Training config changed, so clear caches. self.prev_train_config = chainer.config.train if self.verbosity_level >= 2: print('Clearing schedule cache...') self.schedules.clear() self.in_use_count.clear() if (chainer.config.train is False or chainer.config.enable_backprop is False): key_str = 'test:' + \ ''.join(str(x.shape) + str(x.dtype) for x in in_vars) # If the maximum number of in-use schedules in any iteration # during training mode was exactly 1, assume it should also # be 1 for test mode. if key_str in self.schedules: sched_list = self.schedules[key_str] sched = sched_list[0] else: # avoid "line too long": vb = self.verbosity_level edb = enable_double_backprop sched = StaticScheduleFunction(self, verbosity_level=vb, enable_double_backprop=edb) self.schedules[key_str] = [sched] return sched else: key_str = 'train:' + \ ''.join(str(x.shape) + str(x.dtype) for x in in_vars) self.train_count += 1 if key_str in self.schedules: sched_list = self.schedules[key_str] available_index = self.in_use_count[key_str] if available_index >= len(sched_list): # avoid "line too long": vb = self.verbosity_level edb = enable_double_backprop sched = StaticScheduleFunction(self, verbosity_level=vb, enable_double_backprop=edb) sched_list.append(sched) sched = sched_list[available_index] self.in_use_count[key_str] = available_index + 1 else: # avoid "line too long": vb = self.verbosity_level edb = enable_double_backprop sched = StaticScheduleFunction(self, verbosity_level=vb, enable_double_backprop=edb) self.schedules[key_str] = [sched] self.in_use_count[key_str] = 1 return sched def end_forward(self): """Make in-use schedules available for use in next iteration. Set the in-use status of all schedules to "not in use" so that they can be reused in the next iteration. In the case that test mode is active (`chainer.config.train` is `False`) and the static chain corresponding to this manager was not called more than once in any iteration during training mode, then this method will be called automatically. """ if not self.forward_over: for key in self.in_use_count: self.in_use_count[key] = 0 self.forward_over = True if self.train_count > self.max_in_use_train: self.max_in_use_train = self.train_count if self.verbosity_level >= 2: print('Maximum in-use schedules per training iteration: ', self.max_in_use_train) self.train_count = 0 def __repr__(self): out = 'ScheduleManager:\n' for key_str in self.schedules: out += 'key string: ' + key_str sched_list = self.schedules[key_str] out += ' -> schedule list of length: ' + \ str(len(sched_list)) + '\n' for sched in sched_list: out += str(sched) return out def static_graph(*args, **kwargs): """Decorator to mark a Chain's ``__call__()`` as a static sub-graph. This decorator marks the define-by-run code inside the `__call__()` method of a Chain instance as corresponding to a static computation graph or sub-graph. Such a chain will be referred to as a 'static chain'. This allows various "static graph" optimizations to be performed, which can result in significant speedups for some models. When this decorator is used, the chain's define-by-run code executes during the first iteration as usual. However, while the define-by-run code is executing, a trace is also performed to incrementally create a corresponding static schedule. This static schedule will only contain the subset of the computations inside the define-by-run code that actually needs to run every iteration. Specifically, this will contain the code inside any functions called that were annotated with the `@static_code` decorator, which will include all Chainer built-in functions, as well as any user-defined functions that use `@static_code`. Then, starting from the second iteration, when the static chain is called, its static schedule code will be executed instead of its define-by-run code. However, the user must also be careful of the following: - The user is responsible for applying this decorator correctly. The framework does not check that the define-by-run code corresponds to a static graph. The graph can be different between training and evaluation mode (such as when dropout and/or batch normalization are used), but should otherwise be static. - When `chainer.config.enable_backprop` is enabled, if a backward pass is not performed each iteration, then the user code must call a method `chain.schedule_manager.end_forward()`on the static chain each iteration. - Static graphs allow tradeoffs between computation and memory usage. For example, the `minimize_cache_size` argument will typically result in higher memory usage when set to `False` because all cached schedules are retained. - When this feature is enabled, only the Chainer function and/or link calls inside the chain's `__call__()` method will be included in the static schedule by default. An other code that the user puts in `__call__()`, such as a print statement or code to increment a counter for example, will not automatically get added. We will refer to such code other than Chainer function/link calls as "side-effect" code. Since side-effect code does not get included in the static schedule by default, this means that it will only every execute once, during the first iteration. There is a way to force side-effect code to be included in the static schedule, however: the user can wrapp such code inside a function that is decorated with `@static_code` to ensure that it gets added to the static schedule. For an example of this, refer to the documentation. - This feature is experimental and advanced optimizations such as kernel fusion and various memory optimizations are not implemented yet. Usage: This decorator should only be applied to define-by-run code that actually corresponds to a static subgraph. Refer to the documenation for additional details and examples of correct usage. This decorator should be applied to each of the largest static subgraphs in the model; it can also be applied to a static subgraph that is not the largest subgraph, but that could result in reduced performance. It is not currently allowed to mark a chain as static if it is contained within another chain that is also marked as being static. For example, suppose a static graph `A` contains a static sub-graph `B`. Then, only the chain corresponding to `A` should be marked as static and the chain corresponding to `B` should not be marked as static. The behavior of a static chain depends on the training mode flag, `chainer.config.train`. If it is `True`, then a static chain that is called multiple times will try to use a distinct static schedule object (that is, call a distinct instance of a FunctionNode that implements that static schedule) on each call. The same schedule instance cannot be reused until the forward pass has completed, which is signaled by performing a backward pass through the model. It is therefore important that the backward pass be performed after each forward pass during training. Since this is usually the case, most usages of static chain will not required any modifications to existing code other than applying this decorator. However, if you would like to perform multiple forward passes during training before performing a backward pass, then you must call `chain.schedule_manager.end_forward()` after the end of each forward pass. If test mode is active (`chainer.config.train` is `False`) then it is not necessary to inform the chain at the end of each forward pass because in test mode, a static chain always attempts to reuse existing static schedule objects. The same static schedule can be reused during a single forward pass, because it is not necessary to compute gradients. It is also possible to disable static optimzations while in test mode by setting the decorator argument `force_test_define_by_run=True`. Note: If either 'chainer.config.enable_backprop' or 'chainer.config.train' is set to 'False', then cached static schedules will be reused when possible to reduce memory usage. Double-backprop: Double-backpropagation is not enabled by default. It can be enabled by supplying the keyword argument ``enable_double_backprop=True`` to this decorator. Note: this feature has not been tested yet. Restrictions on input arguments and return values of a static chain: Recall that unlike a function, there is no restrictions on the arguments to a chain. However, there currently are some restrictions when a static chain is used. Specifically, the arguments to a static chain must consist of a variable, list or tuple. In the case of a list or tuple, the elements are required to be an instance of variable, list, or tuple. There can be an arbitrary number of nested lists/ tuples. No other object types are allowed. In addition, keyword arguments are not allowed. The return value of a static chain must be a variable, list, or tuple in which each element of the list or tuple is also a variable, list, or tuple. This decorator can be supplied with the following optional keyword arguments. This is an experimental feature, and the API and arguments might change Args: force_test_define_by_run (bool): If `True`, disable static graph optimizations during test mode (that is, when `chainer.config.train` is False). This may be needed in order for some existing RNN links such as LSTM to work correctly, since some existing links do not correspond to a static graph in some cases. The default is `False`. minimize_cache_size (bool): If `True`, minimize the number of cached static schedules in order to reduce memory usage. For example, if the mini-batch size changes or the training mode changes, the schedules will need to be recomputed, but memory is also saved by not retaining all cached schedules. The default value is `True`. verbosity_level (int): Depending on the value, print additional information: 0: Warnings only. (the default value) 1: Show only information that is collected during the first iteration and when a new static schedule is created. 2: Detailed debugging information, possibly showing new information every iteration. enable_double_backprop (bool): If `True`, enable double-backprop. The default value is `False` (not enabled). Returns: Wrapped ``__call__()`` method with static chain support. """ # todo: consider to allow nested use of this decorator. force_test_define_by_run = False # todo: enable eventually minimize_cache_size = False verbosity_level = 0 enable_double_backprop = False zero_args = False if len(args) == 1 and not kwargs and callable(args[0]): callable_arg = args[0] zero_args = True elif kwargs: if 'force_test_define_by_run' in kwargs: force_test_define_by_run = kwargs['force_test_define_by_run'] if 'minimize_cache_size' in kwargs: minimize_cache_size = kwargs['minimize_cache_size'] if 'verbosity_level' in kwargs: verbosity_level = kwargs['verbosity_level'] if 'enable_double_backprop' in kwargs: enable_double_backprop = kwargs['enable_double_backprop'] def wrap(func): def wrapped_func(*inner_args, **inner_kwargs): # The static subgraph optimization feature can be turned off using # a configuration, in which case this decorator merely calls the # wrapped function without introducing any side effects. if not chainer.config.use_static_graph: return func(*inner_args, **inner_kwargs) if verbosity_level >= 2: print('Calling static chain...') chain = inner_args[0] # The arguments to `__call__()` of the static chain. # These could consist of any combination of nested lists and/or # tuples of variables or arrays. chain_args = inner_args[1:] if chainer.config.train is False and force_test_define_by_run: return func(*inner_args, **inner_kwargs) chain_args_flat, in_unflatten_inds, __ = _flatten_args(chain_args) # Since it is allowed for in_vars to be either variables or arrays, # we force to variables. flat_vars = [] for x in chain_args_flat: # This assumes x is either a variable or ndarray. # todo: check this and handle case when it is not. if not isinstance(x, chainer.Variable): flat_vars.append(chainer.Variable(x)) else: flat_vars.append(x) flat_vars = tuple(flat_vars) if not hasattr(chain, 'schedule_manager'): chain.schedule_manager = ScheduleManager( minimize_cache_size=minimize_cache_size, verbosity_level=verbosity_level) schedule_manager = chain.schedule_manager # To prevent "line too long" error edb = enable_double_backprop chain.static_schedule = \ schedule_manager.get_schedule(flat_vars, enable_double_backprop=edb) if verbosity_level >= 2: print('Current schedule manager info: ', schedule_manager) if not chain.static_schedule.is_empty(): # Call the static schedule code. if verbosity_level >= 2: print('This is the 2nd or greater iteration. Calling ' 'the existing static schedule...') chain.static_schedule.debug_print_ref_counts() out_vars_flat = chain.static_schedule.apply(flat_vars) out_vars = _unflatten_args(out_vars_flat, chain._out_vars_unflatten_inds) else: # This is the first iteration. Calling the define-by-run code. assert isinstance(chain, chainer.Chain) if verbosity_level >= 2: print('This is the first iteration. Calling the ' 'define-by-run code.: ', func) # First check that this chain is not called from inside another # static chain because it is not allowed. if chainer.config.schedule_func is not None: raise RuntimeError('Not allowed to nest static chains: ', chain) new_args = [] new_args.append(chain) new_flat_vars = [] for var in flat_vars: # Replace each input variable with a new variable having # the same data. This is needed so that the chain-local # computation graph will be rooted at the input variables. new_flat_vars.append(chainer.Variable(var.data)) unflat_in_args = _unflatten_args_as_list(new_flat_vars, in_unflatten_inds) for item in unflat_in_args: new_args.append(item) inner_args = tuple(new_args) with chainer.using_config('schedule_func', chain.static_schedule): # Execute the chain's call() method. As the define-by-run # code executes, the static schedule is constructed. out_vars = func(*inner_args, **inner_kwargs) out_vars_flat_dbr, chain._out_vars_unflatten_inds, __ = \ _flatten_args(out_vars) sched_out_vars = list(out_vars_flat_dbr) chain.static_schedule.set_out_variables(sched_out_vars) # Mark the static schedule as complete. chain.static_schedule.build_schedule(chain, new_flat_vars) # Now that the static schedule is available, call it using the # flattened input variables. This will cause the # static schedule function node to be included in the # computational graph. out_vars_flat = chain.static_schedule.apply(flat_vars) out_vars = _unflatten_args(out_vars_flat, chain._out_vars_unflatten_inds) if verbosity_level >= 2: print('Returing from 1st call of the static chain.') return out_vars return wrapped_func if zero_args: return wrap(callable_arg) else: return wrap def _flatten_args(xs): """Flatten the input into a tuple of variables. In the typical case, `xs` is a tuple or list of objects where each object is either a variable, list, or tuple. In the case where it is a list of tuple, the objects in the list or tuple could also be either a variable, list or tuple. Although the non-list and non-tuple items are typically an instance of variable, any object other than list or tuple is allowed. This function simply flattens the hierarchical lists/tuples so that all objects that are deeply contained in `xs` that are non-list and non-tuple will be returned in a single tuple. Args: xs: Returns: The flattened tuple, allong with the indecies and count so that the items can be unflattened later (i.e., by calling `_unflatten_args()`. fixme: does not work if xs is a variable only. """ inds = [] ys = [] i = 0 if not isinstance(xs, (list, tuple)): inds.append(('s', )) return (xs,), inds, 0 for x in xs: if isinstance(x, (list, tuple)): x, sub_inds, total = _flatten_args(x, ) inds.append(('i', i, i+total, sub_inds)) i += total else: x = [x] inds.append(('f', i)) i += 1 ys.extend([y for y in x]) return tuple(ys), inds, i # todo: this only outputs tuples of tuples. Any list in the original input # will be converted to a tuple, changing the types of the input arguments # to the static chain. def _unflatten_args(xs, inds): ys = [] for ind in inds: code = ind[0] if code == 's': return xs[0] elif code == 'i': i_start, i_end, sub_inds = ind[1:] y = _unflatten_args(xs[i_start:i_end], sub_inds) else: i = ind[1] y = xs[i] ys.append(y) return tuple(ys) def _unflatten_args_as_list(xs, inds): ys = [] for ind in inds: code = ind[0] if code == 's': return xs[0] elif code == 'i': i_start, i_end, sub_inds = ind[1:] y = _unflatten_args(xs[i_start:i_end], sub_inds) else: i = ind[1] y = xs[i] ys.append(y) return ys
67,103
44.617947
79
py
chainer
chainer-master/chainer/datasets/text_dataset.py
import io import sys import threading import six from chainer.dataset import dataset_mixin class TextDataset(dataset_mixin.DatasetMixin): """Dataset of a line-oriented text file. This dataset reads each line of text file(s) on every call of the :meth:`__getitem__` operator. Positions of line boundaries are cached so that you can quickliy random access the text file by the line number. .. note:: Cache will be built in the constructor. You can pickle and unpickle the dataset to reuse the cache, but in that case you are responsible to guarantee that files are not modified after the cache has built. Args: paths (str or list of str): Path to the text file(s). If it is a string, this dataset reads a line from the text file and emits it as :class:`str`. If it is a list of string, this dataset reads lines from each text file and emits it as a tuple of :class:`str`. In this case, number of lines in all files must be the same. encoding (str or list of str): Name of the encoding used to decode the file. See the description in :func:`open` for the supported options and how it works. When reading from multiple text files, you can also pass a list of :class:`str` to use different encoding for each file. errors (str or list of str): String that specifies how decoding errors are to be handled. See the description in :func:`open` for the supported options and how it works. When reading from multiple text files, you can also pass a list of :class:`str` to use different error handling policy for each file. newline (str or list of str): Controls how universal newlines mode works. See the description in :func:`open` for the supported options and how it works. When reading from multiple text files, you can also pass a list of :class:`str` to use different mode for each file. filter_func (callable): Function to filter each line of the text file. It should be a function that takes number of arguments equals to the number of files. Arguments are lines loaded from each file. The filter function must return True to accept the line, or return False to skip the line. """ def __init__( self, paths, encoding=None, errors=None, newline=None, filter_func=None): if isinstance(paths, six.string_types): paths = [paths] elif not paths: raise ValueError('at least one text file must be specified') if isinstance(encoding, six.string_types) or encoding is None: encoding = [encoding] * len(paths) if isinstance(errors, six.string_types) or errors is None: errors = [errors] * len(paths) if isinstance(newline, six.string_types) or newline is None: newline = [newline] * len(paths) if not (len(paths) == len(encoding) == len(errors) == len(newline)): raise ValueError( 'length of each option must match with the number of ' 'text files to read') self._paths = paths self._encoding = encoding self._errors = errors self._newline = newline self._fps = None self._open() # Line number is 0-origin. # `lines` is a list of line numbers not filtered; if no filter_func is # given, it is range(linenum)). # `bounds` is a list of cursor positions of line boundaries for each # file, i.e. i-th line of k-th file starts at `bounds[k][i]`. linenum = 0 lines = [] bounds = tuple([[0] for _ in self._fps]) while True: data = [fp.readline() for fp in self._fps] if not all(data): # any of files reached EOF if any(data): # not all files reached EOF raise ValueError( 'number of lines in files does not match') break for i, fp in enumerate(self._fps): bounds[i].append(fp.tell()) if filter_func is not None and filter_func(*data): lines.append(linenum) linenum += 1 if filter_func is None: lines = six.moves.range(linenum) self._bounds = bounds self._lines = lines self._lock = threading.Lock() def __getstate__(self): state = self.__dict__.copy() del state['_fps'] del state['_lock'] return state def __setstate__(self, state): self.__dict__ = state self._open() self._lock = threading.Lock() def __len__(self): return len(self._lines) def _open(self): self._fps = [ io.open( path, mode='rt', encoding=encoding, errors=errors, newline=newline, ) for path, encoding, errors, newline in six.moves.zip(self._paths, self._encoding, self._errors, self._newline) ] def close(self): """Manually closes all text files. In most cases, you do not have to call this method, because files will automatically be closed after TextDataset instance goes out of scope. """ exc = None for fp in self._fps: try: fp.close() except Exception: exc = sys.exc_info() if exc is not None: six.reraise(*exc) def get_example(self, idx): if idx < 0 or len(self._lines) <= idx: raise IndexError linenum = self._lines[idx] self._lock.acquire() try: for k, fp in enumerate(self._fps): fp.seek(self._bounds[k][linenum]) lines = [fp.readline() for fp in self._fps] if len(lines) == 1: return lines[0] return tuple(lines) finally: self._lock.release()
6,272
35.260116
78
py
chainer
chainer-master/chainer/datasets/kuzushiji_mnist.py
import os import numpy import chainer from chainer.dataset import download from chainer.datasets._mnist_helper import make_npz from chainer.datasets._mnist_helper import preprocess_mnist _kuzushiji_mnist_labels = [('o', u'\u304A'), ('ki', u'\u304D'), ('su', u'\u3059'), ('tsu', u'\u3064'), ('na', u'\u306A'), ('ha', u'\u306F'), ('ma', u'\u307E'), ('ya', u'\u3084'), ('re', u'\u308C'), ('wo', u'\u3092')] def get_kuzushiji_mnist_labels(): """Provides a list of labels for the Kuzushiji-MNIST dataset. Returns: List of labels in the form of tuples. Each tuple contains the character name in romaji as a string value and the unicode codepoint for the character. """ return _kuzushiji_mnist_labels def get_kuzushiji_mnist(withlabel=True, ndim=1, scale=1., dtype=None, label_dtype=numpy.int32, rgb_format=False): """Gets the Kuzushiji-MNIST dataset. `Kuzushiji-MNIST (KMNIST) <http://codh.rois.ac.jp/kmnist/>`_ is a set of hand-written Japanese characters represented by grey-scale 28x28 images. In the original images, each pixel is represented by one-byte unsigned integer. This function scales the pixels to floating point values in the interval ``[0, scale]``. This function returns the training set and the test set of the official KMNIST dataset. If ``withlabel`` is ``True``, each dataset consists of tuples of images and labels, otherwise it only consists of images. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. ndim (int): Number of dimensions of each image. The shape of each image is determined depending on ``ndim`` as follows: - ``ndim == 1``: the shape is ``(784,)`` - ``ndim == 2``: the shape is ``(28, 28)`` - ``ndim == 3``: the shape is ``(1, 28, 28)`` scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. rgb_format (bool): if ``ndim == 3`` and ``rgb_format`` is ``True``, the image will be converted to rgb format by duplicating the channels so the image shape is (3, 28, 28). Default is ``False``. Returns: A tuple of two datasets. If ``withlabel`` is ``True``, both datasets are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ dtype = chainer.get_dtype(dtype) train_raw = _retrieve_kuzushiji_mnist_training() train = preprocess_mnist(train_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) test_raw = _retrieve_kuzushiji_mnist_test() test = preprocess_mnist(test_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) return train, test def _retrieve_kuzushiji_mnist_training(): base_url = 'http://codh.rois.ac.jp/' urls = [base_url + 'kmnist/dataset/kmnist/train-images-idx3-ubyte.gz', base_url + 'kmnist/dataset/kmnist/train-labels-idx1-ubyte.gz'] return _retrieve_kuzushiji_mnist('train.npz', urls) def _retrieve_kuzushiji_mnist_test(): base_url = 'http://codh.rois.ac.jp/' urls = [base_url + 'kmnist/dataset/kmnist/t10k-images-idx3-ubyte.gz', base_url + 'kmnist/dataset/kmnist/t10k-labels-idx1-ubyte.gz'] return _retrieve_kuzushiji_mnist('test.npz', urls) def _retrieve_kuzushiji_mnist(name, urls): root = download.get_dataset_directory('pfnet/chainer/kuzushiji_mnist') path = os.path.join(root, name) return download.cache_or_load_file( path, lambda path: make_npz(path, urls), numpy.load)
4,102
40.444444
79
py
chainer
chainer-master/chainer/datasets/concatenated_dataset.py
from chainer.dataset import dataset_mixin class ConcatenatedDataset(dataset_mixin.DatasetMixin): """Dataset which concatenates some base datasets. This dataset wraps some base datasets and works as a concatenated dataset. For example, if a base dataset with 10 samples and another base dataset with 20 samples are given, this dataset works as a dataset which has 30 samples. Args: datasets: The underlying datasets. Each dataset has to support :meth:`__len__` and :meth:`__getitem__`. """ def __init__(self, *datasets): self._datasets = datasets def __len__(self): return sum(len(dataset) for dataset in self._datasets) def get_example(self, i): if i < 0: raise IndexError for dataset in self._datasets: if i < len(dataset): return dataset[i] i -= len(dataset) raise IndexError
939
27.484848
78
py
chainer
chainer-master/chainer/datasets/image_dataset.py
import bisect import io import os import threading import zipfile import numpy try: from PIL import Image available = True except ImportError as e: available = False _import_error = e import six import chainer from chainer.dataset import dataset_mixin def _read_image_as_array(path, dtype): f = Image.open(path) try: image = numpy.asarray(f, dtype=dtype) finally: # Only pillow >= 3.0 has 'close' method if hasattr(f, 'close'): f.close() return image def _postprocess_image(image): if image.ndim == 2: # image is greyscale image = image[..., None] return image.transpose(2, 0, 1) class ImageDataset(dataset_mixin.DatasetMixin): """Dataset of images built from a list of paths to image files. This dataset reads an external image file on every call of the :meth:`__getitem__` operator. The paths to the image to retrieve is given as either a list of strings or a text file that contains paths in distinct lines. Each image is automatically converted to arrays of shape ``channels, height, width``, where ``channels`` represents the number of channels in each pixel (e.g., 1 for grey-scale images, and 3 for RGB-color images). .. note:: **This dataset requires the Pillow package being installed.** In order to use this dataset, install Pillow (e.g. by using the command ``pip install Pillow``). Be careful to prepare appropriate libraries for image formats you want to use (e.g. libpng for PNG images, and libjpeg for JPG images). .. warning:: **You are responsible for preprocessing the images before feeding them to a model.** For example, if your dataset contains both RGB and grayscale images, make sure that you convert them to the same format. Otherwise you will get errors because the input dimensions are different for RGB and grayscale images. Args: paths (str or list of strs): If it is a string, it is a path to a text file that contains paths to images in distinct lines. If it is a list of paths, the ``i``-th element represents the path to the ``i``-th image. In both cases, each path is a relative one from the root path given by another argument. root (str): Root directory to retrieve images from. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). """ def __init__(self, paths, root='.', dtype=None): _check_pillow_availability() if isinstance(paths, six.string_types): with open(paths) as paths_file: paths = [path.strip() for path in paths_file] self._paths = paths self._root = root self._dtype = chainer.get_dtype(dtype) def __len__(self): return len(self._paths) def get_example(self, i): path = os.path.join(self._root, self._paths[i]) image = _read_image_as_array(path, self._dtype) return _postprocess_image(image) class LabeledImageDataset(dataset_mixin.DatasetMixin): """Dataset of image and label pairs built from a list of paths and labels. This dataset reads an external image file like :class:`ImageDataset`. The difference from :class:`ImageDataset` is that this dataset also returns a label integer. The paths and labels are given as either a list of pairs or a text file contains paths/labels pairs in distinct lines. In the latter case, each path and corresponding label are separated by white spaces. This format is same as one used in Caffe. .. note:: **This dataset requires the Pillow package being installed.** In order to use this dataset, install Pillow (e.g. by using the command ``pip install Pillow``). Be careful to prepare appropriate libraries for image formats you want to use (e.g. libpng for PNG images, and libjpeg for JPG images). .. warning:: **You are responsible for preprocessing the images before feeding them to a model.** For example, if your dataset contains both RGB and grayscale images, make sure that you convert them to the same format. Otherwise you will get errors because the input dimensions are different for RGB and grayscale images. Args: pairs (str or list of tuples): If it is a string, it is a path to a text file that contains paths to images in distinct lines. If it is a list of pairs, the ``i``-th element represents a pair of the path to the ``i``-th image and the corresponding label. In both cases, each path is a relative one from the root path given by another argument. root (str): Root directory to retrieve images from. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. """ def __init__(self, pairs, root='.', dtype=None, label_dtype=numpy.int32): _check_pillow_availability() if isinstance(pairs, six.string_types): pairs_path = pairs with open(pairs_path) as pairs_file: pairs = [] for i, line in enumerate(pairs_file): pair = line.strip().split() if len(pair) != 2: raise ValueError( 'invalid format at line {} in file {}'.format( i, pairs_path)) pairs.append((pair[0], int(pair[1]))) self._pairs = pairs self._root = root self._dtype = chainer.get_dtype(dtype) self._label_dtype = label_dtype def __len__(self): return len(self._pairs) def get_example(self, i): path, int_label = self._pairs[i] full_path = os.path.join(self._root, path) image = _read_image_as_array(full_path, self._dtype) label = numpy.array(int_label, dtype=self._label_dtype) return _postprocess_image(image), label class LabeledZippedImageDataset(dataset_mixin.DatasetMixin): """Dataset of zipped image and label pairs. This dataset is zip version of :class:`LabeledImageDataset`. It takes a zipfile like :class:`ZippedImageDataset`. The label file shall contain lines like text file used in :class:`LabeledImageDataset`, but a filename in each line of the label file shall match with a file in the zip archive. Args: zipfilename (str): Path to a zipfile with images labelfilename (str): Path to a label file. ``i``-th line shall contain a filename and an integer label that corresponds to the ``i``-th sample. A filename in the label file shall match with a filename in the zip file given with `zipfilename`. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. """ def __init__(self, zipfilename, labelfilename, dtype=None, label_dtype=numpy.int32): _check_pillow_availability() pairs = [] with open(labelfilename) as pairs_file: for i, line in enumerate(pairs_file): pair = line.strip().split() if len(pair) != 2: raise ValueError( 'invalid format at line {} in file {}'.format( i, pairs_file)) pairs.append((pair[0], int(pair[1]))) self._pairs = pairs self._label_dtype = label_dtype self._zipfile = ZippedImageDataset(zipfilename, dtype=dtype) def __len__(self): return len(self._pairs) def get_example(self, i): path, int_label = self._pairs[i] label = numpy.array(int_label, dtype=self._label_dtype) return self._zipfile.get_example(path), label class MultiZippedImageDataset(dataset_mixin.DatasetMixin): """Dataset of images built from a list of paths to zip files. This dataset reads an external image file in given zipfiles. The zipfiles shall contain only image files. This shall be able to replace ImageDataset and works better on NFS and other networked file systems. The user shall find good balance between zipfile size and number of zipfiles (e.g. granularity) Args: zipfilenames (list of strings): List of zipped archive filename. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). """ def __init__(self, zipfilenames, dtype=None): self._zfs = [ZippedImageDataset(fn, dtype) for fn in zipfilenames] self._zpaths_accumlens = [0] zplen = 0 for zf in self._zfs: zplen += len(zf) self._zpaths_accumlens.append(zplen) def __len__(self): return self._zpaths_accumlens[-1] def get_example(self, i): tgt = bisect.bisect(self._zpaths_accumlens, i) - 1 lidx = i - self._zpaths_accumlens[tgt] return self._zfs[tgt].get_example(lidx) class ZippedImageDataset(dataset_mixin.DatasetMixin): """Dataset of images built from a zip file. This dataset reads an external image file in the given zipfile. The zipfile shall contain only image files. This shall be able to replace ImageDataset and works better on NFS and other networked file systems. If zipfile becomes too large you may consider ``MultiZippedImageDataset`` as a handy alternative. Known issue: pickle and unpickle on same process may cause race condition on ZipFile. Pickle of this class is expected to be sent to different processess via ChainerMN. Args: zipfilename (str): a string to point zipfile path dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). """ def __init__(self, zipfilename, dtype=None): self._zipfilename = zipfilename self._zf = zipfile.ZipFile(zipfilename) self._zf_pid = os.getpid() self._dtype = chainer.get_dtype(dtype) self._paths = [x for x in self._zf.namelist() if not x.endswith('/')] self._lock = threading.Lock() def __len__(self): return len(self._paths) def __getstate__(self): d = self.__dict__.copy() d['_zf'] = None d['_lock'] = None return d def __setstate__(self, state): self.__dict__ = state self._lock = threading.Lock() def get_example(self, i_or_filename): # LabeledZippedImageDataset needs file with filename in zip archive if isinstance(i_or_filename, six.integer_types): zfn = self._paths[i_or_filename] else: zfn = i_or_filename # PIL may seek() on the file -- zipfile won't support it with self._lock: if self._zf is None or self._zf_pid != os.getpid(): self._zf_pid = os.getpid() self._zf = zipfile.ZipFile(self._zipfilename) image_file_mem = self._zf.read(zfn) image_file = io.BytesIO(image_file_mem) image = _read_image_as_array(image_file, self._dtype) return _postprocess_image(image) def _check_pillow_availability(): if not available: raise ImportError('PIL cannot be loaded. Install Pillow!\n' 'The actual import error is as follows:\n' + str(_import_error))
11,769
36.845659
79
py
chainer
chainer-master/chainer/datasets/fashion_mnist.py
import os import numpy import chainer from chainer.dataset import download from chainer.datasets._mnist_helper import make_npz from chainer.datasets._mnist_helper import preprocess_mnist _fashion_mnist_labels = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot'] def get_fashion_mnist_labels(): """Provide a list of the string value names of the labels. Returns: List of string values of the image labels. """ return list(_fashion_mnist_labels) def get_fashion_mnist(withlabel=True, ndim=1, scale=1., dtype=None, label_dtype=numpy.int32, rgb_format=False): """Gets the Fashion-MNIST dataset. `Fashion-MNIST <https://github.com/zalandoresearch/fashion-mnist/>`_ is a set of fashion articles represented by grey-scale 28x28 images. In the original images, each pixel is represented by one-byte unsigned integer. This function scales the pixels to floating point values in the interval ``[0, scale]``. This function returns the training set and the test set of the official Fashion-MNIST dataset. If ``withlabel`` is ``True``, each dataset consists of tuples of images and labels, otherwise it only consists of images. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. ndim (int): Number of dimensions of each image. The shape of each image is determined depending on ``ndim`` as follows: - ``ndim == 1``: the shape is ``(784,)`` - ``ndim == 2``: the shape is ``(28, 28)`` - ``ndim == 3``: the shape is ``(1, 28, 28)`` scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. rgb_format (bool): if ``ndim == 3`` and ``rgb_format`` is ``True``, the image will be converted to rgb format by duplicating the channels so the image shape is (3, 28, 28). Default is ``False``. Returns: A tuple of two datasets. If ``withlabel`` is ``True``, both datasets are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ train_raw = _retrieve_fashion_mnist_training() dtype = chainer.get_dtype(dtype) train = preprocess_mnist(train_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) test_raw = _retrieve_fashion_mnist_test() test = preprocess_mnist(test_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) return train, test def _retrieve_fashion_mnist_training(): base_url = 'http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/' urls = [base_url + 'train-images-idx3-ubyte.gz', base_url + 'train-labels-idx1-ubyte.gz'] return _retrieve_fashion_mnist('train.npz', urls) def _retrieve_fashion_mnist_test(): base_url = 'http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/' urls = [base_url + 't10k-images-idx3-ubyte.gz', base_url + 't10k-labels-idx1-ubyte.gz'] return _retrieve_fashion_mnist('test.npz', urls) def _retrieve_fashion_mnist(name, urls): root = download.get_dataset_directory('pfnet/chainer/fashion-mnist') path = os.path.join(root, name) return download.cache_or_load_file( path, lambda path: make_npz(path, urls), numpy.load)
3,771
38.705263
79
py
chainer
chainer-master/chainer/datasets/tuple_dataset.py
import six class TupleDataset(object): """Dataset of tuples from multiple equal-length datasets. A ``TupleDataset`` combines multiple equal-length datasets into a single dataset of tuples. The ``i``-th tuple contains the ``i``-th example from each of the argument datasets, in the same order that they were supplied. Recall that in Chainer, a dataset is defined as an iterable that supports both ``__getitem__`` and ``__len__``. The ``__getitem__`` method should support indexing by both an integer and a slice. As an example, consider creating a ``TupleDataset`` from two argument datasets ``d1 = [8, 0, 5, 1]`` and ``d2 = [3, 1, 7, 4]`` as ``tuple_dataset = TupleDataset(d1, d2)``. The ``tuple_dataset`` will then contain the examples ``(8, 3), (0, 1), (5, 7), (1, 4)``. Note that this behavior is similar to that of the built-in :func:`zip` function. Args: datasets: Underlying datasets that will be aggregated. Each dataset must be an iterable that implements ``__getitem__`` and ``__len__``. The ``j``-th dataset will be used for the ``j``-th item of each example tuple. All datasets must have the same length. """ def __init__(self, *datasets): if not datasets: raise ValueError('no datasets are given') length = len(datasets[0]) for i, dataset in enumerate(datasets): if len(dataset) != length: raise ValueError( 'dataset of the index {} has a wrong length'.format(i)) self._datasets = datasets self._length = length def __getitem__(self, index): batches = [dataset[index] for dataset in self._datasets] if isinstance(index, slice): length = len(batches[0]) return [tuple([batch[i] for batch in batches]) for i in six.moves.range(length)] else: return tuple(batches) def __len__(self): return self._length
2,024
37.942308
79
py
chainer
chainer-master/chainer/datasets/dict_dataset.py
import six class DictDataset(object): """Dataset of a dictionary of datasets. It combines multiple datasets into one dataset. Each example is represented by a dictionary mapping a key to an example of the corresponding dataset. Args: datasets: Underlying datasets. The keys are used as the keys of each example. All datasets must have the same length. """ def __init__(self, **datasets): if not datasets: raise ValueError('no datasets are given') length = None for key, dataset in six.iteritems(datasets): if length is None: length = len(dataset) elif length != len(dataset): raise ValueError( 'dataset length conflicts at "{}"'.format(key)) self._datasets = datasets self._length = length def __getitem__(self, index): batches = {key: dataset[index] for key, dataset in six.iteritems(self._datasets)} if isinstance(index, slice): length = len(six.next(six.itervalues(batches))) return [{key: batch[i] for key, batch in six.iteritems(batches)} for i in six.moves.range(length)] else: return batches def __len__(self): return self._length
1,334
30.785714
79
py
chainer
chainer-master/chainer/datasets/transform_dataset.py
from chainer.dataset import dataset_mixin class TransformDataset(dataset_mixin.DatasetMixin): """Dataset that indexes the base dataset and transforms the data. This dataset wraps the base dataset by modifying the behavior of the base dataset's :meth:`__getitem__`. Arrays returned by :meth:`__getitem__` of the base dataset with an integer as an argument are transformed by the given function :obj:`transform`. Also, :meth:`__len__` returns the integer returned by the base dataset's :meth:`__len__`. The function :obj:`transform` takes, as an argument, :obj:`in_data`, which is the output of the base dataset's :meth:`__getitem__`, and returns the transformed arrays as output. Please see the following example. Since :obj:`in_data` directly refers to the item in the dataset, take care that :obj:`transform` not modify it. For example, note that the line `img = img - 0.5` bellow is correct since it makes a copy of `img`. However, it would be incorrect to use `img -= 0.5` since that would update the contents of the item in the dataset in place, corrupting it. >>> from chainer.datasets import get_mnist >>> from chainer.datasets import TransformDataset >>> dataset, _ = get_mnist() >>> def transform(in_data): ... img, label = in_data ... img = img - 0.5 # scale to [-0.5, 0.5] ... return img, label >>> dataset = TransformDataset(dataset, transform) Args: dataset: The underlying dataset. The index of this dataset corresponds to the index of the base dataset. This object needs to support functions :meth:`__getitem__` and :meth:`__len__` as described above. transform (callable): A function that is called to transform values returned by the underlying dataset's :meth:`__getitem__`. """ def __init__(self, dataset, transform): self._dataset = dataset self._transform = transform def __len__(self): return len(self._dataset) def get_example(self, i): in_data = self._dataset[i] return self._transform(in_data)
2,158
39.735849
78
py
chainer
chainer-master/chainer/datasets/__init__.py
# import classes and functions from chainer.datasets.cifar import get_cifar10 # NOQA from chainer.datasets.cifar import get_cifar100 # NOQA from chainer.datasets.concatenated_dataset import ConcatenatedDataset # NOQA from chainer.datasets.dict_dataset import DictDataset # NOQA from chainer.datasets.fashion_mnist import get_fashion_mnist # NOQA from chainer.datasets.fashion_mnist import get_fashion_mnist_labels # NOQA from chainer.datasets.image_dataset import ImageDataset # NOQA from chainer.datasets.image_dataset import LabeledImageDataset # NOQA from chainer.datasets.image_dataset import LabeledZippedImageDataset # NOQA from chainer.datasets.image_dataset import MultiZippedImageDataset # NOQA from chainer.datasets.image_dataset import ZippedImageDataset # NOQA from chainer.datasets.kuzushiji_mnist import get_kuzushiji_mnist # NOQA from chainer.datasets.kuzushiji_mnist import get_kuzushiji_mnist_labels # NOQA from chainer.datasets.mnist import get_mnist # NOQA from chainer.datasets.pickle_dataset import open_pickle_dataset # NOQA from chainer.datasets.pickle_dataset import open_pickle_dataset_writer # NOQA from chainer.datasets.pickle_dataset import PickleDataset # NOQA from chainer.datasets.pickle_dataset import PickleDatasetWriter # NOQA from chainer.datasets.ptb import get_ptb_words # NOQA from chainer.datasets.ptb import get_ptb_words_vocabulary # NOQA from chainer.datasets.sub_dataset import get_cross_validation_datasets # NOQA from chainer.datasets.sub_dataset import get_cross_validation_datasets_random # NOQA from chainer.datasets.sub_dataset import split_dataset # NOQA from chainer.datasets.sub_dataset import split_dataset_n # NOQA from chainer.datasets.sub_dataset import split_dataset_n_random # NOQA from chainer.datasets.sub_dataset import split_dataset_random # NOQA from chainer.datasets.sub_dataset import SubDataset # NOQA from chainer.datasets.svhn import get_svhn # NOQA from chainer.datasets.text_dataset import TextDataset # NOQA from chainer.datasets.transform_dataset import TransformDataset # NOQA from chainer.datasets.tuple_dataset import TupleDataset # NOQA
2,144
64
85
py
chainer
chainer-master/chainer/datasets/svhn.py
import os import numpy try: from scipy import io _scipy_available = True except Exception as e: _error = e _scipy_available = False import chainer from chainer.dataset import download from chainer.datasets import tuple_dataset def get_svhn(withlabel=True, scale=1., dtype=None, label_dtype=numpy.int32, add_extra=False): """Gets the SVHN dataset. `The Street View House Numbers (SVHN) dataset <http://ufldl.stanford.edu/housenumbers/>`_ is a dataset similar to MNIST but composed of cropped images of house numbers. The functionality of this function is identical to the counterpart for the MNIST dataset (:func:`~chainer.datasets.get_mnist`), with the exception that there is no ``ndim`` argument. .. note:: `SciPy <https://www.scipy.org/>`_ is required to use this feature. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. add_extra: Use extra training set. Returns: If ``add_extra`` is ``False``, a tuple of two datasets (train and test). Otherwise, a tuple of three datasets (train, test, and extra). If ``withlabel`` is ``True``, all datasets are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ if not _scipy_available: raise RuntimeError('SciPy is not available: %s' % _error) train_raw = _retrieve_svhn_training() dtype = chainer.get_dtype(dtype) train = _preprocess_svhn(train_raw, withlabel, scale, dtype, label_dtype) test_raw = _retrieve_svhn_test() test = _preprocess_svhn(test_raw, withlabel, scale, dtype, label_dtype) if add_extra: extra_raw = _retrieve_svhn_extra() extra = _preprocess_svhn(extra_raw, withlabel, scale, dtype, label_dtype) return train, test, extra else: return train, test def _preprocess_svhn(raw, withlabel, scale, image_dtype, label_dtype): images = raw['x'].transpose(3, 2, 0, 1) images = images.astype(image_dtype) images *= scale / 255. labels = raw['y'].astype(label_dtype).flatten() # labels go from 1-10, with the digit "0" having label 10. # Set "0" to be label 0 to restore expected ordering labels[labels == 10] = 0 if withlabel: return tuple_dataset.TupleDataset(images, labels) else: return images def _retrieve_svhn_training(): url = 'http://ufldl.stanford.edu/housenumbers/train_32x32.mat' return _retrieve_svhn('train.npz', url) def _retrieve_svhn_test(): url = 'http://ufldl.stanford.edu/housenumbers/test_32x32.mat' return _retrieve_svhn('test.npz', url) def _retrieve_svhn_extra(): url = 'http://ufldl.stanford.edu/housenumbers/extra_32x32.mat' return _retrieve_svhn('extra.npz', url) def _retrieve_svhn(name, url): root = download.get_dataset_directory('pfnet/chainer/svhn') path = os.path.join(root, name) return download.cache_or_load_file( path, lambda path: _make_npz(path, url), numpy.load) def _make_npz(path, url): _path = download.cached_download(url) raw = io.loadmat(_path) images = raw['X'].astype(numpy.uint8) labels = raw['y'].astype(numpy.uint8) numpy.savez_compressed(path, x=images, y=labels) return {'x': images, 'y': labels}
3,857
32.258621
79
py
chainer
chainer-master/chainer/datasets/_mnist_helper.py
import gzip import struct import numpy import six from chainer.dataset import download from chainer.datasets import tuple_dataset def make_npz(path, urls): x_url, y_url = urls x_path = download.cached_download(x_url) y_path = download.cached_download(y_url) with gzip.open(x_path, 'rb') as fx, gzip.open(y_path, 'rb') as fy: fx.read(4) fy.read(4) N, = struct.unpack('>i', fx.read(4)) if N != struct.unpack('>i', fy.read(4))[0]: raise RuntimeError('wrong pair of MNIST images and labels') fx.read(8) x = numpy.empty((N, 784), dtype=numpy.uint8) y = numpy.empty(N, dtype=numpy.uint8) for i in six.moves.range(N): y[i] = ord(fy.read(1)) for j in six.moves.range(784): x[i, j] = ord(fx.read(1)) numpy.savez_compressed(path, x=x, y=y) return {'x': x, 'y': y} def preprocess_mnist(raw, withlabel, ndim, scale, image_dtype, label_dtype, rgb_format): images = raw['x'] if ndim == 2: images = images.reshape(-1, 28, 28) elif ndim == 3: images = images.reshape(-1, 1, 28, 28) if rgb_format: images = numpy.broadcast_to(images, (len(images), 3) + images.shape[2:]) elif ndim != 1: raise ValueError('invalid ndim for MNIST dataset') images = images.astype(image_dtype) images *= scale / 255. if withlabel: labels = raw['y'].astype(label_dtype) return tuple_dataset.TupleDataset(images, labels) else: return images
1,614
27.839286
76
py
chainer
chainer-master/chainer/datasets/cifar.py
import os import sys import tarfile import numpy import six.moves.cPickle as pickle import chainer from chainer.dataset import download from chainer.datasets import tuple_dataset def get_cifar10(withlabel=True, ndim=3, scale=1., dtype=None): """Gets the CIFAR-10 dataset. `CIFAR-10 <https://www.cs.toronto.edu/~kriz/cifar.html>`_ is a set of small natural images. Each example is an RGB color image of size 32x32, classified into 10 groups. In the original images, each component of pixels is represented by one-byte unsigned integer. This function scales the components to floating point values in the interval ``[0, scale]``. This function returns the training set and the test set of the official CIFAR-10 dataset. If ``withlabel`` is ``True``, each dataset consists of tuples of images and labels, otherwise it only consists of images. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. ndim (int): Number of dimensions of each image. The shape of each image is determined depending on ndim as follows: - ``ndim == 1``: the shape is ``(3072,)`` - ``ndim == 3``: the shape is ``(3, 32, 32)`` scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). Returns: A tuple of two datasets. If ``withlabel`` is ``True``, both datasets are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ return _get_cifar('cifar-10', withlabel, ndim, scale, dtype) def get_cifar100(withlabel=True, ndim=3, scale=1., dtype=None): """Gets the CIFAR-100 dataset. `CIFAR-100 <https://www.cs.toronto.edu/~kriz/cifar.html>`_ is a set of small natural images. Each example is an RGB color image of size 32x32, classified into 100 groups. In the original images, each component pixels is represented by one-byte unsigned integer. This function scales the components to floating point values in the interval ``[0, scale]``. This function returns the training set and the test set of the official CIFAR-100 dataset. If ``withlabel`` is ``True``, each dataset consists of tuples of images and labels, otherwise it only consists of images. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. ndim (int): Number of dimensions of each image. The shape of each image is determined depending on ndim as follows: - ``ndim == 1``: the shape is ``(3072,)`` - ``ndim == 3``: the shape is ``(3, 32, 32)`` scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). Returns: A tuple of two datasets. If ``withlabel`` is ``True``, both are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ return _get_cifar('cifar-100', withlabel, ndim, scale, dtype) def _get_cifar(name, withlabel, ndim, scale, dtype): root = download.get_dataset_directory(os.path.join('pfnet', 'chainer', 'cifar')) npz_path = os.path.join(root, '{}.npz'.format(name)) url = 'https://www.cs.toronto.edu/~kriz/{}-python.tar.gz'.format(name) def creator(path): archive_path = download.cached_download(url) if name == 'cifar-10': train_x = numpy.empty((5, 10000, 3072), dtype=numpy.uint8) train_y = numpy.empty((5, 10000), dtype=numpy.uint8) test_y = numpy.empty(10000, dtype=numpy.uint8) dir_name = '{}-batches-py'.format(name) with tarfile.open(archive_path, 'r:gz') as archive: # training set for i in range(5): file_name = '{}/data_batch_{}'.format(dir_name, i + 1) d = _pickle_load(archive.extractfile(file_name)) train_x[i] = d['data'] train_y[i] = d['labels'] # test set file_name = '{}/test_batch'.format(dir_name) d = _pickle_load(archive.extractfile(file_name)) test_x = d['data'] test_y[...] = d['labels'] # copy to array train_x = train_x.reshape(50000, 3072) train_y = train_y.reshape(50000) else: # name == 'cifar-100' def load(archive, file_name): d = _pickle_load(archive.extractfile(file_name)) x = d['data'].reshape((-1, 3072)) y = numpy.array(d['fine_labels'], dtype=numpy.uint8) return x, y with tarfile.open(archive_path, 'r:gz') as archive: train_x, train_y = load(archive, 'cifar-100-python/train') test_x, test_y = load(archive, 'cifar-100-python/test') numpy.savez_compressed(path, train_x=train_x, train_y=train_y, test_x=test_x, test_y=test_y) return {'train_x': train_x, 'train_y': train_y, 'test_x': test_x, 'test_y': test_y} raw = download.cache_or_load_file(npz_path, creator, numpy.load) train = _preprocess_cifar(raw['train_x'], raw['train_y'], withlabel, ndim, scale, dtype) test = _preprocess_cifar(raw['test_x'], raw['test_y'], withlabel, ndim, scale, dtype) return train, test def _preprocess_cifar(images, labels, withlabel, ndim, scale, dtype): if ndim == 1: images = images.reshape(-1, 3072) elif ndim == 3: images = images.reshape(-1, 3, 32, 32) else: raise ValueError('invalid ndim for CIFAR dataset') dtype = chainer.get_dtype(dtype) images = images.astype(dtype) images *= scale / 255. if withlabel: labels = labels.astype(numpy.int32) return tuple_dataset.TupleDataset(images, labels) else: return images def _pickle_load(f): if sys.version_info > (3, ): # python3 return pickle.load(f, encoding='latin-1') else: # python2 return pickle.load(f)
6,772
39.076923
79
py
chainer
chainer-master/chainer/datasets/ptb.py
import os import numpy from chainer.dataset import download def get_ptb_words(): """Gets the Penn Tree Bank dataset as long word sequences. `Penn Tree Bank <https://catalog.ldc.upenn.edu/LDC99T42>`_ is originally a corpus of English sentences with linguistic structure annotations. This function uses a variant distributed at `https://github.com/wojzaremba/lstm <https://github.com/wojzaremba/lstm>`_, which omits the annotation and splits the dataset into three parts: training, validation, and test. This function returns the training, validation, and test sets, each of which is represented as a long array of word IDs. All sentences in the dataset are concatenated by End-of-Sentence mark '<eos>', which is treated as one of the vocabulary. Returns: tuple of numpy.ndarray: Int32 vectors of word IDs. .. Seealso:: Use :func:`get_ptb_words_vocabulary` to get the mapping between the words and word IDs. """ train = _retrieve_ptb_words('train.npz', _train_url) valid = _retrieve_ptb_words('valid.npz', _valid_url) test = _retrieve_ptb_words('test.npz', _test_url) return train, valid, test def get_ptb_words_vocabulary(): """Gets the Penn Tree Bank word vocabulary. Returns: dict: Dictionary that maps words to corresponding word IDs. The IDs are used in the Penn Tree Bank long sequence datasets. .. seealso:: See :func:`get_ptb_words` for the actual datasets. """ return _retrieve_word_vocabulary() _train_url = 'https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.train.txt' # NOQA _valid_url = 'https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.valid.txt' # NOQA _test_url = 'https://raw.githubusercontent.com/wojzaremba/lstm/master/data/ptb.test.txt' # NOQA def _retrieve_ptb_words(name, url): def creator(path): vocab = _retrieve_word_vocabulary() words = _load_words(url) x = numpy.empty(len(words), dtype=numpy.int32) for i, word in enumerate(words): x[i] = vocab[word] numpy.savez_compressed(path, x=x) return {'x': x} root = download.get_dataset_directory('pfnet/chainer/ptb') path = os.path.join(root, name) loaded = download.cache_or_load_file(path, creator, numpy.load) return loaded['x'] def _retrieve_word_vocabulary(): def creator(path): words = _load_words(_train_url) vocab = {} index = 0 with open(path, 'w') as f: for word in words: if word not in vocab: vocab[word] = index index += 1 f.write(word + '\n') return vocab def loader(path): vocab = {} with open(path) as f: for i, word in enumerate(f): vocab[word.strip()] = i return vocab root = download.get_dataset_directory('pfnet/chainer/ptb') path = os.path.join(root, 'vocab.txt') return download.cache_or_load_file(path, creator, loader) def _load_words(url): path = download.cached_download(url) words = [] with open(path) as words_file: for line in words_file: if line: words += line.strip().split() words.append('<eos>') return words
3,359
30.111111
98
py
chainer
chainer-master/chainer/datasets/mnist.py
import os import numpy import chainer from chainer.dataset import download from chainer.datasets._mnist_helper import make_npz from chainer.datasets._mnist_helper import preprocess_mnist def get_mnist(withlabel=True, ndim=1, scale=1., dtype=None, label_dtype=numpy.int32, rgb_format=False): """Gets the MNIST dataset. `MNIST <http://yann.lecun.com/exdb/mnist/>`_ is a set of hand-written digits represented by grey-scale 28x28 images. In the original images, each pixel is represented by one-byte unsigned integer. This function scales the pixels to floating point values in the interval ``[0, scale]``. This function returns the training set and the test set of the official MNIST dataset. If ``withlabel`` is ``True``, each dataset consists of tuples of images and labels, otherwise it only consists of images. Args: withlabel (bool): If ``True``, it returns datasets with labels. In this case, each example is a tuple of an image and a label. Otherwise, the datasets only contain images. ndim (int): Number of dimensions of each image. The shape of each image is determined depending on ``ndim`` as follows: - ``ndim == 1``: the shape is ``(784,)`` - ``ndim == 2``: the shape is ``(28, 28)`` - ``ndim == 3``: the shape is ``(1, 28, 28)`` scale (float): Pixel value scale. If it is 1 (default), pixels are scaled to the interval ``[0, 1]``. dtype: Data type of resulting image arrays. ``chainer.config.dtype`` is used by default (see :ref:`configuration`). label_dtype: Data type of the labels. rgb_format (bool): if ``ndim == 3`` and ``rgb_format`` is ``True``, the image will be converted to rgb format by duplicating the channels so the image shape is (3, 28, 28). Default is ``False``. Returns: A tuple of two datasets. If ``withlabel`` is ``True``, both datasets are :class:`~chainer.datasets.TupleDataset` instances. Otherwise, both datasets are arrays of images. """ dtype = chainer.get_dtype(dtype) train_raw = _retrieve_mnist_training() train = preprocess_mnist(train_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) test_raw = _retrieve_mnist_test() test = preprocess_mnist(test_raw, withlabel, ndim, scale, dtype, label_dtype, rgb_format) return train, test def _retrieve_mnist_training(): urls = ['http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz', 'http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz'] return _retrieve_mnist('train.npz', urls) def _retrieve_mnist_test(): urls = ['http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz', 'http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz'] return _retrieve_mnist('test.npz', urls) def _retrieve_mnist(name, urls): root = download.get_dataset_directory('pfnet/chainer/mnist') path = os.path.join(root, name) return download.cache_or_load_file( path, lambda path: make_npz(path, urls), numpy.load)
3,215
40.766234
79
py
chainer
chainer-master/chainer/datasets/pickle_dataset.py
import io import multiprocessing.util import threading import six import six.moves.cPickle as pickle from chainer.dataset import dataset_mixin class PickleDatasetWriter(object): """Writer class that makes PickleDataset. To make :class:`PickleDataset`, a user needs to prepare data using :class:`PickleDatasetWriter`. Args: writer: File like object that supports ``write`` and ``tell`` methods. protocol (int): Valid protocol for :mod:`pickle`. .. seealso: chainer.datasets.PickleDataset """ def __init__(self, writer, protocol=pickle.HIGHEST_PROTOCOL): self._positions = [] self._writer = writer self._protocol = protocol def close(self): self._writer.close() def __enter__(self): return self def __exit__(self, exc_type, exc_value, traceback): self.close() def write(self, x): position = self._writer.tell() pickle.dump(x, self._writer, protocol=self._protocol) self._positions.append(position) def flush(self): if hasattr(self._writer, 'flush'): self._writer.flush() class PickleDataset(dataset_mixin.DatasetMixin): """Dataset stored in a storage using pickle. :mod:`pickle` is the default serialization library of Python. This dataset stores any objects in a storage using :mod:`pickle`. Even when a user wants to use a large dataset, this dataset can stores all data in a large storage like HDD and each data can be randomly accessible. .. testsetup:: import tempfile fs, path_to_data = tempfile.mkstemp() >>> with chainer.datasets.open_pickle_dataset_writer(path_to_data) as w: ... w.write((1, 2.0, 'hello')) ... w.write((2, 3.0, 'good-bye')) ... >>> with chainer.datasets.open_pickle_dataset(path_to_data) as dataset: ... print(dataset[1]) ... (2, 3.0, 'good-bye') .. testcleanup:: import os os.close(fs) Args: reader: File like object. `reader` must support random access. """ def __init__(self, reader): # Only py3 supports `seekable` method if six.PY3 and not reader.seekable(): raise ValueError('reader must support random access') self._reader = reader self._positions = [] reader.seek(0) while True: position = reader.tell() try: pickle.load(reader) except EOFError: break self._positions.append(position) self._lock = threading.RLock() self._register_hook() def _register_hook(self): # TODO: Avoid using undocumented feature multiprocessing.util.register_after_fork( self, PickleDataset._after_fork) def _after_fork(self): if callable(getattr(self._reader, 'after_fork', None)): self._reader.after_fork() def __getstate__(self): state = self.__dict__.copy() del state['_lock'] return state def __setstate__(self, state): self.__dict__.update(state) self._lock = threading.RLock() self._register_hook() def close(self): """Closes a file reader. After a user calls this method, the dataset will no longer be accessible.. """ with self._lock: self._reader.close() def __enter__(self): return self def __exit__(self, exc_type, exc_value, traceback): self.close() def __len__(self): return len(self._positions) def get_example(self, index): with self._lock: self._reader.seek(self._positions[index]) return pickle.load(self._reader) class _FileReader(io.RawIOBase): """A file-like class implemented `after_fork()` hook The method :meth:`after_fork` is called in the child process after forking, and it closes and reopens the file object to avoid race condition caused by open file description. See: https://www.securecoding.cert.org/confluence/x/ZQG7AQ """ def __init__(self, path): super(_FileReader, self).__init__() self._path = path self._fp = None self._open() def _open(self): self._fp = open(self._path, 'rb') def after_fork(self): """Reopens the file to avoid race condition.""" self.close() self._open() def __getstate__(self): state = self.__dict__.copy() del state['_fp'] return state def __setstate__(self, state): self.__dict__.update(state) self._open() # file-like interface def flush(self): self._fp.flush() def close(self): self._fp.close() def fileno(self): return self._fp.fileno() def seekable(self): return self._fp.seekable() def seek(self, offset, whence=io.SEEK_SET): return self._fp.seek(offset, whence) def tell(self): return self._fp.tell() def readinto(self, b): return self._fp.readinto(b) def open_pickle_dataset(path): """Opens a dataset stored in a given path. This is a helper function to open :class:`PickleDataset`. It opens a given file in binary mode, and creates a :class:`PickleDataset` instance. This method does not close the opened file. A user needs to call :func:`PickleDataset.close` or use `with`: .. code-block:: python with chainer.datasets.open_pickle_dataset('path') as dataset: pass # use dataset Args: path (str): Path to a dataset. Returns: chainer.datasets.PickleDataset: Opened dataset. .. seealso: chainer.datasets.PickleDataset """ reader = _FileReader(path) return PickleDataset(reader) def open_pickle_dataset_writer(path, protocol=pickle.HIGHEST_PROTOCOL): """Opens a writer to make a PickleDataset. This is a helper function to open :class:`PickleDatasetWriter`. It opens a given file in binary mode and creates a :class:`PickleDatasetWriter` instance. This method does not close the opened file. A user needs to call :func:`PickleDatasetWriter.close` or use `with`: .. code-block:: python with chainer.datasets.open_pickle_dataset_writer('path') as writer: pass # use writer Args: path (str): Path to a dataset. protocol (int): Valid protocol for :mod:`pickle`. Returns: chainer.datasets.PickleDatasetWriter: Opened writer. .. seealso: chainer.datasets.PickleDataset """ writer = open(path, 'wb') return PickleDatasetWriter(writer, protocol=protocol)
6,689
25.338583
79
py
chainer
chainer-master/chainer/datasets/sub_dataset.py
import numpy import six import warnings from chainer.dataset import dataset_mixin class SubDataset(dataset_mixin.DatasetMixin): """Subset of a base dataset. SubDataset defines a subset of a given base dataset. The subset is defined as an interval of indexes, optionally with a given permutation. If ``order`` is given, then the ``i``-th example of this dataset is the ``order[start + i]``-th example of the base dataset, where ``i`` is a non-negative integer. If ``order`` is not given, then the ``i``-th example of this dataset is the ``start + i``-th example of the base dataset. Negative indexing is also allowed: in this case, the term ``start + i`` is replaced by ``finish + i``. SubDataset is often used to split a dataset into training and validation subsets. The training set is used for training, while the validation set is used to track the generalization performance, i.e. how the learned model works well on unseen data. We can tune hyperparameters (e.g. number of hidden units, weight initializers, learning rate, etc.) by comparing the validation performance. Note that we often use another set called test set to measure the quality of the tuned hyperparameter, which can be made by nesting multiple SubDatasets. There are two ways to make training-validation splits. One is a single split, where the dataset is split just into two subsets. It can be done by :func:`split_dataset` or :func:`split_dataset_random`. The other one is a :math:`k`-fold cross validation, in which the dataset is divided into :math:`k` subsets, and :math:`k` different splits are generated using each of the :math:`k` subsets as a validation set and the rest as a training set. It can be done by :func:`get_cross_validation_datasets`. Args: dataset: Base dataset. start (int): The first index in the interval. finish (int): The next-to-the-last index in the interval. order (sequence of ints): Permutation of indexes in the base dataset. If this is ``None``, then the ascending order of indexes is used. """ def __init__(self, dataset, start, finish, order=None): if start < 0 or finish > len(dataset): raise ValueError('subset overruns the base dataset.') self._dataset = dataset self._start = start self._finish = finish self._size = finish - start if order is not None and len(order) != len(dataset): msg = ('order option must have the same length as the base ' 'dataset: len(order) = {} while len(dataset) = {}'.format( len(order), len(dataset))) raise ValueError(msg) self._order = order def __len__(self): return self._size def get_example(self, i): if i >= 0: if i >= self._size: raise IndexError('dataset index out of range') index = self._start + i else: if i < -self._size: raise IndexError('dataset index out of range') index = self._finish + i if self._order is not None: index = self._order[index] return self._dataset[index] def split_dataset(dataset, split_at, order=None): """Splits a dataset into two subsets. This function creates two instances of :class:`SubDataset`. These instances do not share any examples, and they together cover all examples of the original dataset. Args: dataset: Dataset to split. split_at (int): Position at which the base dataset is split. order (sequence of ints): Permutation of indexes in the base dataset. See the documentation of :class:`SubDataset` for details. Returns: tuple: Two :class:`SubDataset` objects. The first subset represents the examples of indexes ``order[:split_at]`` while the second subset represents the examples of indexes ``order[split_at:]``. """ n_examples = len(dataset) if not isinstance(split_at, (six.integer_types, numpy.integer)): raise TypeError('split_at must be int, got {} instead' .format(type(split_at))) if split_at < 0: raise ValueError('split_at must be non-negative') if split_at > n_examples: raise ValueError('split_at exceeds the dataset size') subset1 = SubDataset(dataset, 0, split_at, order) subset2 = SubDataset(dataset, split_at, n_examples, order) return subset1, subset2 def split_dataset_random(dataset, first_size, seed=None): """Splits a dataset into two subsets randomly. This function creates two instances of :class:`SubDataset`. These instances do not share any examples, and they together cover all examples of the original dataset. The split is automatically done randomly. Args: dataset: Dataset to split. first_size (int): Size of the first subset. seed (int): Seed the generator used for the permutation of indexes. If an integer being convertible to 32 bit unsigned integers is specified, it is guaranteed that each sample in the given dataset always belongs to a specific subset. If ``None``, the permutation is changed randomly. Returns: tuple: Two :class:`SubDataset` objects. The first subset contains ``first_size`` examples randomly chosen from the dataset without replacement, and the second subset contains the rest of the dataset. """ order = numpy.random.RandomState(seed).permutation(len(dataset)) return split_dataset(dataset, first_size, order) def split_dataset_n(dataset, n, order=None): """Splits a dataset into ``n`` subsets. Args: dataset: Dataset to split. n(int): The number of subsets. order (sequence of ints): Permutation of indexes in the base dataset. See the documentation of :class:`SubDataset` for details. Returns: list: List of ``n`` :class:`SubDataset` objects. Each subset contains the examples of indexes ``order[i * (len(dataset) // n):(i + 1) * (len(dataset) // n)]`` . """ n_examples = len(dataset) sub_size = n_examples // n return [SubDataset(dataset, sub_size * i, sub_size * (i + 1), order) for i in six.moves.range(n)] def split_dataset_n_random(dataset, n, seed=None): """Splits a dataset into ``n`` subsets randomly. Args: dataset: Dataset to split. n(int): The number of subsets. seed (int): Seed the generator used for the permutation of indexes. If an integer being convertible to 32 bit unsigned integers is specified, it is guaranteed that each sample in the given dataset always belongs to a specific subset. If ``None``, the permutation is changed randomly. Returns: list: List of ``n`` :class:`SubDataset` objects. Each subset contains ``len(dataset) // n`` examples randomly chosen from the dataset without replacement. """ n_examples = len(dataset) sub_size = n_examples // n order = numpy.random.RandomState(seed).permutation(len(dataset)) return [SubDataset(dataset, sub_size * i, sub_size * (i + 1), order) for i in six.moves.range(n)] def get_cross_validation_datasets(dataset, n_folds=None, order=None, **kwargs): """Creates a set of training/test splits for cross validation. This function generates ``n_folds`` splits of the given dataset. The first part of each split corresponds to the training dataset, while the second part to the test dataset. No pairs of test datasets share any examples, and all test datasets together cover the whole base dataset. Each test dataset contains almost same number of examples (the numbers may differ up to 1). Args: dataset: Dataset to split. n_fold(int): *(deprecated)* `n_fold` is now deprecated for consistency of naming choice. Please use `n_folds` instead. n_folds (int): Number of splits for cross validation. order (sequence of ints): Order of indexes with which each split is determined. If it is ``None``, then no permutation is used. Returns: list of tuples: List of dataset splits. """ if 'n_fold' in kwargs: warnings.warn( 'Argument `n_fold` is deprecated. ' 'Please use `n_folds` instead', DeprecationWarning) n_folds = kwargs['n_fold'] if order is None: order = numpy.arange(len(dataset)) else: order = numpy.array(order) # copy whole_size = len(dataset) borders = [whole_size * i // n_folds for i in six.moves.range(n_folds + 1)] test_sizes = [borders[i + 1] - borders[i] for i in six.moves.range(n_folds)] splits = [] for test_size in reversed(test_sizes): size = whole_size - test_size splits.append(split_dataset(dataset, size, order)) new_order = numpy.empty_like(order) new_order[:test_size] = order[-test_size:] new_order[test_size:] = order[:-test_size] order = new_order return splits def get_cross_validation_datasets_random(dataset, n_folds, seed=None, **kwargs): """Creates a set of training/test splits for cross validation randomly. This function acts almost same as :func:`get_cross_validation_dataset`, except automatically generating random permutation. Args: dataset: Dataset to split. n_fold (int): *(deprecated)* `n_fold` is now deprecated for consistency of naming choice. Please use `n_folds` instead. n_folds (int): Number of splits for cross validation. seed (int): Seed the generator used for the permutation of indexes. If an integer beging convertible to 32 bit unsigned integers is specified, it is guaranteed that each sample in the given dataset always belongs to a specific subset. If ``None``, the permutation is changed randomly. Returns: list of tuples: List of dataset splits. """ if 'n_fold' in kwargs: warnings.warn( 'Argument `n_fold` is deprecated. ' 'Please use `n_folds` instead', DeprecationWarning) n_folds = kwargs['n_fold'] order = numpy.random.RandomState(seed).permutation(len(dataset)) return get_cross_validation_datasets(dataset, n_folds, order)
10,676
38.988764
79
py
chainer
chainer-master/chainer/exporters/__init__.py
from chainer.exporters import caffe # NOQA
44
21.5
43
py
chainer
chainer-master/chainer/exporters/caffe.py
import collections import heapq import os import numpy import six import chainer from chainer import function from chainer import function_node from chainer.links.caffe.protobuf3 import caffe_pb2 as caffe_pb from chainer import variable _function_types = (function.Function, function_node.FunctionNode) def _add_blob(layer, shape, data): # The following part is ridiculously slow!! # TODO(okuta): Replace with C++ extension call blob = layer.blobs.add() blob.shape.dim[:] = shape blob.data[:] = data.flatten() def _dump_graph(outputs): fan_out = collections.defaultdict(int) cand_funcs = [] def add_cand_to_check(cands): for cand in cands: x = cand.creator if x is None: continue if x not in fan_out: # `len(fan_out)` is in order to avoid comparing `x` heapq.heappush(cand_funcs, (-x.rank, len(fan_out), x)) fan_out[x] += 1 add_cand_to_check(outputs) while cand_funcs: _, _, func = heapq.heappop(cand_funcs) assert isinstance(func, _function_types) add_cand_to_check(func.inputs) ret = [] cand_funcs = [] seen_set = set() def add_cand(cands): cands = [cand.creator for cand in cands if cand.creator is not None] for x in cands: if x in seen_set: continue order = 1 if fan_out[x] == 1 and len(cands) == 1: order = -len(seen_set) # Negate since heapq is min-heap # `len(seen_set)` is in order to avoid comparing `x` heapq.heappush(cand_funcs, (order, -x.rank, -len(seen_set), x)) seen_set.add(x) add_cand(outputs) while cand_funcs: _, _, _, func = heapq.heappop(cand_funcs) ret.append(func) add_cand(func.inputs) return ret[::-1] class _RetrieveAsCaffeModel(object): debug = False def __init__(self, prototxt, caffemodel=None): self.caffemodel = caffemodel self.prototxt = prototxt # key:string, val:dict(key: func, val: index) self.naming_map = collections.defaultdict(dict) def _get_layer_name(self, layer): """Generate layer name like "Convolution2DFunction-10-2". The first number means rank of the layer (depth from the top), and the second number is for preventing duplication (different layer objects can have same rank) Args: layer (~chainer.Function_node): Function object Returns: str: A string to be used for the ``name`` field of the graph in the exported Caffe model. """ label = '{}-{}'.format(layer.label, layer.rank) d = self.naming_map[label] if layer not in d.keys(): d[layer] = len(d) + 1 return '{}-{}'.format(label, d[layer]) def _get_parent_name(self, parent_): if parent_ is None: return 'data' return self._get_layer_name(parent_) def _gen_layer_prototxt(self, layer_params, name='layer', depth=0, indent=2): if isinstance(layer_params, (dict, collections.OrderedDict)): s = name + ' {\n' indent_s = ' ' * ((depth + 1) * indent) for key, val in layer_params.items(): s += indent_s + \ self._gen_layer_prototxt(val, name=key, depth=depth + 1) s += ' ' * (depth * indent) s += '}\n' return s elif isinstance(layer_params, bool): return '{}: {}\n'.format(name, 'true' if layer_params else 'false') elif isinstance(layer_params, six.integer_types + (float,)): return '{}: {}\n'.format(name, layer_params) elif isinstance(layer_params, str): return '{}: "{}"\n'.format(name, layer_params) elif isinstance(layer_params, list): s = '' indent_s = ' ' * depth * indent for i, t in enumerate(layer_params): if i != 0: s += indent_s s += self._gen_layer_prototxt(t, name=name, depth=depth + 1) return s else: raise ValueError('Unsupported type: ' + str(type(layer_params))) def dump_function_object(self, func, prototxt, net): assert isinstance(func, _function_types) layer_name = self._get_layer_name(func) parent_layer_names = [self._get_parent_name(input_.creator) for input_ in func.inputs] params = collections.OrderedDict() params['type'] = None params['name'] = layer_name params['bottom'] = parent_layer_names params['top'] = [layer_name] layer = None if net is not None: layer = net.layer.add() if func.label == 'LinearFunction': if len(func.inputs) == 2: _, W = func.inputs b = None else: _, W, b = func.inputs n_out, n_in = W.shape inner_product_param = { 'num_output': n_out, 'bias_term': b is not None, } params['type'] = 'InnerProduct' params['inner_product_param'] = inner_product_param params['bottom'] = params['bottom'][:1] if net is not None: for k, v in six.iteritems(inner_product_param): setattr(layer.inner_product_param, k, v) _add_blob(layer, list(W.shape), W.data) if b is not None: b.retain_data() _add_blob(layer, list(b.shape), b.data) elif func.label in ('Convolution2DFunction', 'Deconvolution2DFunction'): if len(func.inputs) == 2: _, W = func.inputs b = None else: _, W, b = func.inputs n_out, n_in, kw, kh = W.shape convolution_param = { 'num_output': n_out, 'bias_term': b is not None, 'pad_w': func.pw, 'pad_h': func.ph, 'stride_w': func.sx, 'stride_h': func.sy, 'kernel_w': kw, 'kernel_h': kh, 'group': func.groups } params['bottom'] = params['bottom'][:1] if func.label == 'Convolution2DFunction': params['type'] = 'Convolution' else: params['type'] = 'Deconvolution' convolution_param['num_output'] = n_in params['convolution_param'] = convolution_param if net is not None: for k, v in six.iteritems(convolution_param): setattr(layer.convolution_param, k, v) _add_blob(layer, [n_out, n_in, kh, kw], W.data) if b is not None: b.retain_data() _add_blob(layer, [n_out], b.data) elif func.label == 'AveragePooling2D': kw = func.kw kh = func.kh pooling_param = { 'pool': 1, 'pad_w': func.pw, 'pad_h': func.ph, 'stride_w': func.sx, 'stride_h': func.sy, 'kernel_w': kw, 'kernel_h': kh, } params['type'] = 'Pooling' params['pooling_param'] = pooling_param if net is not None: for k, v in six.iteritems(pooling_param): setattr(layer.pooling_param, k, v) elif func.label == 'MaxPoolingND' and func.ndim == 2: kh, kw = func.ksize sy, sx = func.stride ph, pw = func.pad pooling_param = { 'pool': 0, 'pad_w': pw, 'pad_h': ph, 'stride_w': sx, 'stride_h': sy, 'kernel_w': kw, 'kernel_h': kh, } params['type'] = 'Pooling' params['pooling_param'] = pooling_param if net is not None: for k, v in six.iteritems(pooling_param): setattr(layer.pooling_param, k, v) elif func.label == 'LocalResponseNormalization': lrn_param = { 'norm_region': 0, # ACROSS_CHANNELS 'local_size': func.n, 'k': func.k, 'alpha': func.alpha * func.n, 'beta': func.beta, } params['type'] = 'LRN' params['lrn_param'] = lrn_param if net is not None: for k, v in six.iteritems(lrn_param): setattr(layer.lrn_param, k, v) elif func.label == 'FixedBatchNormalization': _, gamma, beta, mean, var = func.inputs batch_norm_param = {'use_global_stats': True, 'eps': func.eps} params['type'] = 'BatchNorm' params['bottom'] = params['bottom'][:1] params['batch_norm_param'] = batch_norm_param if net is not None: for k, v in six.iteritems(batch_norm_param): setattr(layer.batch_norm_param, k, v) _add_blob(layer, [mean.data.size], mean.data) _add_blob(layer, [var.data.size], var.data) _add_blob(layer, [1], numpy.ones((1,), dtype=numpy.float32)) if gamma.data is None and beta.data is None: pass else: bn_name = layer_name + '_bn' params['name'] = bn_name params['top'] = [bn_name] if prototxt is not None: prototxt.write(self._gen_layer_prototxt(params)) if net is not None: layer.name = params['name'] layer.type = params['type'] layer.bottom[:] = params['bottom'] layer.top[:] = params['top'] layer.phase = caffe_pb.TEST del params, layer params = collections.OrderedDict() params['type'] = 'Scale' params['name'] = layer_name params['bottom'] = [bn_name] params['top'] = [layer_name] if net is not None: layer = net.layer.add() beta.retain_data() bias_term = beta.data is not None scale_param = { 'axis': 1, 'bias_term': bias_term, } params['scale_param'] = scale_param if net is not None: for k, v in six.iteritems(scale_param): setattr(layer.scale_param, k, v) _add_blob(layer, [gamma.data.size], gamma.data) if bias_term: _add_blob(layer, [beta.data.size], beta.data) elif func.label == 'ReLU': params['type'] = 'ReLU' elif func.label == 'LeakyReLU': relu_param = {'negative_slope': func.slope} params['type'] = 'ReLU' params['relu_param'] = relu_param if net is not None: for k, v in six.iteritems(relu_param): setattr(layer.relu_param, k, v) elif func.label == 'Concat': axis = func.axis concat_param = {'axis': axis} params['type'] = 'Concat' params['concat_param'] = concat_param if net is not None: for k, v in six.iteritems(concat_param): setattr(layer.concat_param, k, v) elif func.label == 'Softmax': params['type'] = 'Softmax' elif func.label == 'Sigmoid': params['type'] = 'Sigmoid' elif func.label == 'Reshape': input_ = func.inputs[0] parent = input_.creator parent_layer_name = parent_layer_names[0] if 'Reshape' in parent_layer_name: grandparent = parent.inputs[0].creator parent_layer_name = self._get_parent_name(grandparent) reshape_param = {'shape': {'dim': list(func.shape)}} params['type'] = 'Reshape' params['bottom'] = [parent_layer_name] params['reshape_param'] = reshape_param if layer is not None: dim = reshape_param['shape']['dim'] layer.reshape_param.shape.dim[:] = dim elif func.label == '_ + _': params['type'] = 'Eltwise' else: raise Exception( 'Cannot convert, name={}, rank={}, label={}, inputs={}'.format( layer_name, func.rank, func.label, parent_layer_names)) if prototxt is not None: prototxt.write(self._gen_layer_prototxt(params)) if net is not None: layer.name = params['name'] layer.type = params['type'] layer.bottom[:] = params['bottom'] layer.top[:] = params['top'] layer.phase = caffe_pb.TEST def __call__(self, name, inputs, outputs): dumped_list = _dump_graph(outputs) f = None net = None if self.caffemodel is not None: net = caffe_pb.NetParameter() try: if self.prototxt is not None: f = open(self.prototxt, 'wt') f.write('name: "{}"\n'.format(name)) assert len(inputs) == 1 f.write('layer {\n' ' name: "data"\n' ' type: "Input"\n' ' top: "data"\n' ' input_param { shape: {') for i in inputs[0].shape: f.write(' dim: ' + str(i)) f.write(' } }\n' '}\n') for i in dumped_list: self.dump_function_object(i, f, net) finally: if f is not None: f.close() if net is not None: with open(self.caffemodel, 'wb') as f: f.write(net.SerializeToString()) if self.debug: import google.protobuf.text_format with open(self.caffemodel + '.txt', 'w') as f: f.write(google.protobuf.text_format.MessageToString(net)) def export(model, args, directory=None, export_params=True, graph_name='Graph'): """(Experimental) Export a computational graph as Caffe format. Args: model (~chainer.Chain): The model object you want to export in Caffe format. It should have :meth:`__call__` method because the second argument ``args`` is directly given to the model by the ``()`` accessor. args (list of ~chainer.Variable): The arguments which are given to the model directly. directory (str): The directory used for saving the resulting Caffe model. If None, nothing is saved to the disk. export_params (bool): If True, this function exports all the parameters included in the given model at the same time. If False, the exported Caffe model doesn't include any parameter values. graph_name (str): A string to be used for the ``name`` field of the graph in the exported Caffe model. .. note:: Currently, this function supports networks that created by following layer functions. - :func:`~chainer.functions.linear` - :func:`~chainer.functions.convolution_2d` - :func:`~chainer.functions.deconvolution_2d` - :func:`~chainer.functions.max_pooling_2d` - :func:`~chainer.functions.average_pooling_2d` - :func:`~chainer.functions.batch_normalization` - :func:`~chainer.functions.local_response_normalization` - :func:`~chainer.functions.relu` - :func:`~chainer.functions.leaky_relu` - :func:`~chainer.functions.concat` - :func:`~chainer.functions.softmax` - :func:`~chainer.functions.reshape` - :func:`~chainer.functions.add` This function can export at least following networks. - GoogLeNet - ResNet - VGG And, this function use testing (evaluation) mode. .. admonition:: Example >>> from chainer.exporters import caffe >>> >>> class Model(chainer.Chain): ... def __init__(self): ... super(Model, self).__init__() ... with self.init_scope(): ... self.l1 = L.Convolution2D(None, 1, 1, 1, 0) ... self.b2 = L.BatchNormalization(1) ... self.l3 = L.Linear(None, 1) ... ... def __call__(self, x): ... h = F.relu(self.l1(x)) ... h = self.b2(h) ... return self.l3(h) ... >>> x = chainer.Variable(np.zeros((1, 10, 10, 10), np.float32)) >>> caffe.export(Model(), [x], None, True, 'test') """ assert isinstance(args, (tuple, list)) if len(args) != 1: raise NotImplementedError() for i in args: assert isinstance(i, variable.Variable) with function.force_backprop_mode(), chainer.using_config('train', False): output = model(*args) if isinstance(output, variable.Variable): output = [output] assert isinstance(output, (tuple, list)) for i in output: assert isinstance(i, variable.Variable) prototxt = None caffemodel = None if directory is not None: prototxt = os.path.join(directory, 'chainer_model.prototxt') if export_params: caffemodel = os.path.join(directory, 'chainer_model.caffemodel') retriever = _RetrieveAsCaffeModel(prototxt, caffemodel) retriever(graph_name, args, output)
18,050
35.763747
79
py
chainer
chainer-master/chainer/optimizer_hooks/gradient_noise.py
import numpy import chainer from chainer import cuda def exponential_decay_noise(xp, shape, dtype, hook, opt): """Time-dependent annealed Gaussian noise function from the paper: `Adding Gradient Noise Improves Learning for Very Deep Networks <https://arxiv.org/pdf/1511.06807>`_. """ std = numpy.sqrt(hook.eta / numpy.power(1 + opt.t, 0.55)) return xp.random.normal(0, std, shape).astype(dtype) class GradientNoise(object): """Optimizer/UpdateRule hook function for adding gradient noise. This hook function simply adds noise generated by the ``noise_func`` to the gradient. By default it adds time-dependent annealed Gaussian noise to the gradient at every training step: .. math:: g_t \\leftarrow g_t + N(0, \\sigma_t^2) where .. math:: \\sigma_t^2 = \\frac{\\eta}{(1+t)^\\gamma} with :math:`\\eta` selected from {0.01, 0.3, 1.0} and :math:`\\gamma = 0.55`. Args: eta (float): Parameter that defines the scale of the noise. For the default noise function, it is recommended that it be either 0.01, 0.3 or 1.0. noise_func (function): Noise generating function which by default is given by `Adding Gradient Noise Improves Learning for Very Deep Networks <https://arxiv.org/pdf/1511.06807>`_. Attributes: ~optimizer_hooks.GradientNoise.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). ~optimizer_hooks.GradientNoise.call_for_each_param (bool): Specifies if this hook is called for each parameter (``True``) or only once (``False``) by an optimizer to which this hook is registered. This function does not expect users to switch the value from default one, which is `True`. .. versionadded:: 4.0.0 The *timing* parameter. """ name = 'GradientNoise' call_for_each_param = True timing = 'pre' def __init__(self, eta, noise_func=exponential_decay_noise): self.eta = eta self.noise_func = noise_func def __call__(self, rule, param): g = param.grad if g is None: return with chainer.using_device(param.device): xp = param.device.xp noise = self.noise_func(xp, g.shape, g.dtype, self, rule) if xp is cuda.cupy: kernel = cuda.elementwise( 'T noise', 'T g', 'g += noise', 'gradient_noise') kernel(noise, g) else: g += noise
2,840
33.228916
79
py
chainer
chainer-master/chainer/optimizer_hooks/weight_decay.py
import chainer from chainer import cuda class WeightDecay(object): """Optimizer/UpdateRule hook function for weight decay regularization. This hook function adds a scaled parameter to the corresponding gradient. It can be used as a regularization. Args: rate (float): Coefficient for the weight decay. Attributes: ~optimizer_hooks.WeightDecay.rate (float): Coefficient for the weight decay. ~optimizer_hooks.WeightDecay.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). ~optimizer_hooks.WeightDecay.call_for_each_param (bool): Specifies if this hook is called for each parameter (``True``) or only once (``False``) by an optimizer to which this hook is registered. This function does not expect users to switch the value from default one, which is `True`. .. versionadded:: 4.0.0 The *timing* parameter. """ name = 'WeightDecay' call_for_each_param = True timing = 'pre' def __init__(self, rate): self.rate = rate def __call__(self, rule, param): p, g = param.data, param.grad if p is None or g is None: return with chainer.using_device(param.device): rate = self.rate if param._loss_scale is not None: rate *= param._loss_scale if param.device.xp is cuda.cupy: kernel = cuda.elementwise( 'T p, T decay', 'T g', 'g += decay * p', 'weight_decay') kernel(p, rate, g) else: g += rate * p
1,897
34.148148
79
py
chainer
chainer-master/chainer/optimizer_hooks/lasso.py
import chainer from chainer import cuda class Lasso(object): """Optimizer/UpdateRule hook function for Lasso regularization. This hook function adds a scaled parameter to the sign of each weight. It can be used as a regularization. Args: rate (float): Coefficient for the weight decay. Attributes: ~optimizer_hooks.Lasso.rate (float): Coefficient for the weight decay. ~optimizer_hooks.Lasso.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). ~optimizer_hooks.Lasso.call_for_each_param (bool): Specifies if this hook is called for each parameter (``True``) or only once (``False``) by an optimizer to which this hook is registered. This function does not expect users to switch the value from default one, which is `True`. .. versionadded:: 4.0.0 The *timing* parameter. """ name = 'Lasso' call_for_each_param = True timing = 'pre' def __init__(self, rate): self.rate = rate def __call__(self, rule, param): p, g = param.data, param.grad if p is None or g is None: return with chainer.using_device(param.device): xp = param.device.xp sign = xp.sign(p) if xp is cuda.cupy: kernel = cuda.elementwise( 'T s, T decay', 'T g', 'g += decay * s', 'lasso') kernel(sign, self.rate, g) else: g += self.rate * sign
1,773
33.784314
79
py
chainer
chainer-master/chainer/optimizer_hooks/__init__.py
from chainer.optimizer_hooks.gradient_clipping import GradientClipping # NOQA from chainer.optimizer_hooks.gradient_hard_clipping import GradientHardClipping # NOQA from chainer.optimizer_hooks.gradient_lars import GradientLARS # NOQA from chainer.optimizer_hooks.gradient_noise import GradientNoise # NOQA from chainer.optimizer_hooks.lasso import Lasso # NOQA from chainer.optimizer_hooks.weight_decay import WeightDecay # NOQA
436
61.428571
87
py
chainer
chainer-master/chainer/optimizer_hooks/gradient_lars.py
import chainer from chainer import backend class GradientLARS(object): """Optimizer/UpdateRule hook function for layer wise adaptive rate scaling. See: `Large Batch Training of Convolutional Networks <https://arxiv.org/abs/1708.03888>`_. See: `Convergence Analysis of Gradient Descent Algorithms with Proportional Updates <https://arxiv.org/abs/1801.03137>`_. This hook function scales all gradient arrays to fit to the weight norm. In <https://arxiv.org/abs/1708.03888>, .. math:: v_{t+1} &= m * v_t + \\gamma * \\lambda * (\\nabla L(w_t) + \\beta w_t), \\\\ w_{t+1} &= w_{t} - v_{t+1}, where - :math:`\\gamma` : learning_rate - :math:`m` : momentum - :math:`\\beta` : weight_decay - :math:`\\eta` : lars_coeeficient - :math:`\\lambda`: local_lr \ :math:`=\\eta * \ \\frac{\\|w_t\\|}{\\|\\nabla L(w_t)\\| + \\beta * \\|w_t\\|}`. As :math:`lr` in chainer.optimizers.SGD or chainer.optimizers.MomentumSGD corresponds to :math:`\\gamma * \\eta`, we define :math:`clip\\_rate` as :math:`\\frac{\\|w_t\\|}{\\|\\nabla L(w_t)\\| + \\beta * \\|w_t\\|}` and reformulate the aforementioned formula as: :math:`v_{t+1} \ = m * v_t + lr * clip\\_rate * (\\nabla L(w_t) + \\beta w_t)` and implement in this way. So you do not set lars_coeeficient. Args: threashold (float): If weight norm is more than threshold, this function scales all gradient arrays to fit weight norm. (See <https://arxiv.org/abs/1801.03137>) weight_decay (float): Coefficient for the weight decay. eps (float): Small value for the numerical stability. (See <https://arxiv.org/abs/1801.03137>) Attributes: ~optimizer_hooks.GradientLARS.threashold (float): If weight norm is more than threshold, this function scales all gradient arrays to fit weight norm. (See <https://arxiv.org/abs/1801.03137>) ~optimizer_hooks.GradientLARS.weight_decay (float): Coefficient for the weight decay. ~optimizer_hooks.GradientLARS.eps (float): Small value for the numerical stability. (See <https://arxiv.org/abs/1801.03137>) ~optimizer_hooks.GradientLARS.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). ~optimizer_hooks.GradientLARS.call_for_each_param (bool): Specifies if this hook is called for each parameter (``True``) or only once (``False``) by an optimizer to which this hook is registered. This function does not expect users to switch the value from default one, which is `True`. """ name = 'GradientLARS' call_for_each_param = True timing = 'pre' def __init__(self, threshold=1e-2, weight_decay=0.0, eps=1e-9): self.threshold = threshold self.weight_decay = weight_decay self.eps = eps def __call__(self, rule, param): p, g = param.data, param.grad if p is None or g is None: return with chainer.using_device(param.device): xp = param.device.xp if xp is backend.chainerx: # TODO(ecastill): norm in chainerx p_norm = xp.sqrt(xp.sum(p*p)) g_norm = xp.sqrt(xp.sum(g*g)) else: # weight norm p_norm = xp.linalg.norm(p) # grad norm g_norm = xp.linalg.norm(g) local_rate = (p_norm / (self.eps + g_norm + self.weight_decay * p_norm)) rate = xp.where(p_norm > self.threshold, local_rate, 1.0) if xp is backend.cuda: kernel = backend.cuda.elementwise( 'T p, T rate, T weight_decay', 'T g', 'g += weight_decay * p; g *= rate;', 'lars') kernel(p, rate, self.weight_decay, g) else: g += self.weight_decay * p g *= rate
4,461
38.839286
79
py
chainer
chainer-master/chainer/optimizer_hooks/gradient_hard_clipping.py
import chainer from chainer import backend class GradientHardClipping(object): """Optimizer/UpdateRule hook function for gradient clipping. This hook function clips all gradient arrays to be within a lower and upper bound. Args: lower_bound (float): The lower bound of the gradient value. upper_bound (float): The upper bound of the gradient value. Attributes: ~optimizer_hooks.GradientHardClipping.lower_bound (float): The lower bound of the gradient value. ~optimizer_hooks.GradientHardClipping.upper_bound (float): The upper bound of the gradient value. ~optimizer_hooks.GradientHardClipping.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). ~optimizer_hooks.GradientHardClipping.call_for_each_param (bool): \ Specifies if this hook is called for each parameter (``True``) or only once (``False``) by an optimizer to which this hook is registered. This function does not expect users to switch the value from default one, which is `True`. .. versionadded:: 4.0.0 The *timing* parameter. """ name = 'GradientHardClipping' call_for_each_param = True timing = 'pre' def __init__(self, lower_bound, upper_bound): self.lower_bound = lower_bound self.upper_bound = upper_bound def __call__(self, rule, param): grad = param.grad if grad is None: return with chainer.using_device(param.device): xp = param.device.xp # TODO(kshitij12345): Fix when chainerx.clip # supports kwarg `out`. if xp == backend.chainerx \ or isinstance(param.grad, backend.intel64.mdarray): grad[...] = grad.clip(self.lower_bound, self.upper_bound) else: # Save on new object allocation when using numpy and cupy # using kwarg `out` xp.clip(grad, self.lower_bound, self.upper_bound, out=grad)
2,360
38.35
79
py
chainer
chainer-master/chainer/optimizer_hooks/gradient_clipping.py
import collections import numpy import six import chainer from chainer import backend def _sum_sqnorm_grads(params): # Calculates sum of squares of gradients. # Returns a tuple of the sum and the device of the sum. # The device will be `None` in multi-device case. # If the inputs are on a single device, the sum is returned # as an ndarray on the device, so that no synchronization is taken place. # If there are multiple devices, accumulation is done on each device # first, and the total sum is returned as a python float. # TODO(niboshi): Support and test len(params) == 0 params_grouped = collections.defaultdict(list) devices_map = {} # Group params by devices. for param in params: device = param.device params_grouped[device.name].append(param) devices_map[device.name] = device # Calculates partial sums for each device. sq_sums = [] for device_name, paramlist in six.iteritems(params_grouped): device = devices_map[device_name] with chainer.using_device(device): dots = [] for param in paramlist: g = param.grad g = g.ravel() dots.append(g.dot(g)) sq_sums.append(sum(dots)) # Return the total sum. if len(sq_sums) == 1: # single device sqnorm = sq_sums[0] ret_device = params[0].device else: # multi-device sqnorm = sum([float(s) for s in sq_sums]) ret_device = None return sqnorm, ret_device class GradientClipping(object): """Optimizer hook function for gradient clipping. This hook function scales all gradient arrays to fit to the defined L2 norm threshold. Args: threshold (float): L2 norm threshold. Attributes: ~optimizer_hooks.GradientClipping.threshold (float): L2 norm threshold of gradient norm. ~optimizer_hooks.GradientClipping.timing (string): Specifies when this hook should be called by the Optimizer/UpdateRule. Valid values are 'pre' (before any updates) and 'post' (after any updates). .. versionadded:: 4.0.0 The *timing* parameter. """ name = 'GradientClipping' timing = 'pre' def __init__(self, threshold): self.threshold = threshold def __call__(self, opt): sqnorm, device = _sum_sqnorm_grads(list(opt.target.params(False))) if device is None: # Assign a dummy device for using_device. device = backend.CpuDevice() with chainer.using_device(device): norm = device.xp.sqrt(sqnorm) # TODO(niboshi): Could be inf if norm == 0 rate = self.threshold / norm # In NumPy backend, `rate` is already available on CPU and thus # can be compared against 1 without extra overhead. # Otherwise `clip` is used to avoid synchronization. if device.xp is numpy: if rate >= 1: return else: rate = rate.clip(None, 1) for param in opt.target.params(False): grad = param.grad with chainer.using_device(param.device): grad *= rate
3,363
30.439252
79
py
chainer
chainer-master/chainer/iterators/dali_iterator.py
from __future__ import division from chainer.dataset import iterator from chainer import utils class DaliIterator(iterator.Iterator): """(Experimental) Iterator for DALI pipeline. Args: pipeline: DALI pipeline. repeat (bool): If ``True``, it infinitely loops over the dataset. Otherwise, it stops iteration at the end of the first epoch. """ def __init__(self, pipeline, repeat=True): utils.experimental('DaliIterator') self.pipeline = pipeline self._repeat = repeat self._is_build = False self.epoch_size = 1 # dummy self.reset() def __next__(self): if not self._is_build: self.pipeline.build() self._is_build = True self.epoch_size = tuple(self.pipeline.epoch_size().values())[0] if not self._repeat and self.epoch > 0: raise StopIteration self._previous_epoch_detail = self.epoch_detail i = self.current_position i_end = i + self.batch_size N = self.epoch_size if i_end >= N: if self._repeat: self.current_position = i_end - N else: self.current_position = 0 self.epoch += 1 self.is_new_epoch = True else: self.current_position = i_end self.is_new_epoch = False return self.pipeline.run() next = __next__ @property def batch_size(self): return self.pipeline.batch_size @property def epoch_detail(self): return self.epoch + self.current_position / self.epoch_size @property def previous_epoch_detail(self): if self._previous_epoch_detail < 0: return None return self._previous_epoch_detail def serialize(self, serializer): self.current_position = serializer('current_position', self.current_position) self.epoch = serializer('epoch', self.epoch) self.is_new_epoch = serializer('is_new_epoch', self.is_new_epoch) try: self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) except KeyError: # guess previous_epoch_detail for older version self._previous_epoch_detail = self.epoch + \ (self.current_position - self.batch_size) / self.epoch_size if self.epoch_detail > 0: self._previous_epoch_detail = max( self._previous_epoch_detail, 0.) else: self._previous_epoch_detail = -1. def reset(self): self.current_position = 0 self.epoch = 0 self.is_new_epoch = False # use -1 instead of None internally. self._previous_epoch_detail = -1. @property def repeat(self): return self._repeat
2,926
28.867347
75
py
chainer
chainer-master/chainer/iterators/_statemachine.py
import collections import numpy IteratorState = collections.namedtuple('IteratorState', ( 'current_position', 'epoch', 'is_new_epoch', 'order')) def iterator_statemachine(state, batch_size, repeat, order_sampler, dataset_len): i, epoch, _, order = state if not repeat and epoch > 0: return state, None indices_list = [] n = dataset_len if order is None else len(order) if repeat and n == 0: raise ValueError('Epoch size must be positive for an iterator ' 'that repeats.') i_end = i + batch_size is_new_epoch = False while i_end >= n: if order is None: indices_list.append(numpy.arange(i, n, dtype=numpy.intp)) else: indices_list.append(order[i:n]) if order is not None: new_order = order_sampler(order, i) if len(new_order) != len(order): raise ValueError('The size of order does not match ' 'the size of the previous order.') order = new_order epoch += 1 is_new_epoch = True i = 0 if repeat: i_end -= n else: i_end = 0 break # explicit break in case n == 0 if order is None: indices_list.append(numpy.arange(i, i_end, dtype=numpy.intp)) else: indices_list.append(order[i:i_end]) state = IteratorState(i_end, epoch, is_new_epoch, order) indices = numpy.concatenate(indices_list) return state, indices
1,564
26.45614
71
py
chainer
chainer-master/chainer/iterators/multithread_iterator.py
from __future__ import division from multiprocessing import pool import numpy from chainer.dataset import iterator from chainer.iterators import _statemachine from chainer.iterators.order_samplers import ShuffleOrderSampler class MultithreadIterator(iterator.Iterator): """Dataset iterator that loads examples in parallel. This is an implementation of :class:`~chainer.dataset.Iterator` that loads examples with worker threads. It uses the standard :mod:`threading` module to parallelize the loading. Note that this iterator effectively prefetches the examples for the next batch asynchronously after the current batch is returned. This iterator saves ``-1`` instead of ``None`` in snapshots since some serializers do not support ``None``. Args: dataset (~chainer.dataset.Dataset): Dataset to iterate. batch_size (int): Number of examples within each batch. repeat (bool): If ``True``, it infinitely loops over the dataset. Otherwise, it stops iteration at the end of the first epoch. shuffle (bool): If ``True``, the order of examples is shuffled at the beginning of each epoch. Otherwise, examples are extracted in the order of indexes. If ``None`` and no ``order_sampler`` is given, the behavior is the same as the case with ``shuffle=True``. n_threads (int): Number of worker threads. order_sampler (callable): A callable that generates the order of the indices to sample in the next epoch when a epoch finishes. This function should take two arguments: the current order and the current position of the iterator. This should return the next order. The size of the order should remain constant. This option cannot be used when ``shuffle`` is not ``None``. """ def __init__(self, dataset, batch_size, repeat=True, shuffle=None, n_threads=1, order_sampler=None): self.dataset = dataset self.batch_size = batch_size self._repeat = repeat self._shuffle = shuffle if self._shuffle is not None: if order_sampler is not None: raise ValueError('`shuffle` is not `None` and a custom ' '`order_sampler` is set. Please set ' '`shuffle` to `None` to use the custom ' 'order sampler.') else: if self._shuffle: order_sampler = ShuffleOrderSampler() else: if order_sampler is None: order_sampler = ShuffleOrderSampler() self.order_sampler = order_sampler self.n_threads = n_threads self._pool = None self.reset() def reset(self): if self.order_sampler is None: order = None else: order = self.order_sampler(numpy.arange(len(self.dataset)), 0) self._state = _statemachine.IteratorState(0, 0, False, order) self._previous_epoch_detail = -1. # reset internal state self._next = None def finalize(self): pool = self._pool self._next = None self._pool = None if pool is not None: pool.terminate() def __next__(self): if self._next is None: # load for the first iteration self._invoke_prefetch() batch = self._get() self._invoke_prefetch() # prefetch for the next iteration return batch next = __next__ @property def current_position(self): return self._state.current_position @property def epoch(self): return self._state.epoch @property def is_new_epoch(self): return self._state.is_new_epoch @property def epoch_detail(self): return self.epoch + self.current_position / self._epoch_size @property def previous_epoch_detail(self): # use -1 instead of None internally. if self._previous_epoch_detail < 0: return None return self._previous_epoch_detail def serialize(self, serializer): current_position = serializer( 'current_position', self.current_position) epoch = serializer('epoch', self.epoch) is_new_epoch = serializer('is_new_epoch', self.is_new_epoch) order = serializer('_order', self._state.order) self._state = _statemachine.IteratorState( current_position, epoch, is_new_epoch, order) self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) # Old version serialized ``None``. if self._previous_epoch_detail is None: self._previous_epoch_detail = -1. self._next = None @staticmethod def _read(args): dataset, index = args return dataset[index] def _invoke_prefetch(self): assert self._next is None self._next_state, indices = _statemachine.iterator_statemachine( self._state, self.batch_size, self.repeat, self.order_sampler, len(self.dataset)) if indices is None: self._next = None else: if self._pool is None: self._pool = pool.ThreadPool(self.n_threads) args = [(self.dataset, index) for index in indices] self._next = self._pool.map_async(MultithreadIterator._read, args) def _get(self): self._previous_epoch_detail = self.epoch_detail self._state = self._next_state next = self._next if next is None: raise StopIteration self._next = None while not next.ready(): next.wait(0.5) # To avoid interruption bug in Python2 batch = [data for data in next.get()] return batch @property def _epoch_size(self): order = self._state.order if order is None: epoch_size = len(self.dataset) else: epoch_size = len(order) return epoch_size @property def repeat(self): return self._repeat
6,202
32.52973
78
py
chainer
chainer-master/chainer/iterators/multiprocess_iterator.py
from __future__ import division import datetime import multiprocessing from multiprocessing import sharedctypes # type: ignore import signal import sys import threading import warnings import numpy import six from chainer.dataset import iterator from chainer.iterators import _statemachine from chainer.iterators.order_samplers import ShuffleOrderSampler _response_time = 0.1 def _raise_timeout_warning(): warnings.warn( 'Stalled dataset is detected. ' 'See the documentation of MultiprocessIterator for common causes and ' 'workarounds:\n' 'https://docs.chainer.org/en/stable/reference/generated/' 'chainer.iterators.MultiprocessIterator.html', MultiprocessIterator.TimeoutWarning) class MultiprocessIterator(iterator.Iterator): """Dataset iterator that loads examples in parallel. This is an implementation of :class:`~chainer.dataset.Iterator` that loads examples with worker processes. It uses the standard :mod:`multiprocessing` module to parallelize the loading. The dataset is sent to the worker processes in the standard way using pickle. Note that this iterator effectively prefetches the examples for the next batch asynchronously after the current batch is returned. This iterator saves ``-1`` instead of ``None`` in snapshots since some serializers do not support ``None``. .. note:: When you are using OpenCV somewhere in your code and the ``MultiprocessIterator`` is used in the training code, the training loop may get stuck at some point. In such situation, there are several workarounds to prevent the process got stuck. 1. Set the environment variable as follows: ``OMP_NUM_THREADS=1`` 2. Add ``cv2.setNumThreads(0)`` right after ``import cv2`` in your training script. 3. Use :class:`~chainer.iterators.MultithreadIterator` instead of ``MultiprocessIterator``. Args: dataset (~chainer.dataset.Dataset): Dataset to iterate. batch_size (int): Number of examples within each batch. repeat (bool): If ``True``, it infinitely loops over the dataset. Otherwise, it stops iteration at the end of the first epoch. shuffle (bool): If ``True``, the order of examples is shuffled at the beginning of each epoch. Otherwise, examples are extracted in the order of indexes. If ``None`` and no ``order_sampler`` is given, the behavior is the same as the case with ``shuffle=True``. n_processes (int): Number of worker processes. The number of CPUs is used by default. n_prefetch (int): Number of prefetch batches. shared_mem (int): The size of using shared memory per data. If ``None``, size is adjusted automatically. dataset_timeout (float): :class:`MultiprocessIterator.TimeoutWarning` will be issued after this time in seconds elapsed in each dataset realization. ``None`` to disable the warning. You can turn this warning into an error by using :func:`warnings.simplefilter`:: warnings.simplefilter( 'error', chainer.iterators.MultiprocessIterator.TimeoutWarning) order_sampler (callable): A callable that generates the order of the indices to sample in the next epoch when a epoch finishes. This function should take two arguments: the current order and the current position of the iterator. This should return the next order. The size of the order should remain constant. This option cannot be used when ``shuffle`` is not ``None``. maxtasksperchild (int): Number of tasks a worker of prefetch process can complete before it will exit and be replaced with a fresh worker process, to enable unused resources to be freed. If ``None``, worker processes will live as long as the pool. """ class TimeoutWarning(RuntimeWarning): pass _interruption_testing = False # for testing _finalized = False _prefetch_loop = None _comm = None def __init__(self, dataset, batch_size, repeat=True, shuffle=None, n_processes=None, n_prefetch=1, shared_mem=None, order_sampler=None, dataset_timeout=30.0, maxtasksperchild=None): self.dataset = dataset self.batch_size = batch_size self.repeat = repeat self.shuffle = shuffle self.n_processes = n_processes or multiprocessing.cpu_count() self.n_prefetch = max(n_prefetch, 1) self.shared_mem = shared_mem self.dataset_timeout = dataset_timeout self._maxtasksperchild = maxtasksperchild if self.shuffle is not None: if order_sampler is not None: raise ValueError('`shuffle` is not `None` and a custom ' '`order_sampler` is set. Please set ' '`shuffle` to `None` to use the custom ' 'order sampler.') else: if self.shuffle: order_sampler = ShuffleOrderSampler() else: if order_sampler is None: order_sampler = ShuffleOrderSampler() self.order_sampler = order_sampler self._initialize_loop() def _initialize_loop(self): self._comm = _Communicator(self.n_prefetch, self.dataset_timeout) self.reset() self._prefetch_loop = _PrefetchLoop( self.dataset, self.batch_size, self.repeat, self.n_processes, self.n_prefetch, self.shared_mem, self._comm, self.order_sampler, self._interruption_testing, self._maxtasksperchild) # defer launching prefetch thread until creating the worker pool, # not to leave a background thread in forked processes. def __next__(self): measure_mode = False if self._prefetch_loop.thread is None: if self._prefetch_loop.measure_required(): measure_mode = True batch, state = self._prefetch_loop.measure( self.dataset_timeout) self._prefetch_loop.launch_thread() if not measure_mode: batch, state = self._comm.get() self._previous_epoch_detail = self.epoch_detail self._state = state if batch is None: raise StopIteration else: return batch next = __next__ def finalize(self): if self._finalized: return if self._comm is not None: self._comm.terminate() if self._prefetch_loop is not None: self._prefetch_loop.terminate() self._comm = None self._prefetch_loop = None self._finalized = True def __copy__(self): # This function is implemented for backward compatibility. # Please use `reset` normally. other = MultiprocessIterator( self.dataset, self.batch_size, self.repeat, shuffle=None, n_processes=self.n_processes, n_prefetch=self.n_prefetch, shared_mem=self.shared_mem, order_sampler=self.order_sampler) other._reset_state(self.current_position, self.epoch, self.is_new_epoch, self._state.order) other._previous_epoch_detail = self._previous_epoch_detail return other @property def current_position(self): return self._state.current_position @property def epoch(self): return self._state.epoch @property def is_new_epoch(self): return self._state.is_new_epoch @property def epoch_detail(self): return self.epoch + self.current_position / self._epoch_size @property def previous_epoch_detail(self): if self._previous_epoch_detail < 0: return None return self._previous_epoch_detail def serialize(self, serializer): current_position = serializer('current_position', self.current_position) epoch = serializer('epoch', self.epoch) is_new_epoch = serializer('is_new_epoch', self.is_new_epoch) order = self._state.order.copy() try: serializer('order', order) except KeyError: serializer('_order', order) self._reset_state(current_position, epoch, is_new_epoch, order) try: self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) except KeyError: # guess previous_epoch_detail for older version self._previous_epoch_detail = self.epoch + \ (self.current_position - self.batch_size) / self._epoch_size if self.epoch_detail > 0: self._previous_epoch_detail = max( self._previous_epoch_detail, 0.) else: self._previous_epoch_detail = -1. def reset(self): if self.order_sampler is None: order = None else: order = self.order_sampler(numpy.arange(len(self.dataset)), 0) self._reset_state(0, 0, False, order) self._previous_epoch_detail = -1. def _reset_state(self, current_position, epoch, is_new_epoch, order): if self._finalized: raise NotImplementedError( 'Reset of finalized MultiProcessIterator is currently not ' 'supported.') self._state = _statemachine.IteratorState( current_position, epoch, is_new_epoch, order) self._comm.reset(self._state) @property def _epoch_size(self): order = self._state.order if order is None: epoch_size = len(self.dataset) else: epoch_size = len(order) return epoch_size def __getstate__(self): # We trick the serializer to fill a dict for us # this allows us to use the same code for both # chainer and pickle serializers state = {} self.serialize(lambda k, v: state.__setitem__(k, v)) self._reset_state(self.current_position, self.epoch, self.is_new_epoch, state['order']) # Unpickling resets the instance without calling __init__ # Chainer serializers dumps the state in an existing # object hence we need to save the initial parameters too init = self.__dict__.copy() del init['_comm'] del init['_state'] del init['_prefetch_loop'] # TODO(ecastill): When pickling this object there is the risk to copy # the entire dataset. If the dataset is entirely in memory # it can be duplicated when spawning new processes. state['init'] = init return state def __setstate__(self, state): self.__dict__.update(state['init']) self._initialize_loop() # Iterator state is restored after initialization self._reset_state(state['current_position'], state['epoch'], state['is_new_epoch'], state['order']) self._previous_epoch_detail = state['previous_epoch_detail'] class _Communicator(object): STATUS_CONTINUE = 0 STATUS_RESET = 1 STATUS_TERMINATE = 2 def __init__(self, n_prefetch, dataset_timeout): self.n_prefetch = n_prefetch self.dataset_timeout = dataset_timeout self._lock = threading.Lock() self._not_empty_cond = threading.Condition(self._lock) self._not_full_cond = threading.Condition(self._lock) self._batch_queue = [] self._status = _Communicator.STATUS_CONTINUE self._reset_count = 0 @property def is_terminated(self): with self._lock: return self._status == _Communicator.STATUS_TERMINATE # called from iterator def get(self): with self._lock: start = datetime.datetime.now() while not self._batch_queue: self._not_empty_cond.wait(_response_time) dt = datetime.datetime.now() - start if (self.dataset_timeout is not None and dt > datetime.timedelta( seconds=self.dataset_timeout)): _raise_timeout_warning() batch, prefetch_state = self._batch_queue.pop(0) self._not_full_cond.notify() return batch, prefetch_state # called from iterator def reset(self, prefetch_state): with self._lock: self._status = _Communicator.STATUS_RESET self._prefetch_state = prefetch_state self._batch_queue = [] self._not_full_cond.notify() self._reset_count += 1 # called from iterator def terminate(self): with self._lock: self._status = _Communicator.STATUS_TERMINATE self._batch_queue = [] self._not_full_cond.notify() self._reset_count += 1 # called from thread def check(self): with self._lock: status = self._status self._status = _Communicator.STATUS_CONTINUE prefetch_state = None if status == _Communicator.STATUS_RESET: prefetch_state = self._prefetch_state return status, prefetch_state, self._reset_count # called from thread def put(self, batch, prefetch_state, reset_count): with self._lock: if len(self._batch_queue) == self.n_prefetch: self._not_full_cond.wait() if reset_count == self._reset_count: self._batch_queue.append((batch, prefetch_state)) self._not_empty_cond.notify() class _PrefetchLoop(object): _thread = None _pool = None _terminating = False def __init__(self, dataset, batch_size, repeat, n_processes, n_prefetch, mem_size, comm, order_sampler, _interruption_testing, maxtasksperchild): self.dataset = dataset self.batch_size = batch_size self.repeat = repeat self.n_processes = n_processes self.mem_size = mem_size self._comm = comm self.order_sampler = order_sampler self.maxtasksperchild = maxtasksperchild self._allocate_shared_memory() self._interruption_testing = _interruption_testing def terminate(self): self._terminating = True # Terminate the thread first because it depends on the pool. if self._thread is not None: while self._thread.is_alive(): self._thread.join(_response_time) if self._pool is not None: self._pool.terminate() self._thread = None self._pool = None @property def thread(self): return self._thread def measure_required(self): return self.mem_size is None def measure(self, dataset_timeout): # dataset_timeout: timeout in seconds or None status, prefetch_state, _ = self._comm.check() if status == _Communicator.STATUS_RESET: self.prefetch_state = prefetch_state self.prefetch_state, indices = _statemachine.iterator_statemachine( self.prefetch_state, self.batch_size, self.repeat, self.order_sampler, len(self.dataset)) if indices is None: # stop iteration batch = None else: batch_ret = [None] def fetch_batch(): batch_ret[0] = [self.dataset[idx] for idx in indices] if dataset_timeout is None: # Timeout is not set: fetch synchronously fetch_batch() else: # Timeout is set: fetch asynchronously and watch for timeout thr = threading.Thread(target=fetch_batch) thr.daemon = True thr.start() thr.join(dataset_timeout) if thr.is_alive(): _raise_timeout_warning() thr.join() batch = batch_ret[0] self.mem_size = max(map(_measure, batch)) self._allocate_shared_memory() return batch, self.prefetch_state def _allocate_shared_memory(self): if self.measure_required(): self.mem_bulk = None else: self.mem_bulk = \ sharedctypes.RawArray('b', self.batch_size * self.mem_size) def launch_thread(self): self._pool = multiprocessing.Pool( processes=self.n_processes, initializer=_fetch_setup, initargs=(self.dataset, self.mem_size, self.mem_bulk), maxtasksperchild=self.maxtasksperchild) if self._interruption_testing: pids = self._pool.map(_report_pid, range(self.n_processes)) print(' '.join(map(str, pids))) sys.stdout.flush() thread = threading.Thread(target=self._run, name='prefetch_loop') thread.setDaemon(True) thread.start() self._thread = thread return thread def _run(self): # The entry routine of the prefetch thread. alive = True try: while alive: if self._terminating: break alive = self._task() finally: self._pool.close() self._pool.join() def _task(self): # Do a single task in the prefetch thread. # Returns a bool indicating whether the loop should continue running. status, prefetch_state, reset_count = self._comm.check() if status == _Communicator.STATUS_RESET: self.prefetch_state = prefetch_state elif status == _Communicator.STATUS_TERMINATE: return False # stop loop self.prefetch_state, indices = _statemachine.iterator_statemachine( self.prefetch_state, self.batch_size, self.repeat, self.order_sampler, len(self.dataset)) if indices is None: # stop iteration batch = None else: future = self._pool.map_async(_fetch_run, enumerate(indices)) while True: try: data_all = future.get(_response_time) except multiprocessing.TimeoutError: if self._comm.is_terminated: return False else: break batch = [_unpack(data, self.mem_bulk) for data in data_all] self._comm.put(batch, self.prefetch_state, reset_count) return True # Using `parameterized` function (e.g. bound method) with Pool is tricky due to # restrictions imposed by Pickle. Picklable types differ across versions. # Just using top-level function with globals seems to be safest. # it doesn't mean thread safety broken or global variables visible; # notice that each process uses different address space. # To make static linter happy, we first initialize global variables. _fetch_dataset = None _fetch_mem_size = None _fetch_mem_bulk = None def _fetch_setup(dataset, mem_size, mem_bulk): global _fetch_dataset, _fetch_mem_size, _fetch_mem_bulk signal.signal(signal.SIGINT, signal.SIG_IGN) _fetch_dataset = dataset _fetch_mem_size = mem_size _fetch_mem_bulk = mem_bulk def _fetch_run(inputs): i, index = inputs data = _fetch_dataset[index] if _fetch_mem_bulk is not None: offset = i * _fetch_mem_size limit = offset + _fetch_mem_size data = _pack(data, _fetch_mem_bulk, offset, limit) return data def _report_pid(_): # for testing return multiprocessing.current_process().pid class _PackedNdarray(object): def __init__(self, array, mem, offset): self.shape = array.shape self.dtype = array.dtype self.nbytes = array.nbytes self.size = array.size self.offset = offset total = self.offset + self.nbytes if total > len(mem): raise ValueError( 'Shared memory size is too small. expect:{}, actual:{}'.format( total, len(mem))) target = numpy.frombuffer(mem, self.dtype, self.size, self.offset) target[...] = array.ravel() def unpack(self, mem): ret = numpy.frombuffer(mem, self.dtype, self.size, self.offset) ret = ret.reshape(self.shape).copy() return ret def _measure(data): expect = 0 t = type(data) if t is tuple or t is list or t is dict: for v in data: if isinstance(v, numpy.ndarray): expect += v.nbytes return expect def _pack(data, mem, offset, limit): if len(mem) == 0: return data t = type(data) over = False if t is tuple or t is list: ret = [] for v in data: if isinstance(v, numpy.ndarray): if v.nbytes + offset > limit: over = True else: v = _PackedNdarray(v, mem, offset) offset += v.nbytes ret.append(v) data = t(ret) elif t is dict: ret = {} for k, v in six.iteritems(data): if isinstance(v, numpy.ndarray): if v.nbytes + offset > limit: over = True else: v = _PackedNdarray(v, mem, offset) offset += v.nbytes ret[k] = v data = ret elif t is numpy.ndarray: if data.nbytes + offset > limit: over = True else: data = _PackedNdarray(data, mem, offset) offset += data.nbytes if over: expect = _measure(data) warnings.warn( 'Shared memory size is too small.\n' + 'Please set shared_mem option for MultiprocessIterator.\n' + 'Expect shared memory size: {} bytes.\n'.format(expect) + 'Actual shared memory size: {} bytes.'.format(limit - offset), UserWarning) return data def _unpack(data, mem): if len(mem) == 0: return data t = type(data) if t is tuple or t is list: ret = [] for v in data: if isinstance(v, _PackedNdarray): v = v.unpack(mem) ret.append(v) data = t(ret) elif t is dict: ret = {} for k, v in six.iteritems(data): if isinstance(v, _PackedNdarray): v = v.unpack(mem) ret[k] = v data = ret elif t is _PackedNdarray: data = data.unpack(mem) return data
22,878
33.98318
79
py
chainer
chainer-master/chainer/iterators/__init__.py
# import classes and functions from chainer.iterators.multiprocess_iterator import MultiprocessIterator # NOQA from chainer.iterators.multithread_iterator import MultithreadIterator # NOQA from chainer.iterators.serial_iterator import SerialIterator # NOQA from chainer.iterators.dali_iterator import DaliIterator # NOQA from chainer.iterators.order_samplers import OrderSampler # NOQA from chainer.iterators.order_samplers import ShuffleOrderSampler # NOQA
466
45.7
80
py
chainer
chainer-master/chainer/iterators/serial_iterator.py
from __future__ import division import numpy from chainer.dataset import iterator from chainer.iterators import _statemachine from chainer.iterators.order_samplers import ShuffleOrderSampler class SerialIterator(iterator.Iterator): """Dataset iterator that serially reads the examples. This is a simple implementation of :class:`~chainer.dataset.Iterator` that just visits each example in either the order of indexes or a shuffled order. To avoid unintentional performance degradation, the ``shuffle`` option is set to ``True`` by default. For validation, it is better to set it to ``False`` when the underlying dataset supports fast slicing. If the order of examples has an important meaning and the updater depends on the original order, this option should be set to ``False``. This iterator saves ``-1`` instead of ``None`` in snapshots since some serializers do not support ``None``. Args: dataset: Dataset to iterate. batch_size (int): Number of examples within each batch. repeat (bool): If ``True``, it infinitely loops over the dataset. Otherwise, it stops iteration at the end of the first epoch. shuffle (bool): If ``True``, the order of examples is shuffled at the beginning of each epoch. Otherwise, examples are extracted in the order of indexes. If ``None`` and no ``order_sampler`` is given, the behavior is the same as the case with ``shuffle=True``. order_sampler (callable): A callable that generates the order of the indices to sample in the next epoch when a epoch finishes. This function should take two arguments: the current order and the current position of the iterator. This should return the next order. The size of the order should remain constant. This option cannot be used when ``shuffle`` is not ``None``. """ def __init__(self, dataset, batch_size, repeat=True, shuffle=None, order_sampler=None): self.dataset = dataset self.batch_size = batch_size self._repeat = repeat self._shuffle = shuffle if self._shuffle is not None: if order_sampler is not None: raise ValueError('`shuffle` is not `None` and a custom ' '`order_sampler` is set. Please set ' '`shuffle` to `None` to use the custom ' 'order sampler.') else: if self._shuffle: order_sampler = ShuffleOrderSampler() else: if order_sampler is None: order_sampler = ShuffleOrderSampler() self.order_sampler = order_sampler self.reset() def __next__(self): self._previous_epoch_detail = self.epoch_detail self._state, indices = _statemachine.iterator_statemachine( self._state, self.batch_size, self.repeat, self.order_sampler, len(self.dataset)) if indices is None: raise StopIteration batch = [self.dataset[index] for index in indices] return batch next = __next__ @property def current_position(self): return self._state.current_position @property def epoch(self): return self._state.epoch @property def is_new_epoch(self): return self._state.is_new_epoch @property def epoch_detail(self): return self.epoch + self.current_position / self._epoch_size @property def previous_epoch_detail(self): # use -1 instead of None internally. if self._previous_epoch_detail < 0: return None return self._previous_epoch_detail def serialize(self, serializer): current_position = serializer('current_position', self.current_position) epoch = serializer('epoch', self.epoch) is_new_epoch = serializer('is_new_epoch', self.is_new_epoch) order = self._state.order if order is not None: try: serializer('order', order) except KeyError: serializer('_order', order) self._state = _statemachine.IteratorState( current_position, epoch, is_new_epoch, order) try: self._previous_epoch_detail = serializer( 'previous_epoch_detail', self._previous_epoch_detail) except KeyError: # guess previous_epoch_detail for older version self._previous_epoch_detail = self.epoch + \ (self.current_position - self.batch_size) / self._epoch_size if self.epoch_detail > 0: self._previous_epoch_detail = max( self._previous_epoch_detail, 0.) else: self._previous_epoch_detail = -1. def reset(self): if self.order_sampler: order = self.order_sampler( numpy.arange(len(self.dataset)), 0) else: order = None self._state = _statemachine.IteratorState(0, 0, False, order) self._previous_epoch_detail = -1. @property def _epoch_size(self): order = self._state.order if order is None: epoch_size = len(self.dataset) else: epoch_size = len(order) return epoch_size @property def repeat(self): return self._repeat
5,543
35.473684
78
py
chainer
chainer-master/chainer/iterators/order_samplers.py
import numpy class OrderSampler(object): """Base class of all order samplers. Every order sampler subclass has to provide a method :meth:`__call__`. This method is called by an iterator before a new epoch, and it should return a new index order for the next epoch. """ def __call__(self, current_order, current_position): """Sample the next order. Args: current_order (numpy.ndarray): 1-D array of indices. The length should be the same as the dataset to sample data from. current_position (int): The current position of an iterator. Returns: numpy.ndarray: 1-D array of indices. This is the order in which examples are sampled from a dataset in the next epoch. """ raise NotImplementedError class ShuffleOrderSampler(OrderSampler): """Sampler that generates random orders. This is expected to be used together with Chainer's iterators. An order sampler is called by an iterator every epoch. The two initializations below create basically the same objects. >>> dataset = [(1, 2), (3, 4)] >>> it = chainer.iterators.MultiprocessIterator(dataset, 1, shuffle=True) >>> it = chainer.iterators.MultiprocessIterator( ... dataset, 1, order_sampler=chainer.iterators.ShuffleOrderSampler()) Args: random_state (numpy.random.RandomState): Pseudo-random number generator. """ def __init__(self, random_state=None): if random_state is None: random_state = numpy.random.random.__self__ self._random = random_state def __call__(self, current_order, current_position): return self._random.permutation(len(current_order))
1,793
28.9
78
py
chainer
chainer-master/chainer/optimizers/smorms3.py
import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class SMORMS3Hyperparameter(tpe.Protocol): """Protocol class for hyperparameter of Simon Funk's SMORMS3. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float eps = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: SMORMS3Hyperparameter # NOQA _default_hyperparam.lr = 0.001 _default_hyperparam.eps = 1e-16 class SMORMS3Rule(optimizer.UpdateRule): """Update rule for Simon Funk's SMORMS3. See :class:`~chainer.optimizers.SMORMS3` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. eps (float): Small value for the numerical stability. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None, eps=None): super(SMORMS3Rule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if eps is not None: self.hyperparam.eps = eps def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['mem'] = xp.ones_like(param.data) self.state['g'] = xp.zeros_like(param.data) self.state['g2'] = xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return mem, g, g2 = self.state['mem'], self.state['g'], self.state['g2'] r = 1 / (mem + 1) g = (1 - r) * g + r * grad g2 = (1 - r) * g2 + r * grad * grad x = g * g / (g2 + self.hyperparam.eps) param.data -= grad * numpy.minimum(x, self.hyperparam.lr) \ / (numpy.sqrt(g2) + self.hyperparam.eps) mem = 1 + mem * (1 - x) self.state['mem'], self.state['g'], self.state['g2'] = mem, g, g2 def update_core_gpu(self, param): grad = param.grad if grad is None: return if SMORMS3Rule._kernel is None: SMORMS3Rule._kernel = cuda.elementwise( 'T grad, T lr, T eps', 'T param, T mem, T g, T g2', '''T r, x; r = 1 / (mem + 1); g = (1 - r) * g + r * grad; g2 = (1 - r) * g2 + r * grad * grad; x = g * g / (g2 + eps); param -= grad * min(lr, x) / (sqrt(g2) + eps); mem = 1 + mem * (1 - x) ''', 'smorms3') SMORMS3Rule._kernel( grad, self.hyperparam.lr, self.hyperparam.eps, param.data, self.state['mem'], self.state['g'], self.state['g2']) class SMORMS3(optimizer.GradientMethod): """Simon Funk's SMORMS3. See http://sifter.org/~simon/journal/20150420.html. Args: lr (float): Learning rate. eps (float): Small value for the numerical stability. """ def __init__(self, lr=_default_hyperparam.lr, eps=_default_hyperparam.eps): super(SMORMS3, self).__init__() self.hyperparam.lr = lr self.hyperparam.eps = eps lr = optimizer.HyperparameterProxy('lr') eps = optimizer.HyperparameterProxy('eps') def create_update_rule(self): return SMORMS3Rule(self.hyperparam)
3,641
29.605042
86
py
chainer
chainer-master/chainer/optimizers/momentum_sgd.py
import chainer from chainer.backends import cuda from chainer.backends import intel64 from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class MomentumSGDHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of classical momentum SGD. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float momentum = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: MomentumSGDHyperparameter # NOQA _default_hyperparam.lr = 0.01 _default_hyperparam.momentum = 0.9 class MomentumSGDRule(optimizer.UpdateRule): """Update rule for the classical momentum SGD. See :class:`~chainer.optimizers.MomentumSGD` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None, momentum=None): super(MomentumSGDRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if momentum is not None: self.hyperparam.momentum = momentum def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['v'] = xp.zeros_like(param.data) # For iDeep if isinstance(param.data, intel64.mdarray): self.state['v'] = intel64.ideep.array( self.state['v'], itype=intel64.ideep.wgt_array) def update_core_cpu(self, param): grad = param.grad if grad is None: return v = self.state['v'] if isinstance(v, intel64.mdarray): v.inplace_axpby(self.hyperparam.momentum, - self.hyperparam.lr, grad) param.data += v else: v *= self.hyperparam.momentum v -= self.hyperparam.lr * grad param.data += v def update_core_gpu(self, param): grad = param.grad if grad is None: return if MomentumSGDRule._kernel is None: MomentumSGDRule._kernel = cuda.elementwise( 'T grad, T lr, T momentum', 'T param, T v', '''v = momentum * v - lr * grad; param += v;''', 'momentum_sgd') MomentumSGDRule._kernel( grad, self.hyperparam.lr, self.hyperparam.momentum, param.data, self.state['v']) class MomentumSGD(optimizer.GradientMethod): """Momentum SGD optimizer. Args: lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ def __init__(self, lr=_default_hyperparam.lr, momentum=_default_hyperparam.momentum): super(MomentumSGD, self).__init__() self.hyperparam.lr = lr self.hyperparam.momentum = momentum lr = optimizer.HyperparameterProxy('lr') momentum = optimizer.HyperparameterProxy('momentum') def create_update_rule(self): return MomentumSGDRule(self.hyperparam)
3,423
29.571429
90
py
chainer
chainer-master/chainer/optimizers/sgd.py
from chainer.backends import cuda from chainer.backends import intel64 from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class SGDHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of vanilla stochastic gradient descent. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: SGDHyperparameter # NOQA _default_hyperparam.lr = 0.01 class SGDRule(optimizer.UpdateRule): """Update rule of vanilla stochastic gradient descent. See :class:`~chainer.optimizers.SGD` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None): super(SGDRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr def update_core_cpu(self, param): grad = param.grad if grad is None: return if isinstance(param.data, intel64.mdarray): param.data.inplace_axpby(1.0, -self.hyperparam.lr, grad) else: param.data -= self.hyperparam.lr * grad def update_core_gpu(self, param): grad = param.grad if grad is None: return if SGDRule._kernel is None: SGDRule._kernel = cuda.elementwise( 'T grad, T lr', 'T param', 'param -= lr * grad', 'sgd') SGDRule._kernel(grad, self.hyperparam.lr, param.data) class SGD(optimizer.GradientMethod): """Vanilla Stochastic Gradient Descent. Args: lr (float): Learning rate. """ def __init__(self, lr=_default_hyperparam.lr): super(SGD, self).__init__() self.hyperparam.lr = lr lr = optimizer.HyperparameterProxy('lr') def create_update_rule(self): return SGDRule(self.hyperparam)
2,194
25.768293
84
py
chainer
chainer-master/chainer/optimizers/corrected_momentum_sgd.py
import chainer from chainer.backends import cuda from chainer.backends import intel64 from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class CorrectedMomentumSGDHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of corrected momentum SGD. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float momentum = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: CorrectedMomentumSGDHyperparameter # NOQA _default_hyperparam.lr = 0.01 _default_hyperparam.momentum = 0.9 class CorrectedMomentumSGDRule(optimizer.UpdateRule): """Update rule for the corrected momentum SGD. See :class:`~chainer.optimizers.CorrectedMomentumSGD` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ is_elementwise = True def __init__(self, parent_hyperparam=None, lr=None, momentum=None): super(CorrectedMomentumSGDRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if momentum is not None: self.hyperparam.momentum = momentum def init_state(self, param): with chainer.using_device(param.device): self.state['v'] = param.device.xp.zeros_like(param.data) # For iDeep if isinstance(param.data, intel64.mdarray): self.state['v'] = intel64.ideep.array( self.state['v'], itype=intel64.ideep.wgt_array) def update_core_cpu(self, param): grad = param.grad if grad is None: return v = self.state['v'] if isinstance(v, intel64.mdarray): v.inplace_axpby(self.hyperparam.momentum, -1, grad) param.data += self.hyperparam.lr * v else: v *= self.hyperparam.momentum v -= grad param.data += self.hyperparam.lr * v def update_core_gpu(self, param): grad = param.grad if grad is None: return cuda.elementwise( 'T grad, T lr, T momentum', 'T param, T v', '''v = momentum * v - grad; param += lr * v;''', 'momentum_sgd')( grad, self.hyperparam.lr, self.hyperparam.momentum, param.data, self.state['v']) class CorrectedMomentumSGD(optimizer.GradientMethod): """Momentum SGD optimizer. This implements momentum correction discussed in the third section of `Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour <https://arxiv.org/abs/1706.02677>`_. :class:`~chainer.optimizers.MomentumSGD` implements the equation (10) of the paper. This optimizer implements the equation (9). To get better understanding between the two methods, we show the equivalence between the equation (9) and modification of the equation (10) that takes momentum correction into account. First, we set :math:`v_{t} = \\eta_{t} u_t`. We substitute this relation to the equation (10). .. math:: v_{t+1} &= m\\frac{\\eta_{t+1}}{\\eta_{t}}v_t + \\eta_{t+1}g_t \\\\ &= m\\frac{\\eta_{t+1}}{\\eta_{t}}\\eta_{t}u_t + \\eta_{t+1}g_t \\\\ &= \\eta_{t+1}(m u_t + g_t) \\\\ From this result, we derive :math:`u_{t+1} = m u_t + g_t`, which is how update tensors are calculated by :class:`~chainer.optimizers.CorrectedMomentumSGD`. Thus, the equivalence is shown. Args: lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ def __init__(self, lr=_default_hyperparam.lr, momentum=_default_hyperparam.momentum): super(CorrectedMomentumSGD, self).__init__() self.hyperparam.lr = lr self.hyperparam.momentum = momentum lr = optimizer.HyperparameterProxy('lr') momentum = optimizer.HyperparameterProxy('momentum') def create_update_rule(self): return CorrectedMomentumSGDRule(self.hyperparam)
4,407
32.393939
99
py
chainer
chainer-master/chainer/optimizers/msvag.py
from __future__ import division import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class MSVAGHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of M-SVAG. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float beta = None # type: float eta = None # type: float weight_decay_rate = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: MSVAGHyperparameter # NOQA _default_hyperparam.lr = 0.1 _default_hyperparam.beta = 0.9 _default_hyperparam.eta = 1.0 _default_hyperparam.weight_decay_rate = 0 class MSVAGRule(optimizer.UpdateRule): """Update rule of the M-SVAG optimization algorithm. See: `Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients <https://arxiv.org/abs/1705.07774>`_ Modified for proper weight decay. See: `Fixing Weight Decay Regularization in Adam <https://openreview.net/forum?id=rk6qdGgCZ>`_ See :class:`~chainer.optimizers.MSVAG` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. beta (float): Exponential decay rate of the first and second order moment. eta (float): Schedule multiplier, can be used for warm restarts. weight_decay_rate (float): Weight decay rate. """ is_elementwise = True def __init__(self, parent_hyperparam=None, lr=None, beta=None, eta=None, weight_decay_rate=None): super(MSVAGRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if beta is not None: self.hyperparam.beta = beta if eta is not None: self.hyperparam.eta = eta if weight_decay_rate is not None: self.hyperparam.weight_decay_rate = weight_decay_rate self.beta_power = self.hyperparam.beta def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['m'] = xp.zeros_like(param.data) self.state['v'] = xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam m, v = self.state['m'], self.state['v'] rho = (((1.0 - hp.beta) ** 2) * (1.0 - self.beta_power ** 2) / (((1.0 - self.beta_power) ** 2) * (1.0 - hp.beta ** 2))) rho = min(rho, 0.9999) m += (1 - hp.beta) * (grad - m) v += (1 - hp.beta) * (grad * grad - v) mt = m / (1 - self.beta_power) vt = v / (1 - self.beta_power) mt2 = mt ** 2 s = (vt - mt2) / (1 - rho) factor = numpy.clip(mt2 / (mt2 + rho * s), 0, 1) if isinstance(factor, numpy.ndarray): factor[numpy.isnan(factor)] = 0 else: if numpy.isnan(factor): factor = 0 param.data -= hp.eta * (hp.lr * mt * factor + hp.weight_decay_rate * param.data) self.beta_power *= hp.beta def update_core_gpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam rho = (((1.0 - hp.beta) ** 2) * (1.0 - self.beta_power ** 2) / (((1.0 - self.beta_power) ** 2) * (1.0 - hp.beta ** 2))) rho = min(rho, 0.9999) cuda.elementwise( 'T grad, T lr, T one_minus_beta, T eta, \ T weight_decay_rate, T beta_power, T rho', 'T param, T m, T v', '''m += one_minus_beta * (grad - m); v += one_minus_beta * (grad * grad - v); T mt = m / (1.0 - beta_power); T vt = v / (1.0 - beta_power); T mt2 = mt*mt; T s = (vt - mt2) / (1.0 - rho); T factor; if (m == 0 && v == 0) factor = 0.0; else factor = min(1.0, max(0.0, mt2 / (mt2 + rho * s))); param -= eta * (lr * mt * factor + weight_decay_rate * param);''', 'msvag')(grad, hp.lr, 1 - hp.beta, hp.eta, hp.weight_decay_rate, self.beta_power, rho, param.data, self.state['m'], self.state['v']) self.beta_power *= hp.beta class MSVAG(optimizer.GradientMethod): """M-SVAG optimizer. See: `Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients <https://arxiv.org/abs/1705.07774>`_ Modified for proper weight decay (also called AdamW). AdamW introduces the additional parameters ``eta`` and ``weight_decay_rate``, which can be used to properly scale the learning rate, and decouple the weight decay rate from ``alpha``, as shown in the below paper. See: `Fixing Weight Decay Regularization in Adam <https://openreview.net/forum?id=rk6qdGgCZ>`_ Args: lr (float): Learning rate. beta (float): Exponential decay rate of the first and second order moment. eta (float): Schedule multiplier, can be used for warm restarts. weight_decay_rate (float): Weight decay rate. """ def __init__(self, lr=_default_hyperparam.lr, beta=_default_hyperparam.beta, eta=_default_hyperparam.eta, weight_decay_rate=_default_hyperparam.weight_decay_rate): super(MSVAG, self).__init__() self.hyperparam.lr = lr self.hyperparam.beta = beta self.hyperparam.eta = eta self.hyperparam.weight_decay_rate = weight_decay_rate lr = optimizer.HyperparameterProxy('lr') beta = optimizer.HyperparameterProxy('beta') eta = optimizer.HyperparameterProxy('eta') weight_decay_rate = optimizer.HyperparameterProxy('weight_decay_rate') def create_update_rule(self): return MSVAGRule(self.hyperparam)
6,356
31.433673
84
py
chainer
chainer-master/chainer/optimizers/ada_grad.py
import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class AdaGradHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of AdaGrad. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float eps = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: AdaGradHyperparameter # NOQA _default_hyperparam.lr = 0.001 _default_hyperparam.eps = 1e-8 class AdaGradRule(optimizer.UpdateRule): """Update rule of AdaGrad. See :class:`~chainer.optimizers.AdaGrad` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. eps (float): Small value for the numerical stability. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None, eps=None): super(AdaGradRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if eps is not None: self.hyperparam.eps = eps def init_state(self, param): with chainer.using_device(param.device): self.state['h'] = param.device.xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return lr = self.hyperparam.lr eps = self.hyperparam.eps h = self.state['h'] h += grad * grad param.data -= lr * grad / (numpy.sqrt(h) + eps) def update_core_gpu(self, param): grad = param.grad if grad is None: return if AdaGradRule._kernel is None: AdaGradRule._kernel = cuda.elementwise( 'T grad, T lr, T eps', 'T param, T h', '''h += grad * grad; param -= lr * grad / (sqrt(h) + eps);''', 'adagrad') AdaGradRule._kernel(grad, self.hyperparam.lr, self.hyperparam.eps, param.data, self.state['h']) class AdaGrad(optimizer.GradientMethod): """AdaGrad optimizer. See: http://jmlr.org/papers/v12/duchi11a.html Args: lr (float): Learning rate. eps (float): Small value for the numerical stability. """ def __init__(self, lr=_default_hyperparam.lr, eps=_default_hyperparam.eps): super(AdaGrad, self).__init__() self.hyperparam.lr = lr self.hyperparam.eps = eps lr = optimizer.HyperparameterProxy('lr') eps = optimizer.HyperparameterProxy('eps') def create_update_rule(self): return AdaGradRule(self.hyperparam)
2,899
26.619048
86
py
chainer
chainer-master/chainer/optimizers/adam.py
from __future__ import division import math import warnings import numpy import chainer from chainer.backends import cuda from chainer.backends import intel64 from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class AdamHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of Adam. This is only for PEP 544 compliant static type checkers. """ alpha = None # type: float beta1 = None # type: float beta2 = None # type: float eps = None # type: float eta = None # type: float weight_decay_rate = None # type: float amsgrad = None # type: bool adabound = None # type: bool final_lr = None # type: float gamma = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: AdamHyperparameter # NOQA _default_hyperparam.alpha = 0.001 _default_hyperparam.beta1 = 0.9 _default_hyperparam.beta2 = 0.999 _default_hyperparam.eps = 1e-8 _default_hyperparam.eta = 1.0 _default_hyperparam.weight_decay_rate = 0 _default_hyperparam.amsgrad = False _default_hyperparam.adabound = False _default_hyperparam.final_lr = 0.1 _default_hyperparam.gamma = 1e-3 def _learning_rate(hp, t): if t == 0: raise RuntimeError( 'Can\'t determine the learning rate of Adam optimizer ' 'because the update steps have not been started.') fix1 = 1. - math.pow(hp.beta1, t) fix2 = 1. - math.pow(hp.beta2, t) return hp.alpha * math.sqrt(fix2) / fix1 def _get_intermediate_dtype(dtype): # Returns the dtype for intermediate calculation. # For float16 input, float32 is used. # Otherwise the same dtype as the parameter is used. if dtype == numpy.float16: return numpy.float32 return dtype def _inplace_axpby(x, a, b, y): # in-place axpby: x = a * x + b * y if isinstance(x, intel64.mdarray): x.inplace_axpby(a, b, y) else: if a == 1: x += b * y else: x[...] = a * x + b * y class AdamRule(optimizer.UpdateRule): """Update rule of Adam optimization algorithm. See: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980v8>`_ Modified for proper weight decay. See: `Fixing Weight Decay Regularization in Adam <https://openreview.net/forum?id=rk6qdGgCZ>`_ With option to use AMSGrad variant of Adam. See: `On the Convergence of Adam and Beyond <https://openreview.net/forum?id=ryQu7f-RZ>`_ With option to use AdaBound variant of Adam. See: `Adaptive Gradient Methods with Dynamic Bound of Learning Rate <https://openreview.net/forum?id=Bkg3g2R9FX>` See :class:`~chainer.optimizers.Adam` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. weight_decay_rate (float): Weight decay rate. amsgrad (bool): Whether to use the AMSGrad variant of Adam. adabound (bool): Whether to use the AdaBound variant of Adam. final_lr (float): Final (SGD) learning rate in AdaBound. gamma (float): Convergence speed of the bound functions in AdaBound. """ is_elementwise = True _kernel = None _amsgrad_kernel = None _adabound_kernel = None _amsbound_kernel = None # Only used in `update_core_gpu`. # A dummy ndarray to help ElementwiseKernel deduce generic type T as # `dtype`. # It cannot be deduced only by scalar arguments. _dummy = None def __init__(self, parent_hyperparam=None, alpha=None, beta1=None, beta2=None, eps=None, eta=None, weight_decay_rate=None, amsgrad=None, adabound=None, final_lr=None, gamma=None): super(AdamRule, self).__init__( parent_hyperparam or _default_hyperparam) if alpha is not None: self.hyperparam.alpha = alpha if beta1 is not None: self.hyperparam.beta1 = beta1 if beta2 is not None: self.hyperparam.beta2 = beta2 if eps is not None: self.hyperparam.eps = eps if eta is not None: self.hyperparam.eta = eta if weight_decay_rate is not None: self.hyperparam.weight_decay_rate = weight_decay_rate if amsgrad is not None: self.hyperparam.amsgrad = amsgrad if adabound is not None: self.hyperparam.adabound = adabound if final_lr is not None: self.hyperparam.final_lr = final_lr if gamma is not None: self.hyperparam.gamma = gamma if self.hyperparam.adabound: self.initial_alpha = self.hyperparam.alpha def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['m'] = xp.zeros_like(param.data) self.state['v'] = xp.zeros_like(param.data) if self.hyperparam.amsgrad: self.state['vhat'] = xp.zeros_like(param.data) # For iDeep if isinstance(param.data, intel64.mdarray): self.state['m'] = intel64.ideep.array( self.state['m'], itype=intel64.ideep.wgt_array) self.state['v'] = intel64.ideep.array( self.state['v'], itype=intel64.ideep.wgt_array) if self.hyperparam.amsgrad: self.state['vhat'] = intel64.ideep.array( self.state['vhat'], itype=intel64.ideep.wgt_array) def _check_eps(self, interm_dtype): # Checks that the eps does not underflow. hp = self.hyperparam eps = interm_dtype(hp.eps) if hp.eps != 0 and eps == 0: raise ValueError( 'eps of Adam optimizer is too small for {} ({})'.format( interm_dtype.name, hp.eps)) # Note that the converted `eps` (numpy scalar) is discarded here and # the original `hp.eps` is used in calculation, because Python # scalars are faster in cupy elementwise kernels. def update_core_cpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam dtype = _get_intermediate_dtype(param.dtype.type) self._check_eps(dtype) grad = grad.astype(dtype, copy=False) m, v = self.state['m'], self.state['v'] # m += (1 - beta1) * (grad - m) _inplace_axpby(m, 1.0, 1.0 - hp.beta1, grad - m) # v += (1 - beta2) * (grad * grad - v) _inplace_axpby(v, 1.0, 1.0 - hp.beta2, grad*grad - v) if hp.amsgrad: vhat = self.state['vhat'] # For iDeep if isinstance(vhat, intel64.mdarray): vhat[...] = numpy.maximum(vhat, v) else: numpy.maximum(vhat, v, out=vhat) else: vhat = v vhat = vhat.astype(dtype, copy=False) step = self.alpha_t / (numpy.sqrt(vhat) + hp.eps) if hp.adabound: lower, upper = self.bounds step = numpy.clip(step, lower, upper) # param -= # eta * (step * m - weight_decay_rate * param) _inplace_axpby( param.data, 1.0 - hp.eta * hp.weight_decay_rate, -hp.eta, step * m) def update_core_gpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam dtype = _get_intermediate_dtype(param.dtype.type) self._check_eps(dtype) if self._dummy is None: self._dummy = cuda.cupy.empty((0,), dtype=dtype) if hp.adabound: lower, upper = self.bounds if hp.amsgrad and hp.adabound: if AdamRule._amsbound_kernel is None: AdamRule._amsbound_kernel = cuda.elementwise( 'P grad, T alpha_t, T one_minus_beta1, T one_minus_beta2, ' 'T lower, T upper, ' 'T eps, T eta, T weight_decay_rate, raw T dummy', 'P param, P m, P v, P vhat', '''T grad_ = static_cast<T>(grad); T m_ = static_cast<T>(m); T v_ = static_cast<T>(v); T vhat_ = static_cast<T>(vhat); m_ += one_minus_beta1 * (grad_ - m_); v_ += one_minus_beta2 * (grad_ * grad_ - v_); vhat_ = max(vhat_, v_); vhat = static_cast<T>(vhat_); m = static_cast<P>(m_); v = static_cast<P>(v_); param -= eta * (max(min(alpha_t / (sqrt(vhat_) + eps), upper), lower) * m_ + weight_decay_rate * param);''', 'amsbound') AdamRule._amsbound_kernel( grad, self.alpha_t, 1 - hp.beta1, 1 - hp.beta2, lower, upper, hp.eps, hp.eta, hp.weight_decay_rate, self._dummy, param.data, self.state['m'], self.state['v'], self.state['vhat']) elif hp.adabound: if AdamRule._adabound_kernel is None: AdamRule._adabound_kernel = cuda.elementwise( 'P grad, T alpha_t, T one_minus_beta1, T one_minus_beta2, ' 'T lower, T upper, ' 'T eps, T eta, T weight_decay_rate, raw T dummy', 'P param, P m, P v', '''T grad_ = static_cast<T>(grad); T m_ = static_cast<T>(m); T v_ = static_cast<T>(v); m_ += one_minus_beta1 * (grad_ - m_); v_ += one_minus_beta2 * (grad_ * grad_ - v_); m = static_cast<P>(m_); v = static_cast<P>(v_); param -= eta * (max(min(alpha_t / (sqrt(v_) + eps), upper), lower) * m_ + weight_decay_rate * param);''', 'adabound') AdamRule._adabound_kernel( grad, self.alpha_t, 1 - hp.beta1, 1 - hp.beta2, lower, upper, hp.eps, hp.eta, hp.weight_decay_rate, self._dummy, param.data, self.state['m'], self.state['v']) elif hp.amsgrad: if AdamRule._amsgrad_kernel is None: AdamRule._amsgrad_kernel = cuda.elementwise( 'P grad, T alpha_t, T one_minus_beta1, T one_minus_beta2, ' 'T eps, T eta, T weight_decay_rate, raw T dummy', 'P param, P m, P v, P vhat', '''T grad_ = static_cast<T>(grad); T m_ = static_cast<T>(m); T v_ = static_cast<T>(v); T vhat_ = static_cast<T>(vhat); m_ += one_minus_beta1 * (grad_ - m_); v_ += one_minus_beta2 * (grad_ * grad_ - v_); vhat_ = max(vhat_, v_); vhat = static_cast<T>(vhat_); m = static_cast<P>(m_); v = static_cast<P>(v_); param -= eta * (alpha_t * m_ / (sqrt(vhat_) + eps) + weight_decay_rate * param);''', 'adam') AdamRule._amsgrad_kernel( grad, self.alpha_t, 1 - hp.beta1, 1 - hp.beta2, hp.eps, hp.eta, hp.weight_decay_rate, self._dummy, param.data, self.state['m'], self.state['v'], self.state['vhat']) else: if AdamRule._kernel is None: AdamRule._kernel = cuda.elementwise( 'P grad, T alpha_t, T one_minus_beta1, T one_minus_beta2, ' 'T eps, T eta, T weight_decay_rate, raw T dummy', 'P param, P m, P v', '''T grad_ = static_cast<T>(grad); T m_ = static_cast<T>(m); T v_ = static_cast<T>(v); m_ += one_minus_beta1 * (grad_ - m_); v_ += one_minus_beta2 * (grad_ * grad_ - v_); m = static_cast<P>(m_); v = static_cast<P>(v_); param -= eta * (alpha_t * m_ / (sqrt(v_) + eps) + weight_decay_rate * param);''', 'adam') AdamRule._kernel( grad, self.alpha_t, 1 - hp.beta1, 1 - hp.beta2, hp.eps, hp.eta, hp.weight_decay_rate, self._dummy, param.data, self.state['m'], self.state['v']) @property def alpha_t(self): return _learning_rate(self.hyperparam, self.t) @property def lr(self): warnings.warn( 'AdamRule.lr has been renamed to AdamRule.alpha_t. ' 'Use of AdamRule.lr is deprecated in Chainer v6.', DeprecationWarning) return self.alpha_t @property def bounds(self): if self.t == 0: raise RuntimeError( 'Can\'t determine the bounds of AdaBound optimizer ' 'because the update steps have not been started.') hp = self.hyperparam # Workaround to reflect changing `alpha` in `final_lr`. # (by some of `chainer.training.extensions`) final_lr = hp.final_lr * hp.alpha / self.initial_alpha lower = final_lr * (1.0 - 1.0 / (hp.gamma * self.t + 1)) upper = final_lr * (1.0 + 1.0 / (hp.gamma * self.t)) return lower, upper class Adam(optimizer.GradientMethod): """Adam optimizer. See: `Adam: A Method for Stochastic Optimization <https://arxiv.org/abs/1412.6980v8>`_ Modified for proper weight decay (also called :class:`~chainer.optimizers.AdamW`). AdamW introduces the additional parameters ``eta`` and ``weight_decay_rate``, which can be used to properly scale the learning rate, and decouple the weight decay rate from ``alpha``, as shown in the below paper. Note that with the default values ``eta = 1`` and ``weight_decay_rate = 0``, this implementation is identical to the standard Adam method. See: `Fixing Weight Decay Regularization in Adam <https://openreview.net/forum?id=rk6qdGgCZ>`_ A flag ``amsgrad`` to use the :class:`~chainer.optimizers.AMSGrad` variant of Adam from the paper: `On the Convergence of Adam and Beyond <https://openreview.net/forum?id=ryQu7f-RZ>`_ A flag ``adabound`` to use the :class:`~chainer.optimizers.AdaBound` variant of Adam from the paper: `Adaptive Gradient Methods with Dynamic Bound of Learning Rate <https://openreview.net/forum?id=Bkg3g2R9FX>`_ If both ``amsgrad`` and ``adabound`` are ``True``, the optimizer is equivalent to :class:`~chainer.optimizers.AMSBound` proposed in the AdaBound paper. Args: alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. weight_decay_rate (float): Weight decay rate. amsgrad (bool): Whether to use AMSGrad variant of Adam. adabound (bool): Whether to use the AdaBound variant of Adam. final_lr (float): Final (SGD) learning rate in AdaBound. gamma (float): Convergence speed of the bound functions in AdaBound. """ def __init__(self, alpha=_default_hyperparam.alpha, beta1=_default_hyperparam.beta1, beta2=_default_hyperparam.beta2, eps=_default_hyperparam.eps, eta=_default_hyperparam.eta, weight_decay_rate=_default_hyperparam.weight_decay_rate, amsgrad=_default_hyperparam.amsgrad, adabound=_default_hyperparam.adabound, final_lr=_default_hyperparam.final_lr, gamma=_default_hyperparam.gamma): super(Adam, self).__init__() self.hyperparam.alpha = alpha self.hyperparam.beta1 = beta1 self.hyperparam.beta2 = beta2 self.hyperparam.eps = eps self.hyperparam.eta = eta self.hyperparam.weight_decay_rate = weight_decay_rate self.hyperparam.amsgrad = amsgrad self.hyperparam.adabound = adabound self.hyperparam.final_lr = final_lr self.hyperparam.gamma = gamma alpha = optimizer.HyperparameterProxy('alpha') beta1 = optimizer.HyperparameterProxy('beta1') beta2 = optimizer.HyperparameterProxy('beta2') eps = optimizer.HyperparameterProxy('eps') eta = optimizer.HyperparameterProxy('eta') weight_decay_rate = optimizer.HyperparameterProxy('weight_decay_rate') amsgrad = optimizer.HyperparameterProxy('amsgrad') adabound = optimizer.HyperparameterProxy('adabound') final_lr = optimizer.HyperparameterProxy('final_lr') gamma = optimizer.HyperparameterProxy('gamma') def create_update_rule(self): return AdamRule(self.hyperparam) @property def alpha_t(self): return _learning_rate(self.hyperparam, self.t) @property def lr(self): warnings.warn( 'Adam.lr has been renamed to AdamRule.alpha_t. ' 'Use of Adam.lr is deprecated in Chainer v6.', DeprecationWarning) return self.alpha_t class AdamW(Adam): """AdamW optimizer. This class is a special case of :class:`~chainer.optimizers.Adam`. See: `Fixing Weight Decay Regularization in Adam <https://openreview.net/forum?id=rk6qdGgCZ>`_ Args: alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. The default value is 1.0. weight_decay_rate (float): Weight decay rate. The default value is 0. """ def __init__(self, alpha=_default_hyperparam.alpha, beta1=_default_hyperparam.beta1, beta2=_default_hyperparam.beta2, eps=_default_hyperparam.eps, eta=_default_hyperparam.eta, weight_decay_rate=_default_hyperparam.weight_decay_rate): super(AdamW, self).__init__( alpha=alpha, beta1=beta1, beta2=beta2, eps=eps, eta=eta, weight_decay_rate=weight_decay_rate) class AMSGrad(Adam): """AMSGrad optimizer. This class is a special case of :class:`~chainer.optimizers.Adam`. See: `On the Convergence of Adam and Beyond <https://openreview.net/forum?id=ryQu7f-RZ>`_ Args: alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. """ def __init__(self, alpha=_default_hyperparam.alpha, beta1=_default_hyperparam.beta1, beta2=_default_hyperparam.beta2, eps=_default_hyperparam.eps, eta=_default_hyperparam.eta): super(AMSGrad, self).__init__( alpha=alpha, beta1=beta1, beta2=beta2, eps=eps, eta=eta, amsgrad=True) class AdaBound(Adam): """AdaBound optimizer. This class is a special case of :class:`~chainer.optimizers.Adam`. See: `Adaptive Gradient Methods with Dynamic Bound of Learning Rate <https://openreview.net/forum?id=Bkg3g2R9FX>`_ Args: alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. final_lr (float): Final (SGD) learning rate in AdaBound. gamma (float): Convergence speed of the bound functions in AdaBound. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. """ def __init__(self, alpha=_default_hyperparam.alpha, beta1=_default_hyperparam.beta1, beta2=_default_hyperparam.beta2, final_lr=_default_hyperparam.final_lr, gamma=_default_hyperparam.gamma, eps=_default_hyperparam.eps, eta=_default_hyperparam.eta): super(AdaBound, self).__init__( alpha=alpha, beta1=beta1, beta2=beta2, eps=eps, eta=eta, amsgrad=False, adabound=True, final_lr=final_lr, gamma=gamma) class AMSBound(Adam): """AMSBound optimizer. This class is a special case of :class:`~chainer.optimizers.Adam`. See: `Adaptive Gradient Methods with Dynamic Bound of Learning Rate <https://openreview.net/forum?id=Bkg3g2R9FX>`_ Args: alpha (float): Coefficient of learning rate. beta1 (float): Exponential decay rate of the first order moment. beta2 (float): Exponential decay rate of the second order moment. final_lr (float): Final (SGD) learning rate in AdaBound. gamma (float): Convergence speed of the bound functions in AdaBound. eps (float): Small value for the numerical stability. eta (float): Schedule multiplier, can be used for warm restarts. """ def __init__(self, alpha=_default_hyperparam.alpha, beta1=_default_hyperparam.beta1, beta2=_default_hyperparam.beta2, final_lr=_default_hyperparam.final_lr, gamma=_default_hyperparam.gamma, eps=_default_hyperparam.eps, eta=_default_hyperparam.eta): super(AMSBound, self).__init__( alpha=alpha, beta1=beta1, beta2=beta2, eps=eps, eta=eta, amsgrad=True, adabound=True, final_lr=final_lr, gamma=gamma)
22,884
38.321306
83
py
chainer
chainer-master/chainer/optimizers/__init__.py
# import classes and functions from chainer.optimizers.ada_delta import AdaDelta # NOQA from chainer.optimizers.ada_grad import AdaGrad # NOQA from chainer.optimizers.adam import Adam # NOQA from chainer.optimizers.adam import AdamW # NOQA from chainer.optimizers.adam import AMSGrad # NOQA from chainer.optimizers.adam import AdaBound # NOQA from chainer.optimizers.adam import AMSBound # NOQA from chainer.optimizers.corrected_momentum_sgd import CorrectedMomentumSGD # NOQA from chainer.optimizers.momentum_sgd import MomentumSGD # NOQA from chainer.optimizers.msvag import MSVAG # NOQA from chainer.optimizers.nesterov_ag import NesterovAG # NOQA from chainer.optimizers.rmsprop import RMSprop # NOQA from chainer.optimizers.rmsprop_graves import RMSpropGraves # NOQA from chainer.optimizers.sgd import SGD # NOQA from chainer.optimizers.smorms3 import SMORMS3 # NOQA
887
51.235294
82
py
chainer
chainer-master/chainer/optimizers/ada_delta.py
import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class AdaDeltaHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of Zeiler's ADADELTA. This is only for PEP 544 compliant static type checkers. """ rho = None # type: float eps = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: AdaDeltaHyperparameter # NOQA _default_hyperparam.rho = 0.95 _default_hyperparam.eps = 1e-6 class AdaDeltaRule(optimizer.UpdateRule): """Update rule of Zeiler's ADADELTA. See :class:`~chainer.optimizers.AdaDelta` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. rho (float): Exponential decay rate of the first and second order moments. eps (float): Small value for the numerical stability. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, rho=None, eps=None): super(AdaDeltaRule, self).__init__( parent_hyperparam or _default_hyperparam) if rho is not None: self.hyperparam.rho = rho if eps is not None: self.hyperparam.eps = eps def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['msg'] = xp.zeros_like(param.data) self.state['msdx'] = xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return msg, msdx = self.state['msg'], self.state['msdx'] rho = self.hyperparam.rho eps = self.hyperparam.eps msg *= rho msg += (1 - rho) * grad * grad dx = numpy.sqrt((msdx + eps) / (msg + eps)) * grad msdx *= rho msdx += (1 - rho) * dx * dx param.data -= dx def update_core_gpu(self, param): grad = param.grad if grad is None: return if AdaDeltaRule._kernel is None: AdaDeltaRule._kernel = cuda.elementwise( 'T grad, T one_minus_rho, T eps', 'T param, T msg, T msdx', '''msg = msg + one_minus_rho * (grad * grad - msg); T dx = sqrt((msdx + eps) / (msg + eps)) * grad; msdx += one_minus_rho * (dx * dx - msdx); param -= dx;''', 'adadelta') AdaDeltaRule._kernel( grad, 1 - self.hyperparam.rho, self.hyperparam.eps, param.data, self.state['msg'], self.state['msdx']) class AdaDelta(optimizer.GradientMethod): """Zeiler's ADADELTA. See: http://www.matthewzeiler.com/pubs/googleTR2012/googleTR2012.pdf Args: rho (float): Exponential decay rate of the first and second order moments. eps (float): Small value for the numerical stability. """ def __init__(self, rho=_default_hyperparam.rho, eps=_default_hyperparam.eps): super(AdaDelta, self).__init__() self.hyperparam.rho = rho self.hyperparam.eps = eps rho = optimizer.HyperparameterProxy('rho') eps = optimizer.HyperparameterProxy('eps') def create_update_rule(self): return AdaDeltaRule(self.hyperparam)
3,514
29.301724
87
py
chainer
chainer-master/chainer/optimizers/rmsprop_graves.py
import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class RMSpropGravesHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of Alex Graves's RMSprop. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float alpha = None # type: float momentum = None # type: float eps = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: RMSpropGravesHyperparameter # NOQA _default_hyperparam.lr = 1e-4 _default_hyperparam.alpha = 0.95 _default_hyperparam.momentum = 0.9 _default_hyperparam.eps = 1e-4 class RMSpropGravesRule(optimizer.UpdateRule): """Update rule for Alex Graves's RMSprop. See :class:`~chainer.optimizers.RMSpropGraves` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. alpha (float): Exponential decay rate of the first and second order moments of the raw gradient. momentum (float): Exponential decay rate of the first order moment of the adjusted gradient. eps (float): Small value for the numerical stability. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None, alpha=None, momentum=None, eps=None): super(RMSpropGravesRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if alpha is not None: self.hyperparam.alpha = alpha if momentum is not None: self.hyperparam.momentum = momentum if eps is not None: self.hyperparam.eps = eps def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['n'] = xp.zeros_like(param.data) self.state['g'] = xp.zeros_like(param.data) self.state['delta'] = xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return n, g, delta = self.state['n'], self.state['g'], self.state['delta'] hp = self.hyperparam n *= hp.alpha n += (1 - hp.alpha) * grad * grad g *= hp.alpha g += (1 - hp.alpha) * grad delta *= hp.momentum delta -= hp.lr * grad / numpy.sqrt(n - g * g + hp.eps) param.data += delta def update_core_gpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam if RMSpropGravesRule._kernel is None: RMSpropGravesRule._kernel = cuda.elementwise( 'T grad, T lr, T alpha, T momentum, T eps', 'T param, T avg_n, T avg_g, T delta', '''avg_n = alpha * avg_n + (1 - alpha) * grad * grad; avg_g = alpha * avg_g + (1 - alpha) * grad; delta = delta * momentum - lr * grad * rsqrt(avg_n - avg_g * avg_g + eps); param += delta;''', 'rmsprop_graves') RMSpropGravesRule._kernel( grad, hp.lr, hp.alpha, hp.momentum, hp.eps, param.data, self.state['n'], self.state['g'], self.state['delta']) class RMSpropGraves(optimizer.GradientMethod): """Alex Graves's RMSprop. See: https://arxiv.org/abs/1308.0850 Args: lr (float): Learning rate. alpha (float): Exponential decay rate of the first and second order moments of the raw gradient. momentum (float): Exponential decay rate of the first order moment of the adjusted gradient. eps (float): Small value for the numerical stability. """ def __init__(self, lr=_default_hyperparam.lr, alpha=_default_hyperparam.alpha, momentum=_default_hyperparam.momentum, eps=_default_hyperparam.eps): super(RMSpropGraves, self).__init__() self.hyperparam.lr = lr self.hyperparam.alpha = alpha self.hyperparam.momentum = momentum self.hyperparam.eps = eps lr = optimizer.HyperparameterProxy('lr') alpha = optimizer.HyperparameterProxy('alpha') momentum = optimizer.HyperparameterProxy('momentum') eps = optimizer.HyperparameterProxy('eps') def create_update_rule(self): return RMSpropGravesRule(self.hyperparam)
4,709
32.642857
92
py
chainer
chainer-master/chainer/optimizers/rmsprop.py
import numpy import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class RMSpropHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of RMSprop. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float alpha = None # type: float eps = None # type: float eps_inside_sqrt = None # type: bool _default_hyperparam = optimizer.Hyperparameter() # type: RMSpropHyperparameter # NOQA _default_hyperparam.lr = 0.01 _default_hyperparam.alpha = 0.99 _default_hyperparam.eps = 1e-8 _default_hyperparam.eps_inside_sqrt = False class RMSpropRule(optimizer.UpdateRule): """Update rule for RMSprop. See :class:`~chainer.optimizers.RMSprop` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. alpha (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eps_inside_sqrt (bool): When ``True``, gradient will be divided by :math:`\\sqrt{ms + eps}` where ``ms`` is the mean square. When ``False`` (default), gradient will be divided by :math:`\\sqrt{ms} + eps` instead. This option may be convenient for users porting code from other frameworks; see `#4754 <https://github.com/chainer/chainer/issues/4754>`__ for details. """ is_elementwise = True def __init__(self, parent_hyperparam=None, lr=None, alpha=None, eps=None, eps_inside_sqrt=None): super(RMSpropRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if alpha is not None: self.hyperparam.alpha = alpha if eps is not None: self.hyperparam.eps = eps if eps_inside_sqrt is not None: self.hyperparam.eps_inside_sqrt = eps_inside_sqrt def init_state(self, param): with chainer.using_device(param.device): self.state['ms'] = param.device.xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam eps = grad.dtype.type(hp.eps) if hp.eps != 0 and eps == 0: raise ValueError( 'eps of RMSprop optimizer is too small for {} ({})'.format( grad.dtype.name, hp.eps)) ms = self.state['ms'] ms *= hp.alpha ms += (1 - hp.alpha) * grad * grad if hp.eps_inside_sqrt: denom = numpy.sqrt(ms + eps) else: denom = numpy.sqrt(ms) + eps param.data -= hp.lr * grad / denom def update_core_gpu(self, param): grad = param.grad if grad is None: return hp = self.hyperparam eps = grad.dtype.type(hp.eps) if eps == 0: raise ValueError( 'eps of RMSprop optimizer is too small for {} ({})'.format( grad.dtype.name, hp.eps)) if hp.eps_inside_sqrt: denom = 'sqrt(ms + eps)' else: denom = 'sqrt(ms) + eps' kernel = cuda.elementwise( 'T grad, T lr, T alpha, T eps', 'T param, T ms', '''ms = alpha * ms + (1 - alpha) * grad * grad; param -= lr * grad / ({});'''.format(denom), 'rmsprop') kernel(grad, self.hyperparam.lr, self.hyperparam.alpha, eps, param.data, self.state['ms']) class RMSprop(optimizer.GradientMethod): """RMSprop optimizer. See: T. Tieleman and G. Hinton (2012). Lecture 6.5 - rmsprop, COURSERA: Neural Networks for Machine Learning. Args: lr (float): Learning rate. alpha (float): Exponential decay rate of the second order moment. eps (float): Small value for the numerical stability. eps_inside_sqrt (bool): When ``True``, gradient will be divided by :math:`\\sqrt{ms + eps}` where ``ms`` is the mean square. When ``False`` (default), gradient will be divided by :math:`\\sqrt{ms} + eps` instead. This option may be convenient for users porting code from other frameworks; see `#4754 <https://github.com/chainer/chainer/issues/4754>`__ for details. """ def __init__(self, lr=_default_hyperparam.lr, alpha=_default_hyperparam.alpha, eps=_default_hyperparam.eps, eps_inside_sqrt=_default_hyperparam.eps_inside_sqrt): super(RMSprop, self).__init__() self.hyperparam.lr = lr self.hyperparam.alpha = alpha self.hyperparam.eps = eps self.hyperparam.eps_inside_sqrt = eps_inside_sqrt lr = optimizer.HyperparameterProxy('lr') alpha = optimizer.HyperparameterProxy('alpha') eps = optimizer.HyperparameterProxy('eps') eps_inside_sqrt = optimizer.HyperparameterProxy('eps_inside_sqrt') def create_update_rule(self): return RMSpropRule(self.hyperparam)
5,381
33.948052
86
py
chainer
chainer-master/chainer/optimizers/nesterov_ag.py
import chainer from chainer.backends import cuda from chainer import optimizer from chainer import types if types.TYPE_CHECKING: import typing_extensions as tpe class NesterovAGHyperparameter(tpe.Protocol): """Protocol class for hyperparameter of Nesterov's Accelerated Gradient. This is only for PEP 544 compliant static type checkers. """ lr = None # type: float momentum = None # type: float _default_hyperparam = optimizer.Hyperparameter() # type: NesterovAGHyperparameter # NOQA _default_hyperparam.lr = 0.01 _default_hyperparam.momentum = 0.9 class NesterovAGRule(optimizer.UpdateRule): """Update rule for Nesterov's Accelerated Gradient. See :class:`~chainer.optimizers.NesterovAG` for the default values of the hyperparameters. Args: parent_hyperparam (~chainer.optimizer.Hyperparameter): Hyperparameter that provides the default values. lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ is_elementwise = True _kernel = None def __init__(self, parent_hyperparam=None, lr=None, momentum=None): super(NesterovAGRule, self).__init__( parent_hyperparam or _default_hyperparam) if lr is not None: self.hyperparam.lr = lr if momentum is not None: self.hyperparam.momentum = momentum def init_state(self, param): with chainer.using_device(param.device): xp = param.device.xp self.state['v'] = xp.zeros_like(param.data) def update_core_cpu(self, param): grad = param.grad if grad is None: return v = self.state['v'] lr, momentum = self.hyperparam.lr, self.hyperparam.momentum v *= momentum v -= lr * grad param.data += momentum * momentum * v param.data -= (1 + momentum) * lr * grad def update_core_gpu(self, param): grad = param.grad if grad is None: return if NesterovAGRule._kernel is None: NesterovAGRule._kernel = cuda.elementwise( 'T grad, T lr, T momentum', 'T param, T v', ''' v = v * momentum - lr * grad; param += momentum * momentum * v - (1 + momentum) * lr * grad; ''', 'nesterov_ag') NesterovAGRule._kernel( grad, self.hyperparam.lr, self.hyperparam.momentum, param.data, self.state['v']) class NesterovAG(optimizer.GradientMethod): """Nesterov's Accelerated Gradient. See: https://arxiv.org/abs/1212.0901 Args: lr (float): Learning rate. momentum (float): Exponential decay rate of the first order moment. """ def __init__(self, lr=_default_hyperparam.lr, momentum=_default_hyperparam.momentum): super(NesterovAG, self).__init__() self.hyperparam.lr = lr self.hyperparam.momentum = momentum lr = optimizer.HyperparameterProxy('lr') momentum = optimizer.HyperparameterProxy('momentum') def create_update_rule(self): return NesterovAGRule(self.hyperparam)
3,234
28.953704
89
py
chainer
chainer-master/chainer/function_hooks/cuda_profile.py
from chainer.backends import cuda from chainer import function_hook class CUDAProfileHook(function_hook.FunctionHook): name = 'CUDAProfileHook' def __init__(self): cuda.check_cuda_available() if not cuda.cupy.cuda.nvtx_enabled: raise RuntimeError('nvtx is required for CUDAProfileHook') def forward_preprocess(self, function, in_data): cuda.cupy.cuda.nvtx.RangePush(function.label + '.forward') def forward_postprocess(self, function, in_data): cuda.cupy.cuda.nvtx.RangePop() def backward_preprocess(self, function, in_data, out_grad): cuda.cupy.cuda.nvtx.RangePush(function.label + '.backward') def backward_postprocess(self, function, in_data, out_grad): cuda.cupy.cuda.nvtx.RangePop()
780
30.24
70
py
chainer
chainer-master/chainer/function_hooks/timer.py
import os import sys import time import numpy from chainer import backend from chainer.backends import cuda from chainer import function_hook # Select the best-resolution timer function try: _get_time = time.perf_counter except AttributeError: if os.name == 'nt': _get_time = time.clock else: _get_time = time.time class TimerHook(function_hook.FunctionHook): """Function hook for measuring elapsed time of functions. Example: Code example:: from chainer.function_hooks import TimerHook hook = TimerHook() with hook: trainer.run() hook.print_report() Output example:: FunctionName ElapsedTime Occurrence LinearFunction 1.24sec 3900 ReLU 0.59sec 2600 SoftmaxCrossEntropy 0.82sec 1300 Accuracy 0.18sec 700 where *FunctionName* is the name of function that calls the hook, and *ElapsedTime* is the elapsed time the function consumed, and *Occurrence* is the number of calls. Attributes: call_history: List of measurement results. It consists of pairs of the name of the function that calls this hook and the elapsed time the function consumes. """ name = 'TimerHook' table = {'sec': 1, 'ms': 10 ** 3, 'us': 10 ** 6, 'ns': 10 ** 9} def __init__(self): self.call_history = [] self._running_stack = [] self._depth = 0 self._total_time = 0 def _preprocess(self): if self.xp == numpy: start = _get_time() self._running_stack.append(start) else: start = cuda.Event() stop = cuda.Event() start.record() self._running_stack.append((start, stop)) self._depth += 1 def forward_preprocess(self, function, in_data): self.xp = backend.get_array_module(*in_data) self._preprocess() def backward_preprocess(self, function, in_data, out_grad): self.xp = backend.get_array_module(*(in_data + out_grad)) self._preprocess() def _postprocess(self, function): if self.xp == numpy: start = self._running_stack.pop() stop = _get_time() elapsed_time = stop - start else: start, stop = self._running_stack.pop() stop.record() stop.synchronize() # Note that `get_elapsed_time` returns result in milliseconds elapsed_time = cuda.cupy.cuda.get_elapsed_time( start, stop) / 1000 self.call_history.append((function._impl_name, elapsed_time)) assert self._depth > 0 self._depth -= 1 if self._depth == 0: self._total_time += elapsed_time def forward_postprocess(self, function, in_data): xp = backend.get_array_module(*in_data) assert xp == self.xp self._postprocess(function) def backward_postprocess(self, function, in_data, out_grad): xp = backend.get_array_module(*(in_data + out_grad)) assert xp == self.xp self._postprocess(function) def total_time(self): """Returns total elapsed time in seconds.""" return self._total_time def summary(self): """Returns a summary of time profiling in functions. Returns: A summarized dictionary whose keys are function names and values are dictionaries of `elapsed_time` and `occurrence`. """ summary = {} for function_name, elapsed_time in self.call_history: if function_name not in summary: summary[function_name] = {'elapsed_time': 0, 'occurrence': 0} record = summary[function_name] record['elapsed_time'] += elapsed_time record['occurrence'] += 1 return summary def _choose_unit(self, second): """Choose optimal unit.""" factor = 1 for unit in ['sec', 'ms', 'us']: if second * factor >= 1: return factor, unit factor *= 1000.0 return factor, 'ns' def print_report(self, unit='auto', file=sys.stdout): """Prints a summary report of time profiling in functions. Args: unit (str): Supplementary units used for computational times. `sec`, `ms`, `us`, `ns`, `auto`(default) and `auto_foreach` are supported. If `auto`, units of times are aligned to the largest, and if `auto_foreach`, units of times are adjusted for each element. """ entries = [['FunctionName', 'ElapsedTime', 'Occurrence']] auto_foreach = (unit == 'auto_foreach') if unit == 'auto': max_time = max( record['elapsed_time'] for record in self.summary().values()) factor, unit = self._choose_unit(max_time) elif unit != 'auto_foreach': factor = self.table[unit] for function_name, record in self.summary().items(): second = record['elapsed_time'] if auto_foreach: factor, unit = self._choose_unit(second) elapsed_time = '%3.2f%s' % (second * factor, unit) occurrence = str(record['occurrence']) entries.append([function_name, elapsed_time, occurrence]) entry_widths = [] entry_widths.append(max(len(f) for f, _, _ in entries)) entry_widths.append(max(len(e) for _, e, _ in entries)) entry_widths.append(max(len(o) for _, _, o in entries)) template = ' '.join('{:>%d}' % w for w in entry_widths) for function_name, elapsed_time, occurrence in entries: line = template.format(function_name, elapsed_time, occurrence) file.write(line) file.write('\n') if hasattr(file, 'flush'): file.flush()
6,045
33.947977
79
py
chainer
chainer-master/chainer/function_hooks/cupy_memory_profile.py
import collections import sys import typing as tp # NOQA from chainer.backends import cuda from chainer import function_hook try: MemoryHook = cuda.cupy.cuda.memory_hook.MemoryHook # type: tp.Any # to handle https://github.com/python/mypy/issues/2477 # NOQA memory_hook_available = True except Exception as e: _resolution_error = e MemoryHook = object memory_hook_available = False class CupyMemoryProfileHook(function_hook.FunctionHook): """Function hook for measuring memory usage of functions in cupy memory pool. Example: Code example:: from chainer.function_hooks import CupyMemoryProfileHook hook = CupyMemoryProfileHook() with hook: trainer.run() hook.print_report() Output example:: FunctionName UsedBytes AcquiredBytes Occurrence LinearFunction 5.16GB 179.98MB 3900 ReLU 0.99GB 458.97MB 2600 SoftmaxCrossEntropy 0.01GB 5.08MB 1300 Accuracy 0.00GB 0.35MB 700 where *FunctionName* is the name of function that calls the hook, and *UsedBytes* is the memory bytes the function used from cupy memory pool, and *AcquiredBytes* is the actual memory bytes the cupy memory pool acquired from GPU device on the function call, and *Occurrence* is the number of calls. Attributes: call_history: List of measurement results. It consists of the name of the function that calls this hook, the memory bytes the function used from cupy memory pool, and the memory bytes the cupy memory pool acquired from GPU device on the function call. """ name = 'CupyMemoryProfileHook' _units = ['B', 'KB', 'MB', 'GB', 'TB', 'PB', 'EB', 'ZB'] _table = {u: 1024.0 ** i for i, u in enumerate(_units)} def __init__(self): cuda.check_cuda_available() if not memory_hook_available: msg = 'CuPy >= 2.0 is required. %s' % str(_resolution_error) raise RuntimeError(msg) self.call_history = [] self._memory_hook = CupyMemoryCumulativeHook() self._running_stack = [] self._total_used_bytes = 0 self._total_acquired_bytes = 0 def added(self, function=None): self._memory_hook.__enter__() def deleted(self, function=None): self._memory_hook.__exit__() def _preprocess(self): start_used_bytes = self._memory_hook.used_bytes start_acquired_bytes = self._memory_hook.acquired_bytes self._running_stack.append((start_used_bytes, start_acquired_bytes)) def forward_preprocess(self, function, in_data): self._preprocess() def backward_preprocess(self, function, in_data, out_grad): self._preprocess() def _postprocess(self, function): start_used_bytes, start_acquired_bytes = self._running_stack.pop() end_used_bytes = self._memory_hook.used_bytes end_acquired_bytes = self._memory_hook.acquired_bytes used_bytes = end_used_bytes - start_used_bytes acquired_bytes = end_acquired_bytes - start_acquired_bytes depth = len(self._running_stack) self.call_history.append( (function._impl_name, used_bytes, acquired_bytes, depth)) if depth == 0: self._total_used_bytes += used_bytes self._total_acquired_bytes += acquired_bytes def forward_postprocess(self, function, in_data): self._postprocess(function) def backward_postprocess(self, function, in_data, out_grad): self._postprocess(function) def total_used_bytes(self): """Returns total bytes that functions used from cupy memory pool.""" return self._total_used_bytes def total_acquired_bytes(self): """Returns total bytes that cupy memory pool acquired from GPU.""" return self._total_acquired_bytes def summary(self): """Returns a summary of memory profiling in functions. Returns: A summarized dictionary whose keys are function names and values are dictionaries of ``used_bytes``, ``acquired_bytes``, and ``occurrrence``. """ # TODO(sonots): PROBLEM: takes count of nested functions duplicately summary = collections.OrderedDict() for func_name, used_bytes, acquired_bytes, depth in self.call_history: if func_name not in summary: summary[func_name] = {'used_bytes': 0, 'acquired_bytes': 0, 'occurrence': 0} record = summary[func_name] record['used_bytes'] += used_bytes record['acquired_bytes'] += acquired_bytes record['occurrence'] += 1 return summary def _choose_unit(self, size): """Choose optimal unit. Returns: Tuple of denomi (float) and human-readable unit (str). """ denomi = 1.0 if size <= 0: return denomi, self._units[0] for unit in self._units[:-1]: if size / (denomi * 1024) < 1: return denomi, unit denomi *= 1024 return denomi, self._units[-1] def print_report(self, unit='auto', file=sys.stdout): """Prints a summary report of memory profiling in functions. Args: unit (str): Supplementary units used for used memories. `B`, `KB`, `MB`, `GB`, `TB`, `PB`, `EB`, `ZB`, `auto`(default) and `auto_foreach` are supported. If `auto`, units of memories are aligned to the largest values of 'used_bytes' and 'acquired_bytes'. If `auto_foreach`, units of memories are adjusted for each element. """ entries = [[ 'FunctionName', 'UsedBytes', 'AcquiredBytes', 'Occurrence']] if unit == 'auto': max_used = max( record['used_bytes'] for record in self.summary().values()) max_acquired = max( record['acquired_bytes'] for record in self.summary().values()) denomi_used, unit_used = self._choose_unit(max_used) denomi_acquired, unit_acquired = self._choose_unit(max_acquired) elif unit != 'auto_foreach': denomi_used = denomi_acquired = self._table[unit] unit_used = unit_acquired = unit for function_name, record in self.summary().items(): used_bytes = record['used_bytes'] acquired_bytes = record['acquired_bytes'] if unit == 'auto_foreach': denomi_used, unit_used = self._choose_unit(used_bytes) denomi_acquired, unit_acquired = self._choose_unit( acquired_bytes) used_bytes = '%3.2f%s' % (used_bytes / denomi_used, unit_used) acquired_bytes = '%3.2f%s' % ( acquired_bytes / denomi_acquired, unit_acquired) occurrence = str(record['occurrence']) entries.append( [function_name, used_bytes, acquired_bytes, occurrence]) entry_widths = [] entry_widths.append(max(len(f) for f, _, _, _ in entries)) entry_widths.append(max(len(u) for _, u, _, _ in entries)) entry_widths.append(max(len(a) for _, _, a, _ in entries)) entry_widths.append(max(len(o) for _, _, _, o in entries)) template = ' '.join('{:>%d}' % w for w in entry_widths) for function_name, used_bytes, acquired_bytes, occurrence in entries: line = template.format( function_name, used_bytes, acquired_bytes, occurrence) file.write(line) file.write('\n') if hasattr(file, 'flush'): file.flush() class CupyMemoryCumulativeHook(MemoryHook): """A simple memory hook for cupy measuring memory usage cumulatively. Attributes: used_bytes (int): cumulative bytes that application used from cupy memory pool. acquired_bytes (int): cumulative bytes that cupy memory pool acquired from GPU device. """ name = 'CupyMemoryCumulativeHook' def __init__(self): self.used_bytes = 0 self.acquired_bytes = 0 def alloc_preprocess(self, **kwargs): self.acquired_bytes += kwargs['mem_size'] def malloc_preprocess(self, **kwargs): self.used_bytes += kwargs['mem_size']
8,573
38.694444
132
py
chainer
chainer-master/chainer/function_hooks/debug_print.py
import sys import warnings from chainer import backend from chainer import function_hook from chainer import variable class PrintHook(function_hook.FunctionHook): """Function hook that prints debug information. This function hook outputs the debug information of input arguments of ``forward`` and ``backward`` methods involved in the hooked functions at preprocessing time (that is, just before each method is called). Unlike simple "debug print" technique, where users insert print functions at every function to be inspected, we can show the information of all functions involved with single ``with`` statement. Further, this hook enables us to show the information of ``backward`` methods without inserting print functions into Chainer's library code. Args: sep: *(deprecated since v4.0.0)* Ignored. end: Character to be added at the end of print function. file: Output file_like object that that redirect to. flush: If ``True``, this hook forcibly flushes the text stream at the end of preprocessing. .. admonition:: Example The basic usage is to use it with ``with`` statement. >>> from chainer import function_hooks >>> l = L.Linear(10, 10) >>> x = chainer.Variable(np.zeros((1, 10), np.float32)) >>> with chainer.function_hooks.PrintHook(): ... y = l(x) ... z = F.sum(y) ... z.backward() # doctest:+SKIP In this example, ``PrintHook`` shows the debug information of forward propagation of ``LinearFunction`` (which is implicitly called by ``l``) and ``Sum`` (called by ``F.sum``) and backward propagation of ``z`` and ``y``. """ name = 'PrintHook' def __init__(self, sep=None, end='\n', file=sys.stdout, flush=True): if sep is not None: warnings.warn('sep argument in chainer.function_hooks.PrintHook ' 'is deprecated.', DeprecationWarning) self.sep = sep # Keep sep because it was originally documented self.end = end self.file = file self.flush = flush def _print(self, msg): self.file.write(msg + self.end) def _process(self, function, in_data, out_grad=None): self._print('function\t{}'.format(function.label)) self._print('input data') for d in in_data: if d is None: # Some inputs can be removed with `retain_grad`. self._print('(removed)') continue self._print(variable.Variable(d).debug_print()) if out_grad is not None: self._print('output gradient') for d in out_grad: if d is None: v = variable.Variable() else: xp = backend.get_array_module(d) v = variable.Variable(xp.zeros_like(d, dtype=d.dtype)) v.grad = d self._print(v.debug_print()) if self.flush and hasattr(self.file, 'flush'): self.file.flush() def forward_preprocess(self, function, in_data): self._process(function, in_data) def backward_preprocess(self, function, in_data, out_grad): self._process(function, in_data, out_grad)
3,334
35.648352
77
py
chainer
chainer-master/chainer/function_hooks/__init__.py
# import classes and functions from chainer.function_hooks.cuda_profile import CUDAProfileHook # NOQA from chainer.function_hooks.cupy_memory_profile import CupyMemoryProfileHook # NOQA from chainer.function_hooks.debug_print import PrintHook # NOQA from chainer.function_hooks.timer import TimerHook # NOQA
312
51.166667
84
py
chainer
chainer-master/chainer/utils/error.py
def _format_array_props(arrays): # Formats array shapes and dtypes for error messages. assert isinstance(arrays, (list, tuple)) return ', '.join([ None if arr is None else '{}:{}'.format(arr.shape, arr.dtype.name) for arr in arrays])
271
29.222222
57
py
chainer
chainer-master/chainer/utils/type_check.py
import contextlib import functools import operator import sys import threading import numpy import six import chainer from chainer.backends import cuda _thread_local = threading.local() @contextlib.contextmanager def get_function_check_context(f): try: default = _thread_local.current_function except AttributeError: default = None _thread_local.current_function = f try: yield finally: _thread_local.current_function = default class TypeInfo(object): """Type information of an input/gradient array. It contains type information of an array, such as the shape of array and the number of dimensions. This information is independent of CPU or GPU array. """ def __init__(self, shape, dtype): self.shape = shape self.dtype = dtype self.ndim = len(shape) @property def size(self): return functools.reduce(operator.mul, self.shape, 1) class TypeInfoTuple(tuple): """Type information of input/gradient tuples. It is a sub-class of tuple containing :class:`TypeInfo`. The i-th element of this object contains type information of the i-th input/gradient data. As each element is :class:`Expr`, you can easily check its validity. """ def size(self): """Returns an expression representing its length. Returns: Expr: An expression object representing length of the tuple. """ return Variable(len(self), '{0}.size'.format(self.name)) class LightTypeInfoTuple(tuple): """Type information of input/gradient tuples for light-weight check. It is a sub-class of tuple containing :class:`TypeInfo`. The i-th element of this object contains type information of the i-th input/gradient data. """ def size(self): """Returns its length. Returns: int: Length of the tuple. """ return len(self) def get_types(data, name, accept_none, *, shapes=None): assert isinstance(data, tuple) if shapes is None: shapes = tuple([x.shape for x in data]) info = TypeInfoTuple( _get_type(name, i, x, accept_none, shape) for i, (x, shape) in enumerate(zip(data, shapes))) # I don't know a method to set an attribute in an initializer of tuple. info.name = name return info def get_light_types(data, *, shapes=None): assert(isinstance(data, tuple)) if shapes is None: data_ = data else: # For non-default memory format (e.g. NHWC), shapes of data are # different from the semantic shapes (e.g. NCHW). In such cases # semantic shapes are explicitly given as `shapes` argument. # If it is given, TypeInfos with modified shapes are wrapped. data_ = tuple([ TypeInfo(shape, x.dtype) for x, shape in zip(data, shapes)]) return LightTypeInfoTuple(data_) def _get_type(name, index, array, accept_none, shape): var = '{0}[{1}]'.format(name, index) if accept_none and array is None: # case that gradient is not given return Variable(TypeInfo((), None), var) assert isinstance(array, chainer.get_array_types()) return Variable(TypeInfo(shape, array.dtype), var) def _make_un_operator(exp, priority, func): def f(x): return UnaryOperator(priority, x, exp, func) return f def _make_bin_operator(exp, priority, func, right_associative=False): def f(x, y): return BinaryOperator(priority, x, y, exp, func, right_associative) return f def _make_bool_operator(exp, inv, func): def f(x, y): return BoolBinaryOperator(x, y, exp, inv, func) return f def _flip(f): return lambda x, y: f(y, x) class Expr(object): """Abstract syntax tree of an expression. It represents an abstract syntax tree, and isn't a value. You can get its actual value with :meth:`eval` function, and get syntax representation with the :meth:`__str__` method. Each comparison operator (e.g. ``==``) generates a new :class:`Expr` object which represents the result of comparison between two expressions. .. admonition:: Example Let ``x`` and ``y`` be instances of :class:`Expr`, then :: >>> x = Variable(1, 'x') >>> y = Variable(1, 'y') >>> c = (x == y) is also an instance of :class:`Expr`. To evaluate and get its value, call :meth:`eval` method:: >>> c.eval() True Call ``str`` function to get a representation of the original equation:: >>> str(c) 'x == y' You can actually compare an expression with a value:: >>> (x == 1).eval() True Note that you can't use boolean operators such as ``and``, as they try to cast expressions to boolean values:: >>> z = Variable(1, 'z') >>> x == y and y == z # raises an error Traceback (most recent call last): RuntimeError: Don't convert Expr to bool. Please call Expr.eval \ method to evaluate expression. """ def __init__(self, priority): self.priority = priority def eval(self): """Evaluates the tree to get actual value. Behavior of this function depends on an implementation class. For example, a binary operator ``+`` calls the ``__add__`` function with the two results of :meth:`eval` function. """ raise NotImplementedError() def __getattr__(self, name): return GetAttr(self, name) def __getitem__(self, key): return GetItem(self, key) def __call__(self, *args): return Call(self, args) def __nonzero__(self): # When a user calls a boolean operator like `(x == y and z == w)`, # `and` operator evaluate the first expression. # If it returns `True` (and it's default behavior), the `and` operator # returns *the second expression*, not a boolean value. # So, `(x == y and z == w)` returns the result of `z == w`, and # `(x == y and z == w).expect()` raise no errors but only checks # `z == w`. It is confusing. # See also: # https://docs.python.org/3/library/stdtypes.html msg = ('An Expr instance cannot be evaluated as bool. ' 'Please use chainer.utils.type_check.eval() to evaluate an ' 'expression.') raise RuntimeError(msg) def __bool__(self): self.__nonzero__() __eq__ = _make_bool_operator('==', '!=', operator.__eq__) __ne__ = _make_bool_operator('!=', '==', operator.__ne__) __lt__ = _make_bool_operator('<', '>=', operator.__lt__) __le__ = _make_bool_operator('<=', '>', operator.__le__) __gt__ = _make_bool_operator('>', '<=', operator.__gt__) __ge__ = _make_bool_operator('>=', '<', operator.__ge__) # Please refer the Python documentation to know priority of operators. # https://docs.python.org/3/reference/expressions.html __add__ = _make_bin_operator('+', 4, operator.__add__) __radd__ = _flip(__add__) __sub__ = _make_bin_operator('-', 4, operator.__sub__) __rsub__ = _flip(__sub__) __mul__ = _make_bin_operator('*', 5, operator.__mul__) __rmul__ = _flip(__mul__) if sys.version_info < (3, 0, 0): __div__ = _make_bin_operator('/', 5, operator.__div__) # type: ignore # NOQA __rdiv__ = _flip(__div__) else: __truediv__ = _make_bin_operator('/', 5, operator.__truediv__) __rtruediv__ = _flip(__truediv__) __floordiv__ = _make_bin_operator('//', 5, operator.__floordiv__) __rfloordiv__ = _flip(__floordiv__) __mod__ = _make_bin_operator('%', 5, operator.__mod__) __rmod__ = _flip(__mod__) # Only '**' operator is right-associative __pow__ = _make_bin_operator('**', 7, operator.__mod__, right_associative=True) __lshift__ = _make_bin_operator('<<', 3, operator.__lshift__) __rlshift__ = _flip(__lshift__) __rshift__ = _make_bin_operator('>>', 3, operator.__rshift__) __rrshift__ = _flip(__rshift__) __and__ = _make_bin_operator('&', 2, operator.__and__) __rand__ = _flip(__and__) __xor__ = _make_bin_operator('^', 1, operator.__xor__) __rxor__ = _flip(__xor__) __or__ = _make_bin_operator('|', 0, operator.__or__) __ror__ = _flip(__or__) __neg__ = _make_un_operator('-', 6, operator.__neg__) __pos__ = _make_un_operator('+', 6, operator.__pos__) __invert__ = _make_un_operator('~', 6, operator.__invert__) def _eval_expr(v): if isinstance(v, Expr): return v.eval() elif isinstance(v, list): return list(map(_eval_expr, v)) elif isinstance(v, tuple): return tuple(map(_eval_expr, v)) else: return v def _repr(v): if isinstance(v, Expr): return str(v) elif isinstance(v, list): return '[{0}]'.format(', '.join(map(_repr, v))) elif isinstance(v, tuple): if len(v) == 0: return '()' elif len(v) == 1: return '({0},)'.format(_repr(v[0])) else: return '({0})'.format(', '.join(map(_repr, v))) else: return repr(v) class Atom(Expr): def __init__(self): super(Atom, self).__init__(8) class Constant(Atom): def __init__(self, value): super(Constant, self).__init__() self.value = value def __str__(self): return _repr(self.value) def eval(self): return self.value class Variable(Atom): def __init__(self, value, name): super(Variable, self).__init__() self.value = value self.name = name def __str__(self): return self.name def eval(self): return self.value class GetAttr(Atom): def __init__(self, obj, name): super(GetAttr, self).__init__() self.obj = obj self.name = name def __str__(self): if isinstance(self.name, str): return '{0}.{1}'.format(_repr(self.obj), self.name) elif (isinstance(self.name, Constant) and isinstance(self.name.value, str)): return '{0}.{1}'.format(_repr(self.obj), self.name.value) else: return 'getattr({0}, {1})'.format(_repr(self.obj), _repr(self.name)) def eval(self): return getattr(_eval_expr(self.obj), _eval_expr(self.name)) def _str_subscript(exp): if exp is Ellipsis: return '...' elif isinstance(exp, slice): def key_str(v): return '' if v is None else _repr(v) if exp.step is None: return '{0}:{1}'.format(key_str(exp.start), key_str(exp.stop)) else: return '{0}:{1}:{2}'.format(key_str(exp.start), key_str(exp.stop), key_str(exp.step)) elif isinstance(exp, tuple): return ', '.join(map(_str_subscript, exp)) else: return _repr(exp) class GetItem(Atom): def __init__(self, obj, key): super(GetItem, self).__init__() self.obj = obj self.key = key def __str__(self): key = _str_subscript(self.key) return '{0}[{1}]'.format(_repr(self.obj), key) def eval(self): return _eval_expr(self.obj)[_eval_expr(self.key)] class Call(Atom): def __init__(self, obj, args): assert isinstance(args, tuple) super(Call, self).__init__() self.obj = obj self.args = args def __str__(self): return '{0}({1})'.format(_repr(self.obj), ', '.join(map(_repr, self.args))) def eval(self): args = map(_eval_expr, self.args) func = _eval_expr(self.obj) return func(*args) class UnaryOperator(Expr): def __init__(self, priority, term, exp, func): super(UnaryOperator, self).__init__(priority) self.term = term self.exp = exp self.func = func def eval(self): return self.func(_eval_expr(self.term)) def __str__(self): exp = _repr(self.term) if isinstance(self.term, Expr) and self.term.priority < self.priority: exp = '(' + exp + ')' return self.exp + exp class BinaryOperator(Expr): def __init__(self, priority, lhs, rhs, exp, func, right_associative=False): super(BinaryOperator, self).__init__(priority) self.lhs = lhs self.rhs = rhs self.exp = exp self.func = func self.right_associative = right_associative def eval(self): left = self._eval_left() right = self._eval_right() return self.func(left, right) def _eval_left(self): return _eval_expr(self.lhs) def _eval_right(self): return _eval_expr(self.rhs) def __str__(self): # When an infix operator is left-associative, we need to append parens # when rhs has the same priority # e.g. x << (y << z) != x << y << z left = _repr(self.lhs) if isinstance(self.lhs, Expr) and ( self.priority > self.lhs.priority or (self.right_associative and self.priority == self.lhs.priority)): left = '(' + left + ')' right = _repr(self.rhs) if isinstance(self.rhs, Expr) and ( self.priority > self.rhs.priority or (not self.right_associative and self.priority == self.rhs.priority)): right = '(' + right + ')' return '{0} {2} {1}'.format(left, right, self.exp) class Testable(object): def expect(self): raise NotImplementedError() class BoolBinaryOperator(BinaryOperator, Testable): def __init__(self, lhs, rhs, exp, inv, func): BinaryOperator.__init__(self, -1, lhs, rhs, exp, func) self.inv = inv def expect(self): left = self._eval_left() right = self._eval_right() if not self.func(left, right): raise InvalidType( '{0} {1} {2}'.format(self.lhs, self.exp, self.rhs), '{0} {1} {2}'.format(left, self.inv, right)) class InvalidType(Exception): """Raised when types of data for forward/backward are invalid. """ def __init__(self, expect, actual, msg=None): if msg is None: msg = 'Expect: {0}\nActual: {1}'.format(expect, actual) if (hasattr(_thread_local, 'current_function') and _thread_local.current_function is not None): msg = ''' Invalid operation is performed in: {0} (Forward) {1}'''.format(_thread_local.current_function.label, msg) super(InvalidType, self).__init__(msg) self.expect = expect self.actual = actual def __reduce__(self): msg, = self.args return (InvalidType, (self.expect, self.actual, msg)) def _argname(in_types, names): """Assigns user friendly names for the input types. This function also asserts that lengths of in_types and names are the same. Args: in_types (tuple of TypeInfoTuple): Tuple of type information to assign name to. names (tuple of str): Human-readable names of ``in_types``. """ if len(in_types) != len(names): raise InvalidType( '{} argument(s)'.format(str(len(names))), '{} argument(s)'.format(str(len(in_types))), 'Invalid number of arguments') for in_type, name in zip(in_types, names): if isinstance(in_type, Variable): in_type.name = name def expect(*bool_exprs): """Evaluates and tests all given expressions. This function evaluates given boolean expressions in order. When at least one expression is evaluated as ``False``, that means the given condition is not satisfied. You can check conditions with this function. Args: bool_exprs (tuple of Bool expressions): Bool expressions you want to evaluate. """ if in_light_mode(): if not all(bool_exprs): raise InvalidType('', '') else: for expr in bool_exprs: assert isinstance(expr, Testable) expr.expect() def same_types(*arrays): for x in arrays: if not isinstance(x, numpy.ndarray): break else: return True for x in arrays: if not isinstance(x, cuda.ndarray): return False return True def eval(exp): if in_light_mode(): return exp else: return exp.eval() def make_variable(value, name): if in_light_mode(): return value else: return Variable(value, name) def _make_variable_from_array(array, name): if not isinstance(array, chainer.get_array_types()): raise InvalidType( 'isinstance({}, ndarray)'.format(name), 'type({}) == {}'.format(name, type(array)), ) if in_light_mode(): return array else: return Variable(TypeInfo(array.shape, array.dtype), name) class LightMode(object): def __enter__(self): _thread_local.light_mode = True def __exit__(self, exc_type, exc_value, traceback): _thread_local.light_mode = False def _prod_impl(xs): result = 1 for x in xs: result *= x return result _prod = Variable(_prod_impl, 'prod') light_mode = LightMode() def in_light_mode(): try: return _thread_local.light_mode except AttributeError: _thread_local.light_mode = False return False def prod(xs): if in_light_mode(): return _prod_impl(xs) else: return _prod(xs) def expect_broadcast_shapes(*shape_types): """Checks if shapes can be broadcasted together. Args: shapes_types: Type-checked shapes of the arrays to broadcast. """ shapes = [eval(s) for s in shape_types] error = None try: # simulate the shape calculation using zero-sized arrays numpy.broadcast(*[numpy.empty(s + (0,)) for s in shapes]) except ValueError: msgs = ['cannot broadcast inputs of the following shapes:'] for shape_type, shape in six.moves.zip(shape_types, shapes): msgs.append('{} = {}'.format(shape_type, shape)) error = InvalidType('', '', msg='\n'.join(msgs)) if error is not None: raise error
18,493
27.063733
85
py
chainer
chainer-master/chainer/utils/_collections.py
import collections import weakref import six if six.PY3: OrderedDict = collections.OrderedDict else: # Reference counting cannot free keys in old `collections.OrderedDict`, # where a doubly linked list is used to maintain the order. class OrderedDict(object): """Dictionary that remembers insertion order This class wraps `collections.OrderedDict` to free keys by reference counting. """ def __init__(self): self.keys = set() self.dict = collections.OrderedDict() def __contains__(self, key): return weakref.ref(key) in self.dict def __setitem__(self, key, value): self.keys.add(key) self.dict[weakref.ref(key)] = value def __getitem__(self, key): return self.dict[weakref.ref(key)] def items(self): return [(k(), v) for k, v in self.dict.items()] def values(self): return self.dict.values()
994
25.184211
76
py
chainer
chainer-master/chainer/utils/nondeterministic.py
import warnings from chainer import configuration def nondeterministic(f_name): """Function to warn non-deterministic functions If `config.warn_nondeterministic` is True, this function will give a warning that this functions contains a non-deterministic function, such as atomicAdd. """ if configuration.config.warn_nondeterministic: warnings.warn( 'Potentially non-deterministic code is being executed while' ' config.warn_nondeterministic set. Source: ' + f_name)
526
30
75
py
chainer
chainer-master/chainer/utils/conv_nd.py
import itertools import numpy import six from chainer.backends import cuda from chainer.utils.conv import get_conv_outsize from chainer.utils import conv_nd_kernel def as_tuple(x, n): if hasattr(x, '__getitem__'): assert len(x) == n return tuple(x) return (x,) * n def im2col_nd_cpu(img, ksize, stride, pad, pval=0, cover_all=False, dilate=1): n, c = img.shape[0:2] # (n, c, d_1, d_2, ..., d_N) dims = img.shape[2:] ndim = len(dims) dilate = as_tuple(dilate, ndim) assert ndim == len(ksize) == len(stride) == len(pad) outs = tuple(get_conv_outsize(d, k, s, p, cover_all, di) for (d, k, s, p, di) in zip(dims, ksize, stride, pad, dilate)) assert all(out > 0 for out in outs), 'Output sizes should be positive.' # Pad around image. pad_width = ((0, 0), (0, 0)) + tuple( (p, p + s - 1) for (s, p) in zip(stride, pad)) img = numpy.pad(img, pad_width, mode='constant', constant_values=(pval,)) # Make patch array with which we will compute correlation with filter. # shape: (n, c, k_1, k_2, ..., k_N, out_1, out_2, ..., out_N) shape = (n, c) + ksize + outs col = numpy.ndarray(shape, dtype=img.dtype) # Fill the patch array. colon = slice(None) for kxs in itertools.product(*[six.moves.range(k) for k in ksize]): # col[:, :, kx_1, kx_2, ..., kx_N, :, :, ..., :] col_index = (colon, colon) + kxs + (colon,) * ndim # img[:, :, kx_1:kx_lim_1:s_1, ..., kx_N:kx_lim_N:s_N] kx_dilate = tuple(kx * di for (kx, di) in zip(kxs, dilate)) kx_lims = tuple(kx_di + s * out for (kx_di, s, out) in zip(kx_dilate, stride, outs)) img_index = (colon, colon) + tuple( slice(kx_di, kx_lim, s) for (kx_di, kx_lim, s) in zip(kx_dilate, kx_lims, stride)) col[col_index] = img[img_index] return col def im2col_nd_gpu(img, ksize, stride, pad, cover_all=False, dilate=1): n, c = img.shape[0:2] # (n, c, d_1, d_2, ..., d_N) dims = img.shape[2:] ndim = len(dims) dilate = as_tuple(dilate, ndim) assert ndim == len(ksize) == len(stride) == len(pad) outs = tuple(get_conv_outsize(d, k, s, p, cover_all, di) for (d, k, s, p, di) in zip(dims, ksize, stride, pad, dilate)) assert all(out > 0 for out in outs), 'Output sizes should be positive.' # col_shape: (n, c, k_1, k_2, ..., k_N, out_1, out_2, ..., out_N) shape = (n, c) + ksize + outs col = cuda.cupy.empty(shape, dtype=img.dtype) in_params, out_params, operation, name = \ conv_nd_kernel.Im2colNDKernel.generate(ndim) cuda.elementwise(in_params, out_params, operation, name)( img.reduced_view(), *(dims + outs + ksize + stride + pad + dilate + (col,))) return col def im2col_nd(img, ksize, stride, pad, cover_all=False, dilate=1): fn = im2col_nd_gpu if isinstance(img, cuda.ndarray) else im2col_nd_cpu return fn(img, ksize, stride, pad, cover_all=cover_all, dilate=dilate) def col2im_nd_cpu(col, stride, pad, dims, dilate=1): n, c = col.shape[:2] # (n, c, kx_1, ..., kx_N, out_1, ..., out_N) mid = (len(col.shape) - 2) // 2 + 2 ksize = col.shape[2:mid] outs = col.shape[mid:] colon = slice(None) ndim = len(outs) dilate = as_tuple(dilate, ndim) assert len(ksize) == len(stride) == len(pad) == len(dims) == ndim # Image with padded size. img_shape = (n, c) + tuple(d + 2 * p + s - 1 for (d, p, s) in zip(dims, pad, stride)) img = numpy.zeros(img_shape, dtype=col.dtype) for kxs in itertools.product(*[six.moves.range(k) for k in ksize]): # (:, :, kx_1:kx_lim_1:s_1, ..., kx_N:kx_lim_N:s_N) kx_dilate = tuple(kx * di for (kx, di) in zip(kxs, dilate)) kx_lims = tuple(kx_di + s * out for (kx_di, s, out) in zip(kx_dilate, stride, outs)) img_index = (colon, colon) + tuple( slice(kx_di, kx_lim, s) for (kx_di, kx_lim, s) in zip(kx_dilate, kx_lims, stride)) # (:, :, kx_1, kx_2, ..., kx_N, :, :, ..., :) col_index = (colon, colon) + kxs + (colon,) * len(outs) img[img_index] += col[col_index] # (:, :, p_1:d_1 + p_1, p_2:d_2 + p_2, ..., p_N:d_N + p_N] img_index = (colon, colon) + tuple( slice(p, d + p) for (p, d) in zip(pad, dims)) return img[img_index] def col2im_nd_gpu(col, stride, pad, dims, dilate=1): n, c = col.shape[:2] # (n, c, k_1, ..., k_N, out_1, ..., out_N) mid = (len(col.shape) - 2) // 2 + 2 ksize = col.shape[2:mid] outs = col.shape[mid:] ndim = len(dims) dilate = as_tuple(dilate, ndim) assert len(outs) == len(ksize) == len(stride) == len(pad) == ndim img_shape = (n, c) + dims # (n, c, d_1, d_2, ..., d_N) img = cuda.cupy.empty(img_shape, dtype=col.dtype) in_params, out_params, operation, name = \ conv_nd_kernel.Col2imNDKernel.generate(ndim) cuda.elementwise(in_params, out_params, operation, name)( col.reduced_view(), *(dims + outs + ksize + stride + pad + dilate + (img,))) return img def col2im_nd(col, stride, pad, dims, dilate=1): fn = col2im_nd_gpu if isinstance(col, cuda.ndarray) else col2im_nd_cpu return fn(col, stride, pad, dims, dilate)
5,382
36.643357
78
py
chainer
chainer-master/chainer/utils/array.py
import warnings import numpy import six import chainer from chainer.backends import cuda def as_vec(x): warnings.warn( 'chainer.utils.array.as_vec is deprecated. Please refer to ' 'numpy.ravel or other array backend functions to flatten ndarrays.', DeprecationWarning) if x.ndim == 1: return x return x.ravel() def as_mat(x): warnings.warn( 'chainer.utils.array.as_mat is deprecated. Please refer to ' 'numpy.reshape or other array backend functions to reshape ndarrays.', DeprecationWarning) if x.ndim == 2: return x return x.reshape(len(x), -1) def empty_like(x): warnings.warn( 'chainer.utils.array.empty_like is deprecated. Please refer to ' 'numpy.empty_like or other array backend functions to initialize ' 'empty arrays.', DeprecationWarning) if cuda.available and isinstance(x, cuda.ndarray): return cuda.cupy.empty_like(x) else: return numpy.empty_like(x) def size_of_shape(shape): size = 1 for i in shape: size *= i # should not return long in Python 2 return int(size) def sum_to(x, shape): if x.shape == shape: return x if isinstance(x, chainer.Variable): raise TypeError( 'chainer.utils.sum_to does not support Variable input. ' 'Use chainer.functions.sum_to instead.') ndim = len(shape) lead = x.ndim - ndim lead_axis = tuple(six.moves.range(lead)) axis = tuple([i + lead for i, sx in enumerate(shape) if sx == 1]) y = x.sum(lead_axis + axis, keepdims=True) if lead > 0: y = y.squeeze(lead_axis) return y
1,688
24.590909
78
py
chainer
chainer-master/chainer/utils/sparse.py
import numpy import chainer from chainer import backend from chainer import cuda def _add_at(add_at, x, row, col, data): assert data.size > 0 last_nz = data.size - (data != 0)[::-1].argmax() add_at(x, (row[:last_nz], col[:last_nz]), data[:last_nz]) class CooMatrix(object): """A sparse matrix in COO format. Args: data (:ref:`ndarray`): The entries of the matrix. The entries are usually non-zero-elements in the matrix. row (:ref:`ndarray`): The row indices of the matrix entries. col (:ref:`ndarray`): The column indices of the matrix entries. shape (tuple of int): The shape of the matrix in dense format. order ('C', 'F', 'other' or None): If ``'C'``, the maxtix is assumed that its row indices are sorted. If ``'F'``, the matrix is assumed that its column indices are sorted. If ``'other'``, the matrix is assumed as neither 'C' order nor 'F' order. If ``None`` (this is the default), the matrix is automatically checked if it is 'C' order, 'F' order or another. This information will be used by some functions like :func:`~chainer.functions.sparse_matmul` as a hint to improve performance. requires_grad (bool): If ``True``, gradient of this sparse matrix will be computed in back-propagation. .. seealso:: See :func:`~chainer.utils.to_coo` for how to construct a COO matrix from an array. """ def __init__(self, data, row, col, shape, order=None, requires_grad=False): if not (1 <= data.ndim <= 2): raise ValueError('ndim of data must be 1 or 2.') if not (data.ndim == row.ndim == col.ndim): raise ValueError('ndim of data, row and col must be the same.') if len(shape) != 2: raise ValueError('length of shape must be 2.') if not (shape[0] > 0 and shape[1] > 0): raise ValueError('numbers in shape must be greater than 0.') if order not in ('C', 'F', 'other', None): raise ValueError('order must be \'C\', \'F\', \'other\' or None.') self.data = chainer.Variable(data, requires_grad=requires_grad) self.row = row self.col = col self.shape = shape # (row, col) self.order = order if order is None: self.order = get_order(row, col) def to_dense(self): """Returns a dense matrix format of this sparse matrix.""" data = self.data if data.ndim == 1: shape = self.shape elif data.ndim == 2: shape = (data.shape[0], *self.shape) else: assert False xp = data.xp x = xp.zeros(shape, dtype=data.dtype) if data.size > 0: row = self.row col = self.col if xp is numpy: add_at = numpy.add.at elif xp is cuda.cupy: add_at = cuda.cupyx.scatter_add data = data.array if data.ndim == 1: _add_at(add_at, x, row, col, data) elif data.ndim == 2: for i in range(data.shape[0]): _add_at(add_at, x[i], row[i], col[i], data[i]) else: assert False return x def to_coo(x, ldnz=None, requires_grad=False): """Returns a single or a batch of matrices in COO format. Args: x (:ref:`ndarray`): Input dense matrix. The ndim of ``x`` must be two or three. If ndim is two, it is treated as a single matrix. If three, it is treated as batched matrices. ldnz (int): Size of arrays for data, row index and column index to be created. The Actual size becomes max(nnz, ldnz) where nnz is number of non-zero elements in a input dense matrix. requires_grad (bool): If ``True``, gradient of sparse matrix will be computed in back-propagation. Returns: ~chainer.utils.CooMatrix: A sparse matrix or batched sparse matrices in COO format of a given dense matrix or batched dense matrices. .. admonition:: Example Create a :class:`~chainer.utils.CooMatrix` from an array with 2 non-zero elements and 4 zeros and access its attributes. No batch dimension is involved. .. doctest:: >>> data = np.array([[0, 2, 0], [-1, 0, 0]], np.float32) >>> x = chainer.utils.to_coo(data) >>> x.data variable([ 2., -1.]) >>> x.row array([0, 1], dtype=int32) >>> x.col array([1, 0], dtype=int32) >>> x.shape (2, 3) """ xp = backend.get_array_module(x) if x.ndim == 2: _row, _col = xp.where(x != 0) nnz = len(_row) if ldnz is None or ldnz < nnz: ldnz = nnz data = xp.zeros((ldnz), dtype=x.dtype) row = xp.full((ldnz), -1, dtype=xp.int32) col = xp.full((ldnz), -1, dtype=xp.int32) data[:nnz] = x[_row, _col] row[:nnz] = xp.array(_row).astype(xp.int32) col[:nnz] = xp.array(_col).astype(xp.int32) shape = x.shape return CooMatrix(data, row, col, shape, requires_grad=requires_grad) elif x.ndim == 3: # first axis is batch axis nb = x.shape[0] if ldnz is None: ldnz = 0 for i in range(nb): ldnz = max(ldnz, len(xp.where(x[i] != 0)[0])) data = xp.empty((nb, ldnz), dtype=x.dtype) row = xp.empty((nb, ldnz), dtype=xp.int32) col = xp.empty((nb, ldnz), dtype=xp.int32) for i in range(nb): coo = to_coo(x[i], ldnz) data[i] = coo.data.data row[i] = coo.row col[i] = coo.col shape = x.shape[1:] return CooMatrix(data, row, col, shape, requires_grad=requires_grad) else: raise ValueError('ndim of x must be 2 or 3.') def get_order(row, col): """Check if a coo matrix with given row and col is C or F order. Args: row (:ref:`ndarray`): The row indices of the matrix entries. col (:ref:`ndarray`): The column indices of the matrix entries. Returns: Returns ``'C'`` when a coo matrix with given row and column indices is C order, in other words, the row indices are sorted. Returns ``'F'`` when it is F order, in other words, the column indices are sorted. Returns ``'other'`` otherwise. """ if _is_c_order(row, col): return 'C' if _is_c_order(col, row): return 'F' return 'other' def _is_c_order(row, col): """Check if a coo matrix with given row and col is c_order""" if row.shape != col.shape: raise ValueError('shape of row and col must be the same.') if row.ndim != 1: for i in range(row.shape[0]): if not _is_c_order(row[i], col[i]): return False return True xp = backend.get_array_module(row) _row = row[col >= 0] _col = col[row >= 0] if _row[_row < 0].size > 0 or _col[_col < 0].size: raise ValueError('invalid index combination of row and col.') if _row.shape[0] <= 1: return True row_diff = xp.zeros(_row.shape, dtype=_row.dtype) row_diff[1:] = _row[1:] - _row[:-1] if xp.amin(row_diff) < 0: return False col_diff = xp.zeros(_col.shape, dtype=_col.dtype) col_diff[1:] = _col[1:] - _col[:-1] col_diff[(row_diff > 0)] = 0 return xp.amin(col_diff) >= 0
7,680
34.725581
79
py
chainer
chainer-master/chainer/utils/argument.py
import inspect def check_unexpected_kwargs(kwargs, **unexpected): for key, message in unexpected.items(): if key in kwargs: raise ValueError(message) def parse_kwargs(kwargs, *name_and_values, **unexpected): values = [kwargs.pop(name, default_value) for name, default_value in name_and_values] if kwargs: check_unexpected_kwargs(kwargs, **unexpected) caller = inspect.stack()[1] args = ', '.join(repr(arg) for arg in sorted(kwargs.keys())) message = caller[3] + \ '() got unexpected keyword argument(s) {}'.format(args) raise TypeError(message) return tuple(values) def assert_kwargs_empty(kwargs): # It only checks if kwargs is empty. parse_kwargs(kwargs)
773
28.769231
68
py
chainer
chainer-master/chainer/utils/precision.py
import functools import numpy def _fp16_mixed_precision_helper(fn): """Decorator to perform computation in FP32 for FP16 inputs/outputs Decorator to perform forward computation in FP32 for FP16 inputs, returning outputs casted back to FP16. Do nothing for FP32 and FP64 inputs. """ @functools.wraps(fn) def wrapper(self, in_data): flag = all([x.dtype.kind != 'f' or x.dtype == numpy.float16 for x in in_data]) in_data1 = [] for x in in_data: if x.dtype == numpy.float16: in_data1.append(x.astype(numpy.float32)) else: in_data1.append(x) in_data1 = tuple(in_data1) out_data = fn(self, in_data1) if flag: out_data1 = [] for y in out_data: if y is not None and y.dtype == numpy.float32: out_data1.append(y.astype(numpy.float16)) else: out_data1.append(y) out_data = tuple(out_data1) return out_data return wrapper
1,091
27
71
py
chainer
chainer-master/chainer/utils/walker_alias.py
import numpy import chainer from chainer.backends import cuda from chainer import device_resident class WalkerAlias(device_resident.DeviceResident): """Implementation of Walker's alias method. This method generates a random sample from given probabilities :math:`p_1, \\dots, p_n` in :math:`O(1)` time. It is more efficient than :func:`~numpy.random.choice`. This class works on both CPU and GPU. Args: probs (float list): Probabilities of entries. They are normalized with `sum(probs)`. See: `Wikipedia article <https://en.wikipedia.org/wiki/Alias_method>`_ """ def __init__(self, probs): super(WalkerAlias, self).__init__() prob = numpy.array(probs, numpy.float32) prob /= numpy.sum(prob) threshold = numpy.ndarray(len(probs), numpy.float32) values = numpy.ndarray(len(probs) * 2, numpy.int32) il, ir = 0, 0 pairs = list(zip(prob, range(len(probs)))) pairs.sort() for prob, i in pairs: p = prob * len(probs) while p > 1 and ir < il: values[ir * 2 + 1] = i p -= 1.0 - threshold[ir] ir += 1 threshold[il] = p values[il * 2] = i il += 1 # fill the rest for i in range(ir, len(probs)): values[i * 2 + 1] = 0 assert((values < len(threshold)).all()) self.threshold = threshold self.values = values @property def use_gpu(self): # TODO(niboshi): Maybe better to deprecate the property. device = self.device xp = device.xp if xp is cuda.cupy: return True elif xp is numpy: return False raise RuntimeError( 'WalkerAlias.use_gpu attribute is only applicable for numpy or ' 'cupy devices. Use WalkerAlias.device attribute for general ' 'devices.') def device_resident_accept(self, visitor): super(WalkerAlias, self).device_resident_accept(visitor) self.threshold = visitor.visit_array(self.threshold) self.values = visitor.visit_array(self.values) def sample(self, shape): """Generates a random sample based on given probabilities. Args: shape (tuple of int): Shape of a return value. Returns: Returns a generated array with the given shape. If a sampler is in CPU mode the return value is a :class:`numpy.ndarray` object, and if it is in GPU mode the return value is a :class:`cupy.ndarray` object. """ device = self.device xp = device.xp with chainer.using_device(device): if xp is cuda.cupy: return self.sample_gpu(shape) else: return self.sample_xp(xp, shape) def sample_xp(self, xp, shape): thr_dtype = self.threshold.dtype pb = xp.random.uniform(0, len(self.threshold), shape) index = pb.astype(numpy.int32) left_right = ( self.threshold[index] < (pb.astype(thr_dtype) - index.astype(thr_dtype))) left_right = left_right.astype(numpy.int32) return self.values[index * 2 + left_right] def sample_gpu(self, shape): ps = cuda.cupy.random.uniform(size=shape, dtype=numpy.float32) vs = cuda.elementwise( 'T ps, raw T threshold , raw S values, int32 b', 'int32 vs', ''' T pb = ps * b; int index = __float2int_rd(pb); // fill_uniform sometimes returns 1.0, so we need to check index if (index >= b) { index = 0; } int lr = threshold[index] < pb - index; vs = values[index * 2 + lr]; ''', 'walker_alias_sample' )(ps, self.threshold, self.values, len(self.threshold)) return vs
3,976
32.70339
78
py