code
stringlengths
101
5.91M
def _get_required_attr(element: Element, attr: str) -> str: attribute = element.get(attr) if (attribute is None): raise MusicXMLError(f"Attribute '{attr}' is required for an '{element.tag}' element.") return attribute
def get_user_seqs(data_file): lines = open(data_file).readlines() user_seq = [] item_set = set() for line in lines: (user, items) = line.strip().split(' ', 1) items = items.split(' ') items = [int(item) for item in items] user_seq.append(items) item_set = (item_set | set(items)) max_item = max(item_set) num_users = len(lines) num_items = (max_item + 2) valid_rating_matrix = generate_rating_matrix_valid(user_seq, num_users, num_items) test_rating_matrix = generate_rating_matrix_test(user_seq, num_users, num_items) return (user_seq, max_item, valid_rating_matrix, test_rating_matrix)
class BufferReader(): def __init__(self, buffer: bytes): self.buffer = buffer self.read_offset = 0 def bytes_left(self): return (len(self.buffer) - self.read_offset) def unpack_f(self, s_format: str): if (not hasattr(ConstStructs, s_format)): le_format: str = ('<' + s_format) setattr(ConstStructs, s_format, struct.Struct(le_format)) return self.unpack(getattr(ConstStructs, s_format)) def unpack_numpy(self, s: struct.Struct, shape: Tuple): arr = np.ndarray(shape, s.format, self.buffer, self.read_offset).copy() self.advance(s, int(np.prod(shape))) return arr def unpack_torch(self, s: struct.Struct, shape: Tuple): import torch arr = self.unpack_numpy(s, shape) return torch.from_numpy(arr) def unpack_tensorflow(self, s: struct.Struct, shape: Tuple): import tensorflow as tf arr = self.unpack_numpy(s, shape) return tf.constant(arr) def unpack(self, s: struct.Struct): unpack: tuple = s.unpack_from(self.buffer, self.read_offset) self.advance(s) if (len(unpack) == 1): return unpack[0] return unpack def advance(self, s: struct.Struct, times=1): self.read_offset += (s.size * times) def unpack_str(self) -> str: length: int = self.unpack(ConstStructs.ushort) bytes_: bytes = self.unpack_f(('%ds' % length)) return bytes_.decode('utf-8')
class AverageOfMaximumScoreEnsembler(PYODScoreEnsembler): def __init__(self, n_buckets=5, method='static', bootstrap_estimators=False): self.method = method self.n_buckets = n_buckets self.bootstrap_estimators = bootstrap_estimators def _combine(self, scores): return aom(scores, n_buckets=self.n_buckets, method=self.method, bootstrap_estimators=self.bootstrap_estimators)
def calculate_matvec_accumulator_range(matrix, vec_dt): min_weight = matrix.min() max_weight = matrix.max() perceptive_field_elems = matrix.shape[0] min_input = vec_dt.min() max_input = vec_dt.max() acc_min = (perceptive_field_elems * min((min_weight * max_input), (min_weight * min_input), (max_weight * max_input), (max_weight * min_input))) acc_max = (perceptive_field_elems * max((min_weight * max_input), (min_weight * min_input), (max_weight * max_input), (max_weight * min_input))) return (acc_min, acc_max)
def convert_path_to_npy(*, path='train_64x64', outfile='train_64x64.npy'): assert isinstance(path, str), 'Expected a string input for the path' assert os.path.exists(path), "Input path doesn't exist" files = [f for f in listdir(path) if isfile(join(path, f))] print('Number of valid images is:', len(files)) imgs = [] for i in tqdm(range(len(files))): img = scipy.ndimage.imread(join(path, files[i])) img = img.astype('uint8') assert (np.max(img) <= 255) assert (np.min(img) >= 0) assert (img.dtype == 'uint8') assert isinstance(img, np.ndarray) imgs.append(img) (resolution_x, resolution_y) = (img.shape[0], img.shape[1]) imgs = np.asarray(imgs).astype('uint8') assert (imgs.shape[1:] == (resolution_x, resolution_y, 3)) assert (np.max(imgs) <= 255) assert (np.min(imgs) >= 0) print('Total number of images is:', imgs.shape[0]) print('All assertions done, dumping into npy file') np.save(outfile, imgs)
def _format(val: Any, output_format: str='standard', errors: str='coarse') -> Any: val = str(val) result: Any = [] if (val in NULL_VALUES): return [np.nan] if (not validate_it_iva(val)): if (errors == 'raise'): raise ValueError(f'Unable to parse value {val}') error_result = (val if (errors == 'ignore') else np.nan) return [error_result] if (output_format in {'compact', 'standard'}): result = ([iva.compact(val)] + result) return result
class IRBlock(nn.Module): def __init__(self, inplanes, planes, stride=1, downsample=None, norm_type='batch', use_se=True): super(IRBlock, self).__init__() norm_layer = build_norm(norm_type, dimension=2) self.bn0 = norm_layer(inplanes) self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=3, stride=1, padding=1, bias=False) self.bn1 = norm_layer(planes) self.prelu = nn.PReLU(num_parameters=planes) self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride, padding=1, bias=False) self.bn2 = norm_layer(planes) self.downsample = downsample self.stride = stride self.use_se = use_se if self.use_se: self.se = SEBlock(planes) def forward(self, x): residual = x out = self.bn0(x) out = self.conv1(out) out = self.bn1(out) out = self.prelu(out) out = self.conv2(out) out = self.bn2(out) if self.use_se: out = self.se(out) if (self.downsample is not None): residual = self.downsample(x) out += residual return out
def acos_safe(x: Scalar, epsilon: Scalar=epsilon()) -> Scalar: x_safe = Max(((- 1) + epsilon), Min((1 - epsilon), x)) return sympy.acos(x_safe)
def eval_sighan2015_by_model(correct_fn, sighan_path=sighan_2015_path, verbose=True): TP = 0.0 FP = 0.0 FN = 0.0 TN = 0.0 total_num = 0 start_time = time.time() with open(sighan_path, 'r', encoding='utf-8') as f: for line in f: line = line.strip() if line.startswith('#'): continue parts = line.split('\t') if (len(parts) != 2): continue src = parts[0] tgt = parts[1] r = correct_fn(src) (tgt_pred, pred_detail) = (r['target'], r['errors']) if verbose: print() print('input :', src) print('truth :', tgt) print('predict:', tgt_pred, pred_detail) if (src == tgt): if (tgt == tgt_pred): TN += 1 print('right') else: FP += 1 print('wrong') elif (tgt == tgt_pred): TP += 1 print('right') else: FN += 1 print('wrong') total_num += 1 spend_time = (time.time() - start_time) acc = ((TP + TN) / total_num) precision = ((TP / (TP + FP)) if (TP > 0) else 0.0) recall = ((TP / (TP + FN)) if (TP > 0) else 0.0) f1 = ((((2 * precision) * recall) / (precision + recall)) if ((precision + recall) != 0) else 0) print(f'Sentence Level: acc:{acc:.4f}, precision:{precision:.4f}, recall:{recall:.4f}, f1:{f1:.4f}, cost time:{spend_time:.2f} s, total num: {total_num}') return (acc, precision, recall, f1)
def florisPw(u_stream, tis, xs, ys, yws): if (curl == True): fi.floris.farm.set_wake_model('curl') fi.reinitialize_flow_field(wind_speed=u_stream) fi.reinitialize_flow_field(turbulence_intensity=tis) fi.reinitialize_flow_field(layout_array=[xs, ys]) fi.calculate_wake(yaw_angles=yws) floris_power_0 = fi.get_farm_power() return round((floris_power_0 / 1000000.0), 2)
def StackedRNN(inners, num_layers, lstm=False, dropout=0, train=True): num_directions = len(inners) total_layers = (num_layers * num_directions) def forward(input, hidden, weight, batch_sizes): assert (len(weight) == total_layers) next_hidden = [] if lstm: hidden = list(zip(*hidden)) for i in range(num_layers): all_output = [] for (j, inner) in enumerate(inners): l = ((i * num_directions) + j) (hy, output) = inner(input, hidden[l], weight[l], batch_sizes) next_hidden.append(hy) all_output.append(output) input = torch.cat(all_output, (input.dim() - 1)) if ((dropout != 0) and (i < (num_layers - 1))): input = F.dropout(input, p=dropout, training=train, inplace=False) if lstm: (next_h, next_c) = zip(*next_hidden) next_hidden = (torch.cat(next_h, 0).view(total_layers, *next_h[0].size()), torch.cat(next_c, 0).view(total_layers, *next_c[0].size())) else: next_hidden = torch.cat(next_hidden, 0).view(total_layers, *next_hidden[0].size()) return (next_hidden, input) return forward
class ResizeShortestEdge(T.Augmentation): def __init__(self, short_edge_length, max_size=sys.maxsize, sample_style='range', interp=Image.BILINEAR, clip_frame_cnt=1): super().__init__() assert (sample_style in ['range', 'choice', 'range_by_clip', 'choice_by_clip']), sample_style self.is_range = ('range' in sample_style) if isinstance(short_edge_length, int): short_edge_length = (short_edge_length, short_edge_length) if self.is_range: assert (len(short_edge_length) == 2), f"short_edge_length must be two values using 'range' sample style. Got {short_edge_length}!" self._cnt = 0 self._init(locals()) def get_transform(self, image): if ((self._cnt % self.clip_frame_cnt) == 0): if self.is_range: self.size = np.random.randint(self.short_edge_length[0], (self.short_edge_length[1] + 1)) else: self.size = np.random.choice(self.short_edge_length) if (self.size == 0): return NoOpTransform() self._cnt = 0 self._cnt += 1 (h, w) = image.shape[:2] scale = ((self.size * 1.0) / min(h, w)) if (h < w): (newh, neww) = (self.size, (scale * w)) else: (newh, neww) = ((scale * h), self.size) if (max(newh, neww) > self.max_size): scale = ((self.max_size * 1.0) / max(newh, neww)) newh = (newh * scale) neww = (neww * scale) neww = int((neww + 0.5)) newh = int((newh + 0.5)) return T.ResizeTransform(h, w, newh, neww, self.interp)
class TestDivision(object): def test_division_int(self): x = np.array([5, 10, 90, 100, (- 5), (- 10), (- 90), (- 100), (- 120)]) if ((5 / 10) == 0.5): assert_equal((x / 100), [0.05, 0.1, 0.9, 1, (- 0.05), (- 0.1), (- 0.9), (- 1), (- 1.2)]) else: assert_equal((x / 100), [0, 0, 0, 1, (- 1), (- 1), (- 1), (- 1), (- 2)]) assert_equal((x // 100), [0, 0, 0, 1, (- 1), (- 1), (- 1), (- 1), (- 2)]) assert_equal((x % 100), [5, 10, 90, 0, 95, 90, 10, 0, 80]) def test_division_complex(self): msg = 'Complex division implementation check' x = np.array([(1.0 + (1.0 * 1j)), (1.0 + (0.5 * 1j)), (1.0 + (2.0 * 1j))], dtype=np.complex128) assert_almost_equal(((x ** 2) / x), x, err_msg=msg) msg = 'Complex division overflow/underflow check' x = np.array([1e+110, 1e-110], dtype=np.complex128) y = ((x ** 2) / x) assert_almost_equal((y / x), [1, 1], err_msg=msg) def test_zero_division_complex(self): with np.errstate(invalid='ignore', divide='ignore'): x = np.array([0.0], dtype=np.complex128) y = (1.0 / x) assert_(np.isinf(y)[0]) y = (complex(np.inf, np.nan) / x) assert_(np.isinf(y)[0]) y = (complex(np.nan, np.inf) / x) assert_(np.isinf(y)[0]) y = (complex(np.inf, np.inf) / x) assert_(np.isinf(y)[0]) y = (0.0 / x) assert_(np.isnan(y)[0]) def test_floor_division_complex(self): msg = 'Complex floor division implementation check' x = np.array([(0.9 + 1j), ((- 0.1) + 1j), (0.9 + (0.5 * 1j)), (0.9 + (2.0 * 1j))], dtype=np.complex128) y = np.array([0.0, (- 1.0), 0.0, 0.0], dtype=np.complex128) assert_equal(np.floor_divide((x ** 2), x), y, err_msg=msg) msg = 'Complex floor division overflow/underflow check' x = np.array([1e+110, 1e-110], dtype=np.complex128) y = np.floor_divide((x ** 2), x) assert_equal(y, [1e+110, 0], err_msg=msg) def test_floor_division_signed_zero(self): x = np.zeros(10) assert_equal(np.signbit((x // 1)), 0) assert_equal(np.signbit(((- x) // 1)), 1)
def test_alias_delay_initialization1(capture): class B(m.A): def __init__(self): super(B, self).__init__() def f(self): print('In python f()') with capture: a = m.A() m.call_f(a) del a pytest.gc_collect() assert (capture == 'A.f()') with capture: b = B() m.call_f(b) del b pytest.gc_collect() assert (capture == '\n PyA.PyA()\n PyA.f()\n In python f()\n PyA.~PyA()\n ')
class RandomSplit(NamedTuple): train_sentences: List[Sentence] dev_sentences: List[Sentence] test_sentences: List[Sentence]
class PlyProperty(object): def __init__(self, name, val_dtype): _check_name(name) self._name = str(name) self.val_dtype = val_dtype def _get_val_dtype(self): return self._val_dtype def _set_val_dtype(self, val_dtype): self._val_dtype = _data_types[_lookup_type(val_dtype)] val_dtype = property(_get_val_dtype, _set_val_dtype) def name(self): return self._name def dtype(self, byte_order='='): return (byte_order + self.val_dtype) def _from_fields(self, fields): return _np.dtype(self.dtype()).type(next(fields)) def _to_fields(self, data): (yield _np.dtype(self.dtype()).type(data)) def _read_bin(self, stream, byte_order): try: return _read_array(stream, self.dtype(byte_order), 1)[0] except IndexError: raise StopIteration def _write_bin(self, data, stream, byte_order): _write_array(stream, _np.dtype(self.dtype(byte_order)).type(data)) def __str__(self): val_str = _data_type_reverse[self.val_dtype] return ('property %s %s' % (val_str, self.name)) def __repr__(self): return ('PlyProperty(%r, %r)' % (self.name, _lookup_type(self.val_dtype)))
def build_alphabet(data=None, names=None, name=None): if ((name is not None) and ((data is not None) or (names is not None))): raise ValueError('name cannot be specified with any other argument') if (isinstance(names, (int, Integer)) or (names == Infinity) or ((data is None) and (names is not None))): (data, names) = (names, data) if isinstance(data, (int, Integer)): if (names is None): from sage.sets.integer_range import IntegerRange return IntegerRange(Integer(data)) if isinstance(names, str): return TotallyOrderedFiniteSet([(names + ('%d' % i)) for i in range(data)]) if (isinstance(names, collections.abc.Sequence) and (len(names) == data)): return TotallyOrderedFiniteSet(names) raise ValueError('invalid value for names') if (data == Infinity): data = NonNegativeIntegers() if (isinstance(data, (tuple, list, str, range)) or (data in Sets())): if (names is not None): if (not isinstance(names, str)): raise TypeError('names must be a string when data is a set') return Family(data, (lambda i: (names + str(i))), name=names) if (data in Sets()): return data return TotallyOrderedFiniteSet(data) if (name is not None): if (not isinstance(name, str)): raise TypeError('name must be a string') if ((name == 'positive integers') or (name == 'PP')): from sage.sets.positive_integers import PositiveIntegers return PositiveIntegers() if ((name == 'natural numbers') or (name == 'NN')): return NonNegativeIntegers() data = [] for alpha_name in name.split(' '): try: data.extend(list(set_of_letters[alpha_name])) except KeyError: raise TypeError('name is not recognized') return TotallyOrderedFiniteSet(data) if (data is None): from sage.sets.pythonclass import Set_PythonType return Set_PythonType(object) raise ValueError('unable to construct an alphabet from the given parameters')
def imsave(path, img, channel_first=False, as_uint16=False, auto_scale=True, **kwargs): best_backend = backend_manager.get_best_backend(path, 'save') best_backend.imsave(path, img, channel_first=channel_first, as_uint16=as_uint16, auto_scale=auto_scale, **kwargs)
_module() class VeryDeepVgg(BaseModule): def __init__(self, leaky_relu=True, input_channels=3, init_cfg=[dict(type='Xavier', layer='Conv2d'), dict(type='Uniform', layer='BatchNorm2d')]): super().__init__(init_cfg=init_cfg) ks = [3, 3, 3, 3, 3, 3, 2] ps = [1, 1, 1, 1, 1, 1, 0] ss = [1, 1, 1, 1, 1, 1, 1] nm = [64, 128, 256, 256, 512, 512, 512] self.channels = nm cnn = Sequential() def conv_relu(i, batch_normalization=False): n_in = (input_channels if (i == 0) else nm[(i - 1)]) n_out = nm[i] cnn.add_module('conv{0}'.format(i), nn.Conv2d(n_in, n_out, ks[i], ss[i], ps[i])) if batch_normalization: cnn.add_module('batchnorm{0}'.format(i), nn.BatchNorm2d(n_out)) if leaky_relu: cnn.add_module('relu{0}'.format(i), nn.LeakyReLU(0.2, inplace=True)) else: cnn.add_module('relu{0}'.format(i), nn.ReLU(True)) conv_relu(0) cnn.add_module('pooling{0}'.format(0), nn.MaxPool2d(2, 2)) conv_relu(1) cnn.add_module('pooling{0}'.format(1), nn.MaxPool2d(2, 2)) conv_relu(2, True) conv_relu(3) cnn.add_module('pooling{0}'.format(2), nn.MaxPool2d((2, 2), (2, 1), (0, 1))) conv_relu(4, True) conv_relu(5) cnn.add_module('pooling{0}'.format(3), nn.MaxPool2d((2, 2), (2, 1), (0, 1))) conv_relu(6, True) self.cnn = cnn def out_channels(self): return self.channels[(- 1)] def forward(self, x): output = self.cnn(x) return output
class SideObstacleSetBBreakoutWorld(RandomSideObstacleBreakoutWorld): side_obstacle_width_range_start = 15 side_obstacle_width_range_end = 20
class FeatureWrapper(torch.utils.data.Dataset): def __init__(self, data_source, feature_path): self.data_source = data_source self.features = torch.load(feature_path) def __len__(self): return len(self.data_source) def __getitem__(self, idx): item = self.data_source[idx] feature = self.features[item['impath']] output = {'img': feature, 'label': item['label'], 'classname': item['classname'], 'impath': item['impath'], 'idx': idx} return output
def count_parameters(model): total_params = 0 for (name, parameter) in model.named_parameters(): if (not parameter.requires_grad): continue params = parameter.numel() print(name, params) total_params += params print(f'Total Trainable Params: {total_params}') return total_params
class Runtime(): def __init__(self, worker_pool_factory): self._get_worker_pool = worker_pool_factory def inline(cls): return cls(inline_pool_factory) def run(self, query, args, combiner=union_combiner, randomize=True, chunksize=1, progress=False, profile=False, print_error=True): with perf_count('Executing query in Runtime', enable=profile): with _WorkerPoolContext(self._get_worker_pool(query)) as pool: total_work = len(args) with tqdm(total=total_work, disable=(not progress)) as pbar: with perf_count('Executing in workers', enable=profile): args_with_err = [] with perf_count('Dispatching tasks', enable=profile): if randomize: random.shuffle(args) async_results = pool.map(_create_tasks(args, chunksize), _get_callback(pbar, args_with_err, print_error)) combined_result = None for future in async_results: try: r = future.get() except TaskException: continue if (combined_result is None): combined_result = r else: combined_result = combiner(combined_result, r) if ((combined_result is None) and (total_work > 0)): raise RekallRuntimeException('All tasks failed!') return (combined_result, args_with_err) def get_result_iterator(self, query, args, randomize=True, chunksize=1, print_error=True, dispatch_size=mp.cpu_count()): with _WorkerPoolContext(self._get_worker_pool(query)) as pool: args_with_err = [] if randomize: random.shuffle(args) tasks = _create_tasks(args, chunksize) if ((dispatch_size is None) or (dispatch_size <= 0)): dispatch_size = len(tasks) outstanding_tasks = tasks async_results = [] num_finished_tasks = 0 while (num_finished_tasks < len(tasks)): num_to_yield = len(async_results) if ((num_to_yield <= (dispatch_size / 2)) and (len(outstanding_tasks) > 0)): task_batch = outstanding_tasks[:dispatch_size] outstanding_tasks = outstanding_tasks[dispatch_size:] async_results.extend(pool.map(task_batch, _get_callback(None, args_with_err, print_error))) if randomize: future_to_yield = _pop_future_to_yield(async_results) else: future_to_yield = async_results.pop(0) num_finished_tasks += 1 try: r = future_to_yield.get() except TaskException: continue (yield r) if (len(args_with_err) > 0): raise RekallRuntimeException('The following tasks failed: {0}'.format(args_with_err))
def are_projectively_equivalent(P, Q, base_ring): from sage.matrix.constructor import matrix return (matrix(base_ring, [P, Q]).rank() < 2)
.xfail(_IS_WASM, reason='cannot start subprocess') def test_imports_strategies(): good_import = '\n from sklearn.experimental import enable_halving_search_cv\n from sklearn.model_selection import HalvingGridSearchCV\n from sklearn.model_selection import HalvingRandomSearchCV\n ' assert_run_python_script(textwrap.dedent(good_import)) good_import_with_model_selection_first = '\n import sklearn.model_selection\n from sklearn.experimental import enable_halving_search_cv\n from sklearn.model_selection import HalvingGridSearchCV\n from sklearn.model_selection import HalvingRandomSearchCV\n ' assert_run_python_script(textwrap.dedent(good_import_with_model_selection_first)) bad_imports = "\n import pytest\n\n with pytest.raises(ImportError, match='HalvingGridSearchCV is experimental'):\n from sklearn.model_selection import HalvingGridSearchCV\n\n import sklearn.experimental\n with pytest.raises(ImportError, match='HalvingRandomSearchCV is experimental'):\n from sklearn.model_selection import HalvingRandomSearchCV\n " assert_run_python_script(textwrap.dedent(bad_imports))
def script_model_defines_attr(script_model, attr): script_attr = getattr(script_model, attr, None) if (script_attr is None): return False default_attr = get_function_from_type(torch.jit.RecursiveScriptModule, attr) if (default_attr is None): return False return (script_attr != default_attr)
class CollectionNode(Node): def __init__(self, tag, value, start_mark=None, end_mark=None, flow_style=None): self.tag = tag self.value = value self.start_mark = start_mark self.end_mark = end_mark self.flow_style = flow_style
def do_structure(cfg): if isinstance(cfg, CfgNode): model = build_model(cfg) else: model = instantiate(cfg.model) logger.info(('Model Structure:\n' + str(model)))
def write_model_card(model_card_dir, src_lang, tgt_lang, model_name): texts = {'en': "Machine learning is great, isn't it?", 'ru': ' - , ?', 'de': 'Maschinelles Lernen ist groartig, nicht wahr?'} scores = {'wmt19-de-en-6-6-base': [0, 38.37], 'wmt19-de-en-6-6-big': [0, 39.9]} pair = f'{src_lang}-{tgt_lang}' readme = f''' --- language: - {src_lang} - {tgt_lang} thumbnail: tags: - translation - wmt19 - allenai license: apache-2.0 datasets: - wmt19 metrics: - bleu --- # FSMT ## Model description This is a ported version of fairseq-based [wmt19 transformer]( for {src_lang}-{tgt_lang}. For more details, please, see [Deep Encoder, Shallow Decoder: Reevaluating the Speed-Quality Tradeoff in Machine Translation]( 2 models are available: * [wmt19-de-en-6-6-big]( * [wmt19-de-en-6-6-base]( ## Intended uses & limitations #### How to use ```python from transformers import FSMTForConditionalGeneration, FSMTTokenizer mname = "allenai/{model_name}" tokenizer = FSMTTokenizer.from_pretrained(mname) model = FSMTForConditionalGeneration.from_pretrained(mname) input = "{texts[src_lang]}" input_ids = tokenizer.encode(input, return_tensors="pt") outputs = model.generate(input_ids) decoded = tokenizer.decode(outputs[0], skip_special_tokens=True) print(decoded) # {texts[tgt_lang]} ``` #### Limitations and bias ## Training data Pretrained weights were left identical to the original model released by allenai. For more details, please, see the [paper]( ## Eval results Here are the BLEU scores: model | transformers -------| {model_name} | {scores[model_name][1]} The score was calculated using this code: ```bash git clone cd transformers export PAIR={pair} export DATA_DIR=data/$PAIR export SAVE_DIR=data/$PAIR export BS=8 export NUM_BEAMS=5 mkdir -p $DATA_DIR sacrebleu -t wmt19 -l $PAIR --echo src > $DATA_DIR/val.source sacrebleu -t wmt19 -l $PAIR --echo ref > $DATA_DIR/val.target echo $PAIR PYTHONPATH="src:examples/seq2seq" python examples/seq2seq/run_eval.py allenai/{model_name} $DATA_DIR/val.source $SAVE_DIR/test_translations.txt --reference_path $DATA_DIR/val.target --score_path $SAVE_DIR/test_bleu.json --bs $BS --task translation --num_beams $NUM_BEAMS ``` ## Data Sources - [training, etc.]( - [test set]( ### BibTeX entry and citation info ``` {{kasai2020deep, title={{Deep Encoder, Shallow Decoder: Reevaluating the Speed-Quality Tradeoff in Machine Translation}}, author={{Jungo Kasai and Nikolaos Pappas and Hao Peng and James Cross and Noah A. Smith}}, year={{2020}}, eprint={{2006.10369}}, archivePrefix={{arXiv}}, primaryClass={{cs.CL}} }} ``` ''' model_card_dir.mkdir(parents=True, exist_ok=True) path = os.path.join(model_card_dir, 'README.md') print(f'Generating {path}') with open(path, 'w', encoding='utf-8') as f: f.write(readme)
def DrawGLScene(): glClear((GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT)) glLoadIdentity() glutSwapBuffers()
def step_decay(optimizer, step, lr, decay_step, gamma): lr = (lr * (gamma ** (step / decay_step))) for param_group in optimizer.param_groups: param_group['lr'] = lr return lr
def test_channel_first() -> None: env = DummyAtari(grayscale=False) assert env.observation_space.shape (width, height, channel) = env.observation_space.shape wrapper = ChannelFirst(env) (observation, _) = wrapper.reset() assert (observation.shape == (channel, width, height)) (observation, _, _, _, _) = wrapper.step(wrapper.action_space.sample()) assert (observation.shape == (channel, width, height)) dqn = DQNConfig().create() dqn.build_with_env(wrapper) dqn.predict(np.expand_dims(observation, axis=0))
class SemiSupervisedTrainingPlan(TrainingPlan): def __init__(self, module: BaseModuleClass, n_classes: int, *, classification_ratio: int=50, lr: float=0.001, weight_decay: float=1e-06, n_steps_kl_warmup: Union[(int, None)]=None, n_epochs_kl_warmup: Union[(int, None)]=400, reduce_lr_on_plateau: bool=False, lr_factor: float=0.6, lr_patience: int=30, lr_threshold: float=0.0, lr_scheduler_metric: Literal[('elbo_validation', 'reconstruction_loss_validation', 'kl_local_validation')]='elbo_validation', **loss_kwargs): super().__init__(module=module, lr=lr, weight_decay=weight_decay, n_steps_kl_warmup=n_steps_kl_warmup, n_epochs_kl_warmup=n_epochs_kl_warmup, reduce_lr_on_plateau=reduce_lr_on_plateau, lr_factor=lr_factor, lr_patience=lr_patience, lr_threshold=lr_threshold, lr_scheduler_metric=lr_scheduler_metric, **loss_kwargs) self.loss_kwargs.update({'classification_ratio': classification_ratio}) self.n_classes = n_classes def log_with_mode(self, key: str, value: Any, mode: str, **kwargs): self.log(f'{mode}_{key}', value, **kwargs) def compute_and_log_metrics(self, loss_output: LossOutput, metrics: dict[(str, ElboMetric)], mode: str): super().compute_and_log_metrics(loss_output, metrics, mode) if (loss_output.classification_loss is None): return classification_loss = loss_output.classification_loss true_labels = loss_output.true_labels.squeeze() logits = loss_output.logits predicted_labels = torch.argmax(logits, dim=(- 1)) accuracy = tmf.classification.multiclass_accuracy(predicted_labels, true_labels, self.n_classes, average='micro') f1 = tmf.classification.multiclass_f1_score(predicted_labels, true_labels, self.n_classes, average='micro') ce = tmf.classification.multiclass_calibration_error(logits, true_labels, self.n_classes) self.log_with_mode(METRIC_KEYS.CLASSIFICATION_LOSS_KEY, classification_loss, mode, on_step=False, on_epoch=True, batch_size=loss_output.n_obs_minibatch) self.log_with_mode(METRIC_KEYS.ACCURACY_KEY, accuracy, mode, on_step=False, on_epoch=True, batch_size=loss_output.n_obs_minibatch) self.log_with_mode(METRIC_KEYS.F1_SCORE_KEY, f1, mode, on_step=False, on_epoch=True, batch_size=loss_output.n_obs_minibatch) self.log_with_mode(METRIC_KEYS.CALIBRATION_ERROR_KEY, ce, mode, on_step=False, on_epoch=True, batch_size=loss_output.n_obs_minibatch) def training_step(self, batch, batch_idx): if (len(batch) == 2): full_dataset = batch[0] labelled_dataset = batch[1] else: full_dataset = batch labelled_dataset = None if ('kl_weight' in self.loss_kwargs): self.loss_kwargs.update({'kl_weight': self.kl_weight}) input_kwargs = {'feed_labels': False, 'labelled_tensors': labelled_dataset} input_kwargs.update(self.loss_kwargs) (_, _, loss_output) = self.forward(full_dataset, loss_kwargs=input_kwargs) loss = loss_output.loss self.log('train_loss', loss, on_epoch=True, batch_size=loss_output.n_obs_minibatch, prog_bar=True) self.compute_and_log_metrics(loss_output, self.train_metrics, 'train') return loss def validation_step(self, batch, batch_idx): if (len(batch) == 2): full_dataset = batch[0] labelled_dataset = batch[1] else: full_dataset = batch labelled_dataset = None input_kwargs = {'feed_labels': False, 'labelled_tensors': labelled_dataset} input_kwargs.update(self.loss_kwargs) (_, _, loss_output) = self.forward(full_dataset, loss_kwargs=input_kwargs) loss = loss_output.loss self.log('validation_loss', loss, on_epoch=True, batch_size=loss_output.n_obs_minibatch) self.compute_and_log_metrics(loss_output, self.val_metrics, 'validation')
def _train_test_metrics(args_namespace): return train_test_metrics(args_namespace.train_dataset_path, args_namespace.test_dataset_path, args_namespace.output_path, args_namespace.config_path, args_namespace.exclude_slot_metrics, args_namespace.include_errors, args_namespace.verbosity)
def test_load_csvs_folder_does_not_exist(): error_message = re.escape("The folder 'demo/' cannot be found.") with pytest.raises(ValueError, match=error_message): load_csvs('demo/')
def apply_filters(sentence, filters): for f in filters: sentence = f(sentence) return sentence
(config_name='config', config_path='conf') def main(cfg: DictConfig) -> None: logging.info(('\n' + OmegaConf.to_yaml(cfg))) train_device = _get_device(cfg.framework.gpu) env_device = _get_device(cfg.framework.env_gpu) logging.info(('Using training device %s.' % str(train_device))) logging.info(('Using env device %s.' % str(env_device))) gripper_mode = Discrete() if (cfg.method.name == 'PathARM'): arm_action_mode = TrajectoryActionMode(cfg.method.trajectory_points) else: arm_action_mode = EndEffectorPoseViaPlanning() action_mode = MoveArmThenGripper(arm_action_mode, gripper_mode) task_files = [t.replace('.py', '') for t in os.listdir(task.TASKS_PATH) if ((t != '__init__.py') and t.endswith('.py'))] if (cfg.rlbench.task not in task_files): raise ValueError(('Task %s not recognised!.' % cfg.rlbench.task)) task_class = task_file_to_task_class(cfg.rlbench.task) cfg.rlbench.cameras = (cfg.rlbench.cameras if isinstance(cfg.rlbench.cameras, ListConfig) else [cfg.rlbench.cameras]) obs_config = _create_obs_config(cfg.rlbench.cameras, cfg.rlbench.camera_resolution) env = CustomRLBenchEnv(task_class=task_class, observation_config=obs_config, action_mode=action_mode, dataset_root=cfg.rlbench.demo_path, episode_length=cfg.rlbench.episode_length, headless=True, time_in_state=True) cwd = os.getcwd() logging.info(('CWD:' + os.getcwd())) existing_seeds = len(list(filter((lambda x: ('seed' in x)), os.listdir(cwd)))) for seed in range(existing_seeds, (existing_seeds + cfg.framework.seeds)): logging.info(('Starting seed %d.' % seed)) run_seed(cfg, env, cfg.rlbench.cameras, train_device, env_device, seed)
_utils.test(debug=True) def test_assign_ann(): def func_ann(): a: ti.i32 = 1 b: ti.f32 = a assert (a == 1) assert (b == 1.0) func_ann()
class SeqCLDataset(th.utils.data.Dataset): def __init__(self, data: Sequence): super().__init__() self.d = data def __getitem__(self, node_id): item = self.d.get_tokens(node_id) neighbours = self.d.neighbours[node_id] k = np.random.choice(neighbours, 1) item = self.d.get_NB_tokens(item, k[0]) return item def __len__(self): return self.d.n_nodes
def s_load(file_obj): cur_elt = [] for line in file_obj: if (line == b'\n'): encoded_elt = b''.join(cur_elt) try: pickled_elt = base64.b64decode(encoded_elt) elt = loads(pickled_elt) except EOFError: print('EOF found while unpickling data') print(pickled_elt) raise StopIteration cur_elt = [] (yield elt) else: cur_elt.append(line)
.parametrize('embedding_size,cross_num,hidden_size,sparse_feature_num', [(8, 0, (32,), 2), (8, 1, (32,), 2)]) def test_DCNMix(embedding_size, cross_num, hidden_size, sparse_feature_num): model_name = 'DCN-Mix' sample_size = SAMPLE_SIZE (x, y, feature_columns) = get_test_data(sample_size, sparse_feature_num=sparse_feature_num, dense_feature_num=sparse_feature_num) model = DCNMix(linear_feature_columns=feature_columns, dnn_feature_columns=feature_columns, cross_num=cross_num, dnn_hidden_units=hidden_size, dnn_dropout=0.5, device=get_device()) check_model(model, model_name, x, y)
def lr_grad(w, X, y, lam=0): y[(y == 0)] = (- 1) z = torch.sigmoid((y * X.mv(w))) return (X.t().mv(((z - 1) * y)) + ((lam * X.size(0)) * w))
def register_optimizer_class(cls, name=None): _init_optimizer_classes_dict() if (not name): name = cls.__name__ _check_valid_optimizer(cls) assert (name.lower() not in _OptimizerClassesDict) _OptimizerClassesDict[name.lower()] = cls if name.endswith('Optimizer'): name = name[:(- len('Optimizer'))] assert (name.lower() not in _OptimizerClassesDict) _OptimizerClassesDict[name.lower()] = cls
class AttnSkipUpBlock2D(nn.Module): def __init__(self, in_channels: int, prev_output_channel: int, out_channels: int, temb_channels: int, dropout: float=0.0, num_layers: int=1, resnet_eps: float=1e-06, resnet_time_scale_shift: str='default', resnet_act_fn: str='swish', resnet_pre_norm: bool=True, attn_num_head_channels=1, attention_type='default', output_scale_factor=np.sqrt(2.0), upsample_padding=1, add_upsample=True): super().__init__() self.attentions = nn.ModuleList([]) self.resnets = nn.ModuleList([]) self.attention_type = attention_type for i in range(num_layers): res_skip_channels = (in_channels if (i == (num_layers - 1)) else out_channels) resnet_in_channels = (prev_output_channel if (i == 0) else out_channels) self.resnets.append(ResnetBlock2D(in_channels=(resnet_in_channels + res_skip_channels), out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min((resnet_in_channels + (res_skip_channels // 4)), 32), groups_out=min((out_channels // 4), 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm)) self.attentions.append(AttentionBlock(out_channels, num_head_channels=attn_num_head_channels, rescale_output_factor=output_scale_factor, eps=resnet_eps)) self.upsampler = FirUpsample2D(in_channels, out_channels=out_channels) if add_upsample: self.resnet_up = ResnetBlock2D(in_channels=out_channels, out_channels=out_channels, temb_channels=temb_channels, eps=resnet_eps, groups=min((out_channels // 4), 32), groups_out=min((out_channels // 4), 32), dropout=dropout, time_embedding_norm=resnet_time_scale_shift, non_linearity=resnet_act_fn, output_scale_factor=output_scale_factor, pre_norm=resnet_pre_norm, use_nin_shortcut=True, up=True, kernel='fir') self.skip_conv = nn.Conv2d(out_channels, 3, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) self.skip_norm = torch.nn.GroupNorm(num_groups=min((out_channels // 4), 32), num_channels=out_channels, eps=resnet_eps, affine=True) self.act = nn.SiLU() else: self.resnet_up = None self.skip_conv = None self.skip_norm = None self.act = None def forward(self, hidden_states, res_hidden_states_tuple, temb=None, skip_sample=None): for resnet in self.resnets: res_hidden_states = res_hidden_states_tuple[(- 1)] res_hidden_states_tuple = res_hidden_states_tuple[:(- 1)] hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1) hidden_states = resnet(hidden_states, temb) hidden_states = self.attentions[0](hidden_states) if (skip_sample is not None): skip_sample = self.upsampler(skip_sample) else: skip_sample = 0 if (self.resnet_up is not None): skip_sample_states = self.skip_norm(hidden_states) skip_sample_states = self.act(skip_sample_states) skip_sample_states = self.skip_conv(skip_sample_states) skip_sample = (skip_sample + skip_sample_states) hidden_states = self.resnet_up(hidden_states, temb) return (hidden_states, skip_sample)
(Output('anomaly-attribute-options', 'children'), Output('anomaly_exception_modal', 'is_open'), Output('anomaly_exception_modal_content', 'children'), [Input('anomaly-btn', 'n_clicks'), Input('anomaly_exception_modal_close', 'n_clicks')], [State('log-type-select', 'value'), State('attribute-name-options', 'value'), State('file-select', 'value'), State('parsing-algo-select', 'value'), State('vectorization-algo-select', 'value'), State('categorical-encoder-select', 'value'), State('ad-algo-select', 'value'), State('time-interval', 'value'), State('ad-param-table', 'children'), State('ad-parsing-param-table', 'children')]) def click_run(btn_click, modal_close, log_type, attributes, filename, parsing_algo, vectorization_algo, categorical_encoder, ad_algo, time_interval, ad_param_table, parsing_param_table): ctx = dash.callback_context if ctx.triggered: prop_id = ctx.triggered[0]['prop_id'].split('.')[0] if (prop_id == 'anomaly-btn'): try: interval_map = {0: '1s', 1: '1min', 2: '1h', 3: '1d'} freq = interval_map[time_interval] file_path = os.path.join(file_manager.base_directory, filename) ad_params = log_anomaly_demo.parse_parameters(param_info=log_anomaly_demo.get_parameter_info(ad_algo), params={p['Parameter']: p['Value'] for p in ad_param_table['props']['data'] if p['Parameter']}) parsing_params = LogPattern().parse_parameters(param_info=LogPattern().get_parameter_info(parsing_algo), params={p['Parameter']: p['Value'] for p in parsing_param_table['props']['data'] if p['Parameter']}) config = _ad_config_sample() config.open_set_data_loader_config.filepath = file_path config.open_set_data_loader_config.dataset_name = log_type config.feature_extractor_config.group_by_category = attributes config.feature_extractor_config.group_by_time = freq config.log_parser_config.parsing_algorithm = parsing_algo config_class = LogPattern().get_config_class(parsing_algo) config.log_parser_config.parsing_algo_params = config_class.from_dict(parsing_params) config.log_vectorizer_config.algo_name = vectorization_algo config.categorical_encoder_config.algo_name = categorical_encoder config.anomaly_detection_config.algo_name = ad_algo config_class = log_anomaly_demo.get_config_class(ad_algo) config.anomaly_detection_config.algo_params = config_class.from_dict(ad_params) log_anomaly_demo.execute_anomaly_detection(config) return (create_attribute_component(log_anomaly_demo.get_attributes()), False, '') except Exception as error: return (html.Div(), True, str(error)) elif (prop_id == 'anomaly_exception_modal_close'): return (html.Div(), False, '') else: return (html.Div(), False, '')
def shift_stats_container(sc, num_of_shifting_factors): shifting_factor = np.random.random(num_of_shifting_factors) shifted_sc = shift_statistics(sc, shifting_factor) return (shifted_sc, shifting_factor)
def update_plot(policy, max_length=np.inf): queue.put(['demo', policy.get_param_values(), max_length])
def init_params(net): for m in net.modules(): if (isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d)): init.kaiming_normal_(m.weight, mode='fan_out') if (m.bias.data is not None): init.constant_(m.bias, 0) elif isinstance(m, nn.BatchNorm2d): init.constant_(m.weight, 1) init.constant_(m.bias, 0) elif isinstance(m, nn.Linear): if (m.bias.data is not None): init.constant_(m.bias, 0)
class VQVAE(tfk.Model): def __init__(self, encoder, decoder, codebook_size, beta=0.25): super(VQVAE, self).__init__() self.encoder = encoder self.decoder = decoder self.quantizer = VectorQuantizerEMA(codebook_size) self.beta = beta def quantize(self, x): if (not self.built): self(x) z_e = self.encoder(x) flat_z_e = tf.reshape(z_e, shape=((- 1), z_e.shape[(- 1)])) codebook = self.quantizer.codebook distances = ((tf.reduce_sum((flat_z_e ** 2), axis=1, keepdims=True) - (2 * tf.matmul(flat_z_e, codebook))) + tf.reduce_sum((codebook ** 2), axis=0, keepdims=True)) encoding_indices = tf.argmax((- distances), axis=1) encoding_indices = tf.reshape(encoding_indices, tf.concat([tf.shape(z_e)[:(- 1)], [1]], axis=0)) return encoding_indices def dequantize(self, x): flat_x = tf.reshape(x, shape=((- 1), x.shape[(- 1)])) codebook = tf.transpose(self.quantizer.codebook) z_q = tf.nn.embedding_lookup(codebook, flat_x) z_q = tf.reshape(z_q, tf.concat([tf.shape(x)[:(- 1)], [(- 1)]], axis=0)) x_rec = self.decoder(z_q) return x_rec def call(self, x, training=False): z_e = self.encoder(x) z_q = self.quantizer(z_e, training=training) commitment_loss = tf.reduce_mean(tf.square((tf.stop_gradient(z_q) - z_e))) self.add_loss((self.beta * commitment_loss)) x_rec = self.decoder(z_q) return x_rec
class PythonInliner(ast.NodeTransformer): def __init__(self, target_id, target_ast): self.target_id = target_id self.target_ast = target_ast def visit_Name(self, node: ast.AST): if (node.id == self.target_id): return ast.copy_location(self.target_ast, node) else: return self.generic_visit(node)
def random_hex(): r = (lambda : np.random.randint(0, 255)) return ('#%02X%02X%02X' % (r(), r(), r()))
def get_parent(config: Dict[(str, Any)]) -> Optional[str]: if (config['load_from'] is None): return None return config['load_from'].rsplit('/', maxsplit=1)[0]
class BaseTransformersCLICommand(ABC): def register_subcommand(parser: ArgumentParser): raise NotImplementedError() def run(self): raise NotImplementedError()
def xcorr_slow(x, kernel): batch = x.size()[0] out = [] for i in range(batch): px = x[i] pk = kernel[i] px = px.view(1, px.size()[0], px.size()[1], px.size()[2]) pk = pk.view((- 1), px.size()[1], pk.size()[1], pk.size()[2]) po = F.conv2d(px, pk) out.append(po) out = torch.cat(out, 0) return out
def _map_slice_value_raw(v: Union[(None, slice, int, numpy.number, numpy.ndarray, Tensor[T])]) -> Union[(None, slice, int, numpy.number, T)]: if (v is None): return None if isinstance(v, slice): return slice(_map_slice_value_raw(v.start), _map_slice_value_raw(v.stop), _map_slice_value_raw(v.step)) if isinstance(v, (int, numpy.number)): return v if isinstance(v, numpy.ndarray): assert (v.ndim <= 1), f'strided_slice: expect scalar or vector, got array with shape {v.shape}' return v if isinstance(v, Tensor): assert (len(v.dims) <= 1), f'strided_slice: expect scalar or vector, got Tensor with dims {v.dims}' return v.raw_tensor raise TypeError(f'strided_slice: got unexpected value of type {type(v).__name__}')
class SageNotebookInteractiveShell(SageShellOverride, InteractiveShell): def init_display_formatter(self): from sage.repl.rich_output.backend_ipython import BackendIPythonNotebook backend = BackendIPythonNotebook() backend.get_display_manager().switch_backend(backend, shell=self)
_module() class VFNet(SingleStageDetector): 'Implementation of `VarifocalNet\n (VFNet).< def __init__(self, backbone, neck, bbox_head, train_cfg=None, test_cfg=None, pretrained=None, init_cfg=None): super(VFNet, self).__init__(backbone, neck, bbox_head, train_cfg, test_cfg, pretrained, init_cfg)
def test_get_path_accessible(accessible_path, workspace_root): workspace = Workspace(workspace_root, True) full_path = workspace.get_path(accessible_path) assert full_path.is_absolute() assert full_path.is_relative_to(workspace_root)
class ModelArguments(): model_name_or_path: str = field(metadata={'help': 'Path to pretrained model or model identifier from huggingface.co/models'}) config_name: Optional[str] = field(default=None, metadata={'help': 'Pretrained config name or path if not the same as model_name'}) tokenizer_name: Optional[str] = field(default=None, metadata={'help': 'Pretrained tokenizer name or path if not the same as model_name'}) cache_dir: Optional[str] = field(default=None, metadata={'help': 'Where to store the pretrained models downloaded from huggingface.co'}) use_fast_tokenizer: bool = field(default=True, metadata={'help': 'Whether to use one of the fast tokenizer (backed by the tokenizers library) or not.'}) model_revision: str = field(default='main', metadata={'help': 'The specific model version to use (can be a branch name, tag name or commit id).'}) use_auth_token: bool = field(default=False, metadata={'help': 'Will use the token generated when running `transformers-cli login` (necessary to use this script with private models).'}) resize_position_embeddings: Optional[bool] = field(default=None, metadata={'help': "Whether to automatically resize the position embeddings if `max_source_length` exceeds the model's position embeddings."}) from_pt: bool = field(default=False, metadata={'help': 'Whether to load the model checkpoint from .pt file'}) multiencoder_type: Optional[str] = field(default=None, metadata={'help': "Currently only 'bart' is supported"}) multiencoder_max_num_chunks: Optional[int] = field(default=None, metadata={'help': 'Max passages/encoders to use in multiencoder'}) multiencoder_stride: bool = field(default=False, metadata={'help': 'Whether to stride'})
def test_orthogonal_procrustes_checkfinite_exception(): np.random.seed(1234) (m, n) = (2, 3) A_good = np.random.randn(m, n) B_good = np.random.randn(m, n) for bad_value in (np.inf, (- np.inf), np.nan): A_bad = A_good.copy() A_bad[(1, 2)] = bad_value B_bad = B_good.copy() B_bad[(1, 2)] = bad_value for (A, B) in ((A_good, B_bad), (A_bad, B_good), (A_bad, B_bad)): assert_raises(ValueError, orthogonal_procrustes, A, B)
def srwl_uti_cryst_ASF(_s, _mat='Si'): fa = None fa0 = 0 if (_mat == 'Si'): fa = [6.2915, 2.4386, 3.0353, 32.3337, 1.9891, 0.6785, 1.541, 81.6937, 1.1407] fa0 = 13.985 else: raise Exception(strMatDataNotDefined) s2 = (_s * _s) if (s2 != 0): f0 = (((((fa[0] * exp(((- fa[1]) * s2))) + (fa[2] * exp(((- fa[3]) * s2)))) + (fa[4] * exp(((- fa[5]) * s2)))) + (fa[6] * exp(((- fa[7]) * s2)))) + fa[8]) else: f0 = fa0 return f0
def nparray(named_tensor): proto = named_tensor.transformer_metadata.pop() metadata = {'int_to_float': proto.int_to_float, 'int_list': proto.int_list, 'bool_list': proto.bool_list} array_shape = tuple(metadata['int_list']) flat_array = np.frombuffer(named_tensor.data_bytes, dtype=np.float32) nparray = np.reshape(flat_array, newshape=array_shape, order='C') return nparray
class Table(object): def __init__(self, dataset, version): self.dataset = dataset self.version = version self.name = f'{self.dataset}_{self.version}' L.info(f'start building data {self.name}...') self.data = pd.read_pickle(((DATA_ROOT / self.dataset) / f'{self.version}.pkl')) self.data_size_mb = ((self.data.values.nbytes / 1024) / 1024) self.row_num = self.data.shape[0] self.col_num = len(self.data.columns) self.parse_columns() L.info(f'build finished: {self}') def parse_columns(self): self.columns = OrderedDict([(col, Column(col, self.data[col])) for col in self.data.columns]) def __repr__(self): return f'''Table {self.name} ({self.row_num} rows, {self.data_size_mb:.2f}MB, columns: {os.linesep.join([repr(c) for c in self.columns.values()])})''' def get_minmax_dict(self): minmax_dict = {} for (i, col) in enumerate(self.columns.values()): minmax_dict[i] = (col.minval, col.maxval) return minmax_dict def normalize(self, scale=1): data = copy.deepcopy(self.data) for (cname, col) in self.columns.items(): data[cname] = (col.normalize(data[cname].values) * scale) return data def digitalize(self): data = copy.deepcopy(self.data) for (cname, col) in self.columns.items(): if is_categorical(col.dtype): data[cname] = col.discretize(data[cname]) elif col.has_nan: data[cname].fillna(0, inplace=True) return data def get_max_muteinfo_order(self): order = [] max_entropy = float('-inf') first_col = None for c in self.columns.keys(): e = entropy(self.data[c].value_counts()) if (e > max_entropy): first_col = c max_entropy = e assert (first_col is not None), (first_col, max_entropy) order.append(first_col) sep = '|' chosen_data = (self.data[first_col].astype(str) + sep) while (len(order) < self.col_num): max_muinfo = float('-inf') next_col = None for c in self.columns.keys(): if (c in order): continue m = mutual_info_score(chosen_data, self.data[c]) if (m > max_muinfo): next_col = c max_muinfo = m assert (next_col is not None), (next_col, max_entropy) order.append(next_col) chosen_data = ((chosen_data + sep) + self.data[next_col].astype(str)) return (order, [self.data.columns.get_loc(c) for c in order]) def get_muteinfo(self, digital_data=None): data = (digital_data if (digital_data is not None) else self.digitalize()) muteinfo_dict = {} for c1 in self.columns.keys(): muteinfo_dict[c1] = {} for c2 in self.columns.keys(): if ((c1 != c2) and (c2 in muteinfo_dict)): assert (c1 in muteinfo_dict[c2]), muteinfo_dict.keys() muteinfo_dict[c1][c2] = muteinfo_dict[c2][c1] else: muteinfo_dict[c1][c2] = mutual_info_score(data[c1], data[c2]) return pd.DataFrame().from_dict(muteinfo_dict)
def get_cpp_decl_type(typename, ensure_temp_safe=True): if ensure_temp_safe: typename = TEMP_SAFE_CPP_DECL_TYPE.get(typename, typename) return typename
class Fantasizer(GreedyAcquisitionFunctionBuilder[FantasizerModelOrStack]): def __init__(self, base_acquisition_function_builder: Optional[(AcquisitionFunctionBuilder[SupportsPredictJoint] | SingleModelAcquisitionBuilder[SupportsPredictJoint])]=None, fantasize_method: str='KB'): tf.debugging.Assert((fantasize_method in ['KB', 'sample']), [tf.constant([])]) if (base_acquisition_function_builder is None): base_acquisition_function_builder = ExpectedImprovement() if isinstance(base_acquisition_function_builder, SingleModelAcquisitionBuilder): base_acquisition_function_builder = base_acquisition_function_builder.using(OBJECTIVE) self._builder = base_acquisition_function_builder self._fantasize_method = fantasize_method self._base_acquisition_function: Optional[AcquisitionFunction] = None self._fantasized_acquisition: Optional[AcquisitionFunction] = None self._fantasized_models: Mapping[(Tag, (_fantasized_model | ModelStack[SupportsPredictJoint]))] = {} def _update_base_acquisition_function(self, models: Mapping[(Tag, FantasizerModelOrStack)], datasets: Optional[Mapping[(Tag, Dataset)]]) -> AcquisitionFunction: if (self._base_acquisition_function is not None): self._base_acquisition_function = self._builder.update_acquisition_function(self._base_acquisition_function, models, datasets) else: self._base_acquisition_function = self._builder.prepare_acquisition_function(models, datasets) return self._base_acquisition_function def _update_fantasized_acquisition_function(self, models: Mapping[(Tag, FantasizerModelOrStack)], datasets: Optional[Mapping[(Tag, Dataset)]], pending_points: TensorType) -> AcquisitionFunction: tf.debugging.assert_rank(pending_points, 2) fantasized_data = {tag: _generate_fantasized_data(fantasize_method=self._fantasize_method, model=model, pending_points=pending_points) for (tag, model) in models.items()} if (datasets is None): datasets = fantasized_data else: datasets = {tag: (data + fantasized_data[tag]) for (tag, data) in datasets.items()} if (self._fantasized_acquisition is None): self._fantasized_models = {tag: _generate_fantasized_model(model, fantasized_data[tag]) for (tag, model) in models.items()} self._fantasized_acquisition = self._builder.prepare_acquisition_function(cast(Dict[(Tag, SupportsPredictJoint)], self._fantasized_models), datasets) else: for (tag, model) in self._fantasized_models.items(): if isinstance(model, ModelStack): observations = tf.split(fantasized_data[tag].observations, model._event_sizes, axis=(- 1)) for (submodel, obs) in zip(model._models, observations): submodel.update_fantasized_data(Dataset(fantasized_data[tag].query_points, obs)) else: model.update_fantasized_data(fantasized_data[tag]) self._builder.update_acquisition_function(self._fantasized_acquisition, cast(Dict[(Tag, SupportsPredictJoint)], self._fantasized_models), datasets) return self._fantasized_acquisition def prepare_acquisition_function(self, models: Mapping[(Tag, FantasizerModelOrStack)], datasets: Optional[Mapping[(Tag, Dataset)]]=None, pending_points: Optional[TensorType]=None) -> AcquisitionFunction: for model in models.values(): if (not (isinstance(model, FantasizerModelType) or (isinstance(model, ModelStack) and all((isinstance(m, FantasizerModelType) for m in model._models))))): raise NotImplementedError(f'Fantasizer only works with FastUpdateModel models that also support predict_joint, get_kernel and get_observation_noise, or with ModelStack stacks of such models; received {model.__repr__()}') if (pending_points is None): return self._update_base_acquisition_function(models, datasets) else: return self._update_fantasized_acquisition_function(models, datasets, pending_points) def update_acquisition_function(self, function: AcquisitionFunction, models: Mapping[(Tag, FantasizerModelOrStack)], datasets: Optional[Mapping[(Tag, Dataset)]]=None, pending_points: Optional[TensorType]=None, new_optimization_step: bool=True) -> AcquisitionFunction: if (pending_points is None): return self._update_base_acquisition_function(models, datasets) else: return self._update_fantasized_acquisition_function(models, datasets, pending_points)
class State(Borg): def __init__(self, session: Optional[SparkSession]=None): Borg.__init__(self) if (not hasattr(self, 'logger_set')): self.logger = logger_with_settings() self.logger_set = True if (session is None): if (not hasattr(self, 'session')): self.session = get_spark_session() else: self.session = session
class DebugPrompts(Prompts): def in_prompt_tokens(self, cli=None): return [(Token.Prompt, 'debug: ')] def continuation_prompt_tokens(self, cli=None, width=None): return [(Token.Prompt, '.....: ')] def rewrite_prompt_tokens(self): return [(Token.Prompt, '-----> ')] def out_prompt_tokens(self): return [(Token.OutPrompt, '')]
class IndexVocab(Configurable): ROOT = 0 def __init__(self, *args, **kwargs): super(IndexVocab, self).__init__(*args, **kwargs) self.placeholder = None def generate_placeholder(self): if (self.placeholder is None): self.placeholder = tf.placeholder(tf.int32, shape=[None, None], name=self.name) return self.placeholder def set_feed_dict(self, data, feed_dict): feed_dict[self.placeholder] = data return def setup(self): self.placeholder = None return def index(self, token): return (0 if (token == '_') else int(token)) def depth(self): return None def conll_idx(self): return self._conll_idx def __getitem__(self, key): if isinstance(key, basestring): return int(key) elif isinstance(key, (int, long, np.int32, np.int64)): return str(key) elif hasattr(key, '__iter__'): return [self[k] for k in key] else: raise ValueError('key to BaseVocab.__getitem__ must be (iterable of) string or integer') return
class ResNetABN(nn.Module): def __init__(self, block, layers, num_classes=10, num_bns=2, first_layer_conv=3): self.inplanes = 64 self.num_bns = num_bns self.num_classes = num_classes super(ResNetABN, self).__init__() self.conv1 = conv3x3(3, 64, kernel_size=first_layer_conv) self.bn1 = MultiBatchNorm(64, self.num_bns) self.layer1 = self._make_layer(block, 64, layers[0], stride=1, num_bns=self.num_bns) self.layer2 = self._make_layer(block, 128, layers[1], stride=2, num_bns=self.num_bns) self.layer3 = self._make_layer(block, 256, layers[2], stride=2, num_bns=self.num_bns) self.layer4 = self._make_layer(block, 512, layers[3], stride=2, num_bns=self.num_bns) self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) self.fc = nn.Linear((512 * block.expansion), num_classes) def _make_layer(self, block, planes, blocks, stride=1, num_bns=2): downsample = None if ((stride != 1) or (self.inplanes != (planes * block.expansion))): downsample = TwoInputSequential(Conv2d(self.inplanes, (planes * block.expansion), kernel_size=1, stride=stride, bias=False), MultiBatchNorm((planes * block.expansion), num_bns)) layers = [] layers.append(block(self.inplanes, planes, stride, downsample, num_bns=num_bns)) self.inplanes = (planes * block.expansion) for i in range(1, blocks): layers.append(block(self.inplanes, planes, num_bns=num_bns)) return TwoInputSequential(*layers) def forward(self, x, rf=False, domain_label=None): if (domain_label is None): domain_label = (0 * torch.ones(x.shape[0], dtype=torch.long).cuda()) x = self.conv1(x) (x, _) = self.bn1(x, domain_label) x = F.relu(x) (x, _) = self.layer1(x, domain_label) (x, _) = self.layer2(x, domain_label) (x, _) = self.layer3(x, domain_label) (x, _) = self.layer4(x, domain_label) x = self.avgpool(x) feat = x.view(x.size(0), (- 1)) x = self.fc(feat) if rf: return (feat, x) else: return x
.parametrize(['packet_params', 'expected_params'], [({'nu_line': 0.1, 'next_line_id': 0, 'is_last_line': True}, {'tardis_error': None, 'd_line': 1e+99}), ({'nu_line': 0.2, 'next_line_id': 1, 'is_last_line': False}, {'tardis_error': None, 'd_line': 7.e+17}), ({'nu_line': 0.5, 'next_line_id': 1, 'is_last_line': False}, {'tardis_error': utils.MonteCarloException, 'd_line': 0.0}), ({'nu_line': 0.6, 'next_line_id': 0, 'is_last_line': False}, {'tardis_error': utils.MonteCarloException, 'd_line': 0.0})]) def test_calculate_distance_line(packet_params, expected_params, static_packet, model): nu_line = packet_params['nu_line'] is_last_line = packet_params['is_last_line'] time_explosion = model.time_explosion doppler_factor = frame_transformations.get_doppler_factor(static_packet.r, static_packet.mu, time_explosion) comov_nu = (static_packet.nu * doppler_factor) d_line = 0 obtained_tardis_error = None try: d_line = calculate_distances.calculate_distance_line(static_packet, comov_nu, is_last_line, nu_line, time_explosion) except utils.MonteCarloException: obtained_tardis_error = utils.MonteCarloException assert_almost_equal(d_line, expected_params['d_line']) assert (obtained_tardis_error == expected_params['tardis_error'])
def test_trainable_variables(): (trackable_layer, variables, modules, module_variables) = setup_layer_modules_variables() all_vars = (variables + module_variables) trainable_variables = [v for v in all_vars if v.trainable] assert (to_tensor_set(trackable_layer.trainable_variables) == to_tensor_set(trainable_variables))
.skip(reason='Covered more efficiently by test_train.test_run_experiment') def test_experiment_config_parser(tmp_path): tmp_data_dir = (tmp_path / 'tmpdata') cfg_fname = os.path.join(Config.get_dir(), 'experiments.json') cfg = memcnn.experiment.factory.load_experiment_config(cfg_fname, ['cifar10', 'resnet110']) memcnn.experiment.factory.experiment_config_parser(cfg, str(tmp_data_dir), workers=None)
class NumberConverter(BaseConverter): weight = 50 def __init__(self, map, fixed_digits=0, min=None, max=None, signed=False): if signed: self.regex = self.signed_regex BaseConverter.__init__(self, map) self.fixed_digits = fixed_digits self.min = min self.max = max self.signed = signed def to_python(self, value): if (self.fixed_digits and (len(value) != self.fixed_digits)): raise ValidationError() value = self.num_convert(value) if (((self.min is not None) and (value < self.min)) or ((self.max is not None) and (value > self.max))): raise ValidationError() return value def to_url(self, value): value = self.num_convert(value) if self.fixed_digits: value = (('%%0%sd' % self.fixed_digits) % value) return str(value) def signed_regex(self): return ('-?' + self.regex)
def spline_basis(pseudo: torch.Tensor, kernel_size: torch.Tensor, is_open_spline: torch.Tensor, degree: int) -> Tuple[(torch.Tensor, torch.Tensor)]: return torch.ops.torch_spline_conv.spline_basis(pseudo, kernel_size, is_open_spline, degree)
class Options(): def __init__(self): self.parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter) self.parser.add_argument('--dataset', default='cifar10', help='folder | cifar10 | mnist ') self.parser.add_argument('--dataroot', default='', help='path to dataset') self.parser.add_argument('--batchsize', type=int, default=64, help='input batch size') self.parser.add_argument('--workers', type=int, help='number of data loading workers', default=8) self.parser.add_argument('--droplast', action='store_true', default=True, help='Drop last batch size.') self.parser.add_argument('--isize', type=int, default=32, help='input image size.') self.parser.add_argument('--nc', type=int, default=3, help='input image channels') self.parser.add_argument('--nz', type=int, default=100, help='size of the latent z vector') self.parser.add_argument('--ngf', type=int, default=64) self.parser.add_argument('--ndf', type=int, default=64) self.parser.add_argument('--extralayers', type=int, default=0, help='Number of extra layers on gen and disc') self.parser.add_argument('--device', type=str, default='gpu', help='Device: gpu | cpu') self.parser.add_argument('--gpu_ids', type=str, default='0', help='gpu ids: e.g. 0 0,1,2, 0,2. use -1 for CPU') self.parser.add_argument('--ngpu', type=int, default=1, help='number of GPUs to use') self.parser.add_argument('--name', type=str, default='experiment_name', help='name of the experiment') self.parser.add_argument('--model', type=str, default='ganomaly', help='chooses which model to use. ganomaly') self.parser.add_argument('--display_server', type=str, default=' help='visdom server of the web display') self.parser.add_argument('--display_port', type=int, default=8097, help='visdom port of the web display') self.parser.add_argument('--display_id', type=int, default=0, help='window id of the web display') self.parser.add_argument('--display', action='store_true', help='Use visdom.') self.parser.add_argument('--outf', default='./output', help='folder to output images and model checkpoints') self.parser.add_argument('--manualseed', default=(- 1), type=int, help='manual seed') self.parser.add_argument('--abnormal_class', default='car', help='Anomaly class idx for mnist and cifar datasets') self.parser.add_argument('--proportion', type=float, default=0.1, help='Proportion of anomalies in test set.') self.parser.add_argument('--metric', type=str, default='roc', help='Evaluation metric.') self.parser.add_argument('--print_freq', type=int, default=100, help='frequency of showing training results on console') self.parser.add_argument('--save_image_freq', type=int, default=100, help='frequency of saving real and fake images') self.parser.add_argument('--save_test_images', action='store_true', help='Save test images for demo.') self.parser.add_argument('--load_weights', action='store_true', help='Load the pretrained weights') self.parser.add_argument('--resume', default='', help='path to checkpoints (to continue training)') self.parser.add_argument('--phase', type=str, default='train', help='train, val, test, etc') self.parser.add_argument('--iter', type=int, default=0, help='Start from iteration i') self.parser.add_argument('--niter', type=int, default=15, help='number of epochs to train for') self.parser.add_argument('--beta1', type=float, default=0.5, help='momentum term of adam') self.parser.add_argument('--lr', type=float, default=0.0002, help='initial learning rate for adam') self.parser.add_argument('--w_adv', type=float, default=1, help='Adversarial loss weight') self.parser.add_argument('--w_con', type=float, default=50, help='Reconstruction loss weight') self.parser.add_argument('--w_enc', type=float, default=1, help='Encoder loss weight.') self.isTrain = True self.opt = None def parse(self): self.opt = self.parser.parse_args() self.opt.isTrain = self.isTrain str_ids = self.opt.gpu_ids.split(',') self.opt.gpu_ids = [] for str_id in str_ids: id = int(str_id) if (id >= 0): self.opt.gpu_ids.append(id) if (self.opt.device == 'gpu'): torch.cuda.set_device(self.opt.gpu_ids[0]) args = vars(self.opt) if (self.opt.name == 'experiment_name'): self.opt.name = ('%s/%s' % (self.opt.model, self.opt.dataset)) expr_dir = os.path.join(self.opt.outf, self.opt.name, 'train') test_dir = os.path.join(self.opt.outf, self.opt.name, 'test') if (not os.path.isdir(expr_dir)): os.makedirs(expr_dir) if (not os.path.isdir(test_dir)): os.makedirs(test_dir) file_name = os.path.join(expr_dir, 'opt.txt') with open(file_name, 'wt') as opt_file: opt_file.write(' Options \n') for (k, v) in sorted(args.items()): opt_file.write(('%s: %s\n' % (str(k), str(v)))) opt_file.write(' End \n') return self.opt
def resize_target(target, size): new_target = np.zeros((target.shape[0], size, size), np.int32) for (i, t) in enumerate(target.numpy()): new_target[(i, ...)] = cv2.resize(t, ((size,) * 2), interpolation=cv2.INTER_NEAREST) return torch.from_numpy(new_target).long()
def _get_test_keep_instance_predicate(cfg: CfgNode): general_keep_predicate = _maybe_create_general_keep_instance_predicate(cfg) return general_keep_predicate
class ConvTBC(torch.nn.Module): def __init__(self, in_channels, out_channels, kernel_size, padding=0): super(ConvTBC, self).__init__() self.in_channels = in_channels self.out_channels = out_channels self.kernel_size = _single(kernel_size) self.padding = _single(padding) self.weight = torch.nn.Parameter(torch.Tensor(self.kernel_size[0], in_channels, out_channels)) self.bias = torch.nn.Parameter(torch.Tensor(out_channels)) self.reset_parameters() def reset_parameters(self): nn.init.xavier_normal_(self.weight) nn.init.zeros_(self.bias) def conv_tbc(self, input: Tensor): return torch.conv_tbc(input.contiguous(), self.weight, self.bias, self.padding[0]) def forward(self, input: Tensor): return self.conv_tbc(input) def __repr__(self): s = '{name}({in_channels}, {out_channels}, kernel_size={kernel_size}, padding={padding}' if (self.bias is None): s += ', bias=False' s += ')' return s.format(name=self.__class__.__name__, **self.__dict__)
class _TensorMixin(_TensorMixinBase): def from_tensor(x) -> Tensor: assert x.get_shape().is_fully_defined() x_shape = x.get_shape().as_list() return _t.Tensor(name=str(x.op.name), shape=x_shape, batch_dim_axis=None, dtype=x.dtype.name, placeholder=x) def template_from_constant(x, name, dtype=None, shape=None, with_batch_dim=False, sparse_dim=None, feature_dim=None) -> Tensor: import numpy if (dtype is None): if isinstance(x, bool): dtype = 'bool' elif isinstance(x, int): dtype = 'int32' elif isinstance(x, float): dtype = 'float32' elif isinstance(x, numpy.ndarray): dtype = str(x.dtype) else: raise TypeError(('%r: cannot handle value %r of type %r' % (name, x, type(x)))) shape_ = (x.shape if isinstance(x, numpy.ndarray) else ()) if (shape is not None): if (len(shape) > len(shape_) == 0): pass else: assert (len(shape) == len(shape_)), ('%r: shape does not match in ndim, %r vs %r' % (name, shape, shape_)) else: shape = shape_ dim_tags = [] for (i, d) in enumerate(shape): d_ = (shape_[i] if (len(shape_) > 0) else None) if isinstance(d, Dim): if (len(shape_) > 0): assert (d.dimension == d_) elif isinstance(d, int): if (len(shape_) > 0): assert (d == d_) d = Dim(kind=(Dim.Types.Spatial if (i < (len(shape) - 1)) else Dim.Types.Feature), description=('%s:static:%i' % (name, i)), auto_generated=True, dimension=d) else: raise TypeError(('%r shape[%i] invalid type %r in shape %r' % (name, i, type(d), shape))) dim_tags.append(d) if (with_batch_dim and (batch_dim not in dim_tags)): dim_tags.insert(0, batch_dim) return _t.Tensor(name=name, dim_tags=dim_tags, dtype=dtype, sparse_dim=sparse_dim, feature_dim=feature_dim) def _handle_extra_kwargs(self, *, shape=None, sparse=None, dim=NotSpecified, batch_dim_axis=NotSpecified, dim_tags=None, placeholder=None, size_placeholder=None, auto_create_placeholders=False, vocab=None, same_dim_tags_as=None, **kwargs): assert isinstance(self, _t.Tensor) (shape, sparse, dim, batch_dim_axis, dim_tags) if (vocab is not None): from returnn.datasets.util.vocabulary import Vocabulary if isinstance(vocab, str): vocab = Vocabulary(vocab) elif isinstance(vocab, dict): vocab = Vocabulary.create_vocab(**vocab) assert isinstance(vocab, Vocabulary) assert self.sparse, ('%s should represent indices of %s' % (self, vocab)) assert (self.dim == vocab.num_labels), ('%s dims do not match with vocab %s' % (self, vocab)) self.sparse_dim.vocab = vocab if kwargs: self._extra = _TensorExtra(tensor=self, **kwargs) if size_placeholder: self.size_placeholder = size_placeholder if same_dim_tags_as: for (_axis, _dim_tag) in sorted(same_dim_tags_as.items()): _axis = self.get_axis_from_description(_axis) assert isinstance(_dim_tag, Dim) base_tag = self._dims[_axis] if (base_tag != _dim_tag): base_tag.declare_same_as(_dim_tag) self._dims = ((self._dims[:_axis] + (_dim_tag,)) + self._dims[(_axis + 1):]) if (placeholder is not None): self.raw_tensor = placeholder elif auto_create_placeholders: from returnn.tf.frontend_low_level._backend import TFBackend self.raw_tensor = TFBackend.create_placeholder_raw(self) _auto_create_size_placeholders_on_dim_tags(name=self.name, dim_tags=self._dims) self._adapt_batch_consistent_dim_tags() self.sanity_check(assume_complete=False) def _raw_backend(self) -> Optional[Type[Backend]]: import returnn.frontend._backend as _backend_api if (self._raw_tensor is None): return None return _backend_api.get_backend_by_raw_tensor_type(type(self._raw_tensor)) def control_flow_ctx(self) -> Optional[ControlFlowContext]: if (not self._extra): return None return self._extra.control_flow_ctx _flow_ctx.setter def control_flow_ctx(self, value: Optional[ControlFlowContext]): if (value == self.control_flow_ctx): return self._make_extra().control_flow_ctx = value def available_for_inference(self) -> bool: if (not self._extra): return True return self._extra.available_for_inference _for_inference.setter def available_for_inference(self, value: bool): if (value == self.available_for_inference): return self._make_extra().available_for_inference = value def _make_extra(self: Tensor) -> _TensorExtra: if (not self._extra): self._extra = _TensorExtra(tensor=self) return self._extra def sanity_check(self, ignore_placeholder=False, assume_complete=True): special_axes_dict = {'time_dim_axis': self.time_dim_axis, 'feature_dim_axis': self.feature_dim_axis} batch_dim_axis = self.batch_dim_axis batch_ndim = self.batch_ndim for (axis_name, axis) in special_axes_dict.items(): assert ((axis is None) or (0 <= axis < batch_ndim)), ('%s: axis %s (%i) invalid' % (self, axis_name, axis)) if (batch_dim_axis is not None): for (axis_name, axis) in special_axes_dict.items(): assert (axis != batch_dim_axis), ('%s: axis %s (%i) must be different from batch_dim_axis (%i)' % (self, axis_name, axis, batch_dim_axis)) if (self.sparse_dim is not None): assert (special_axes_dict['feature_dim_axis'] is None), ('%s: If sparse, there cannot be a feature dim axis.' % self) for (axis, tag) in enumerate(self._dims): if tag.is_batch_dim(): assert (axis == batch_dim_axis), ('%s: invalid %s' % (self, tag)) continue if (tag.batch and self.batch): assert ((tag.batch == self.batch) or self.batch.is_broadcast()) if tag.dyn_size_ext: assert (tag.dyn_size_ext.dtype in {'int32', 'int64'}) if tag.dyn_size_ext.have_batch_axis(): assert (tag.batch == tag.dyn_size_ext.batch) if ((not ignore_placeholder) and (self._raw_tensor is not None)): backend = self._raw_backend raw_shape = backend.get_known_shape_raw(self._raw_tensor) assert (len(raw_shape) == batch_ndim), f'Mismatching shape ndim: Raw tensor {raw_shape} vs Tensor {self}' for i in range(batch_ndim): if (self._dims[i].dimension is None): continue if (raw_shape[i] != self._dims[i].dimension): raise Exception((f'''Mismatching shape: Raw tensor {raw_shape} vs Tensor {self}; ''' + backend.format_graph_output(self._raw_tensor, max_depth=3))) backend.set_known_shape_raw(self._raw_tensor, self.batch_shape) assert (backend.get_dtype_name_raw(self._raw_tensor) == self.dtype), f'{self} dtype {self.dtype} does not match raw tensor dtype {backend.get_dtype_name_raw(self._raw_tensor)}' if assume_complete: for tag in self._dims: if tag.is_batch_dim(): continue if tag.is_dynamic(): assert tag.dyn_size_ext, ('%s sanity_check: dynamic dim %s undefined' % (self, tag)) if (not ignore_placeholder): if (tag.dyn_size_ext.placeholder is None): tag.complete_dyn_size() if (self.placeholder is not None): assert (tag.dyn_size_ext.placeholder is not None), ('%s sanity_check: dynamic dim %s value unknown' % (self, tag)) assert tag.is_dim_known() def get_runtime_sanity_check_op(self: Tensor): assert (self._raw_tensor is not None) return self._raw_backend.runtime_sanity_checks(self) def verify_out_shape(self, out_shape, allow_missing_implicit_dims=False): actual_dims_str = ('{%s}' % ', '.join([str(d) for d in (list(self.dim_tags) + sorted(self.dim_tags_set_implicit_only_wrapped))])) expected_dims_str = ('{%s}' % ', '.join([str(d) for d in sorted(out_shape)])) self_dim_tags = self.dim_tags_set_implicit self_dim_tags_implicit_only = self.dim_tags_set_implicit_only_wrapped if (not out_shape): if self_dim_tags: raise VerifyOutShapeException((('%s verify_out_shape:\n' % self) + ('Actual dims: %s\nExpected empty out_shape: %s' % (actual_dims_str, expected_dims_str)))) return if (not isinstance(out_shape, set)): raise TypeError(('%s verify_out_shape: expects a set but got %s' % (self, type(out_shape)))) remaining = set(self_dim_tags) for dim in out_shape: if isinstance(dim, Dim): dim_tag = dim elif isinstance(dim, _m.ImplicitDim): dim_tag = dim.tag if (dim not in self_dim_tags_implicit_only): raise VerifyOutShapeException(('%s verify_out_shape:\nActual dims: %s\nExpected out_shape: %s\n%s is not an implicit dim in self' % (self, actual_dims_str, expected_dims_str, dim))) elif isinstance(dim, _m.OptionalDim): dim_tag = dim.tag if (dim_tag not in remaining): continue else: raise TypeError(('%s verify_out_shape with out_shape %s: expect dim tags but got %s' % (self, out_shape, type(dim)))) if (dim_tag not in remaining): if (dim_tag in self_dim_tags): raise VerifyOutShapeException(((('%s verify_out_shape does not match:\n' % self) + ('Actual dims: %s\nExpected out_shape: %s\n' % (actual_dims_str, expected_dims_str))) + ('Dim %s multiple times in out_shape' % dim))) raise VerifyOutShapeException(((('%s verify_out_shape:\n' % self) + ('Actual dims: %s\nExpected out_shape: %s\n' % (actual_dims_str, expected_dims_str))) + ('Dim %s not in self' % dim))) remaining.discard(dim_tag) if remaining: if (allow_missing_implicit_dims and remaining.issubset(self.dim_tags_set_implicit_only)): pass else: raise VerifyOutShapeException(((('%s verify_out_shape missing dims:\n' % self) + ('Actual dims: %s\nExpected out_shape: %s\n' % (actual_dims_str, expected_dims_str))) + ('Missing dims: %s' % ', '.join(map(str, sorted(remaining)))))) def get_placeholder_kwargs(self, with_batch=True): return dict(name=self.name, dtype=self.dtype, shape=(self.batch_shape if with_batch else self.shape)) def get_axes_with_size(self): return [i for (i, dim) in enumerate(self.shape) if (dim is None)] def get_kwargs(self, *, include_special_axes=True): keys = ['name', 'dims', 'dtype'] if include_special_axes: if ((self.version <= 1) and (self.time_dim_axis_or_unspecified is not NotSpecified)): keys += ['time_dim_axis'] if (self.feature_dim_axis_or_unspecified is not NotSpecified): keys += ['feature_dim_axis'] if self.sparse_dim: keys += ['sparse_dim'] if (self.version == 1): keys += ['version'] if self._extra: if (self.batch is not None): keys += ['batch'] if (self.beam is not None): keys += ['beam'] if self.control_flow_ctx: keys += ['control_flow_ctx'] if (not self.available_for_inference): keys += ['available_for_inference'] return {key: getattr(self, key) for key in keys} def get_description(self, with_name=True, with_placeholder=False, catch_exceptions=False): keys = [] if self.sparse: keys.append('dtype') keys.append('sparse_dim') elif (self.dtype != 'float32'): keys.append('dtype') if with_placeholder: keys.append('placeholder') if (not self.available_for_inference): keys.append('available_for_inference') if (self.beam is not None): if ((not self.batch) or (self.batch.beam != self.beam)): keys.append('beam') args = [] if with_name: name = getattr(self, 'name', None) args += [(repr(name) if name else '<undefined>')] try: batch_shape_meta = ('[%s]' % ','.join(self.get_batch_axes_short_description())) except Exception as exc: if catch_exceptions: batch_shape_meta = ('<!%s: %s>' % (type(exc).__name__, exc)) else: raise args += [batch_shape_meta] for key in keys: try: value_repr = repr(getattr(self, key)) except Exception as exc: if catch_exceptions: value_repr = ('<!%s: %s>' % (type(exc).__name__, exc)) else: raise args += [('%s=%s' % (key, value_repr))] if self.control_flow_ctx: try: value_repr = self.control_flow_ctx.repr_inner() except Exception as exc: if catch_exceptions: value_repr = ('<!%s: %s>' % (type(exc).__name__, exc)) else: raise args += [('ctx=' + value_repr)] return ('Tensor{%s}' % ', '.join(args)) def get_batch_axes_short_description(self, special_axes=True): res = [] for (axis, dim_tag) in enumerate(self.dim_tags): descriptions = [] if (axis == self.batch_dim_axis): if self.batch: descriptions.append(self.batch.short_repr()) else: descriptions.append('B?') if special_axes: if (axis == self.time_dim_axis): descriptions.append('T') if (axis == self.feature_dim_axis): descriptions.append('F') if (self.batch_shape[axis] is None): if (axis == self.batch_dim_axis): pass else: descriptions.append(dim_tag.short_repr()) elif ((axis != self.batch_dim_axis) or (not self.batch)): descriptions.append(dim_tag.short_repr()) res.append(('|'.join(descriptions) or '?')) return res def get_compare_key(self): return (self.dtype, self.shape, self.batch_dim_axis, self.feature_dim_axis, self.time_dim_axis, self.dim_tags, self.batch, self.beam) def __repr__(self): return self.get_description(catch_exceptions=True) def __hash__(self): return id(self) def _sis_hash(self): if ((self.raw_tensor is not None) and hasattr(self.raw_tensor, '_sis_hash')): return self.raw_tensor._sis_hash() from sisyphus.hash import sis_hash_helper return sis_hash_helper(self.get_kwargs()) def __getstate__(self): d = {k: getattr(self, k) for k in self.__slots__} d['_raw_tensor'] = None return d def __setstate__(self, state): for (k, v) in state.items(): setattr(self, k, v) def reset(self: Tensor): self._raw_tensor = None self.batch = None def _adapt_batch_consistent_dim_tags(self): if (not self._extra): return if (not self.batch): return dims = tuple((tag.get_for_batch_ctx(batch=self.batch, ctx=self.control_flow_ctx) for tag in self._dims)) assert all(dims) dims: Tuple[(Dim, ...)] self._dims = dims def copy(self, name: Optional[str]=None) -> _t.Tensor: data = _t.Tensor(**self.get_kwargs()) data._raw_tensor = self._raw_tensor if name: data.name = name return data def copy_as_batch_major(self) -> _t.Tensor: return self.copy_with_batch_dim_axis(0) def copy_as_time_major(self) -> _t.Tensor: assert (self.time_dim_axis is not None) return self.copy_with_time_dim_axis(0) def copy_with_batch_dim_axis(self, batch_dim_axis) -> _t.Tensor: assert (self.batch_dim_axis is not None) return self.copy_move_axis(self.batch_dim_axis, batch_dim_axis) def copy_with_time_dim_axis(self, time_dim_axis) -> _t.Tensor: assert (self.time_dim_axis is not None) return self.copy_move_axis(self.time_dim_axis, time_dim_axis) def copy_transpose(self: Tensor, perm: Sequence[Union[(int, Dim)]], *, allow_int: bool=True) -> _t.Tensor: assert (len(perm) == len(self._dims)), f'{self}: invalid perm {perm!r} length' if (not perm): return self.copy() if (allow_int and isinstance(perm[0], int)): assert all((isinstance(a, int) for a in perm)), f'{self}: invalid perm {perm!r} types' assert (set(perm) == set(range(len(perm)))), f'{self}: invalid perm {perm!r}' return self._copy_compatible_to_dims_with_perm([self._dims[i] for i in perm], perm) else: assert all((isinstance(a, Dim) for a in perm)), f'{self}: invalid perm {perm!r} types' return self.copy_compatible_to_dims(perm) def copy_move_axis(self, old_axis, new_axis) -> _t.Tensor: if (old_axis < 0): old_axis += self.batch_ndim assert (old_axis >= 0) assert (0 <= old_axis < self.batch_ndim) if (new_axis < 0): new_axis += self.batch_ndim assert (new_axis >= 0) assert (0 <= new_axis < self.batch_ndim) if (old_axis == new_axis): return self.copy() perm = list(range(self.batch_ndim)) old = perm.pop(old_axis) perm.insert(new_axis, old) return self.copy_transpose(perm) def copy_swap_axes(self, axis1, axis2) -> _t.Tensor: if (axis1 < 0): axis1 += self.batch_ndim assert (0 <= axis1 < self.batch_ndim) if (axis2 < 0): axis2 += self.batch_ndim assert (0 <= axis2 < self.batch_ndim) if (axis1 == axis2): return self.copy() perm = list(range(self.batch_ndim)) (perm[axis1], perm[axis2]) = (perm[axis2], perm[axis1]) return self.copy_transpose(perm) def copy_as_bt_or_tb_major(self) -> _t.Tensor: assert (self.have_batch_axis() and self.have_time_axis()) if (self.batch_dim_axis == 0): return self.copy_with_time_dim_axis(1) if (self.time_dim_axis == 0): return self.copy_with_batch_dim_axis(1) if (self.batch_dim_axis > self.time_dim_axis): return self.copy_as_time_major().copy_as_bt_or_tb_major() return self.copy_as_batch_major().copy_as_bt_or_tb_major() def copy_with_feature_dim_axis(self, feature_dim_axis) -> _t.Tensor: assert (self.feature_dim_axis is not None) return self.copy_move_axis(self.feature_dim_axis, feature_dim_axis) def copy_as_batch_feature_major(self) -> _t.Tensor: assert (self.batch_dim_axis is not None) assert (self.feature_dim_axis is not None) data = self.copy_as_batch_major() data = data.copy_with_feature_dim_axis(1) return data def copy_as_time_batch_major(self) -> _t.Tensor: assert (self.have_batch_axis() and self.have_time_axis()) data = self.copy_as_bt_or_tb_major() if (data.time_dim_axis == 1): data = data.copy_move_axis(0, 1) return data def copy_as_batch_spatial_major(self) -> _t.Tensor: data = self.copy_as_batch_major() if (data.feature_dim_axis is not None): data = data.copy_with_feature_last() if data.size_placeholder: for (i, (j, size)) in enumerate(sorted(data.size_placeholder.items())): data = data.copy_move_axis(data.get_batch_axis(j), (i + 1)) if (data.feature_dim_axis is not None): assert (data.feature_dim_axis == (data.batch_ndim - 1)) if (data.feature_dim_axis_or_unspecified is not NotSpecified): if (data._default_feature_dim_axis() == data.feature_dim_axis): data.feature_dim_axis = NotSpecified return data def copy_with_feature_last(self) -> _t.Tensor: assert (self.feature_dim_axis is not None) return self.copy_with_feature_dim_axis((- 1)) def copy_add_batch_dim(self, batch_dim_axis, batch=None, dim_tag=None) -> _t.Tensor: if self.have_batch_axis(): raise Exception(f'{self} copy_add_batch_dim: already has batch-dim at axis {self.batch_dim_axis}, cannot add tag {dim_tag!r}') assert (self.batch_dim_axis is None) if (batch_dim_axis < 0): assert (((batch_dim_axis + self.batch_ndim) + 1) >= 0) batch_dim_axis += (self.batch_ndim + 1) assert (0 <= batch_dim_axis <= self.batch_ndim) data_opts = self.get_kwargs(include_special_axes=False) placeholder = self.placeholder if (placeholder is not None): backend = self._raw_backend placeholder = backend.expand_dims_raw(placeholder, batch_dim_axis) if batch: batch_dim_ = batch.dim elif dim_tag: if dim_tag.dyn_size_ext: assert (dim_tag.dyn_size_ext.dims == ()) assert (dim_tag.dyn_size_ext.raw_tensor is not None) batch_dim_ = dim_tag.dyn_size_ext.raw_tensor elif dim_tag.dimension: batch_dim_ = dim_tag.dimension else: raise Exception(f'{self} copy_add_batch_dim: unknown batch dim for {dim_tag!r}') else: raise Exception(f'{self} copy_add_batch_dim: unknown batch dim ') if ((not isinstance(batch_dim_, int)) or (batch_dim_ != 1)): placeholder = backend.expand_raw(placeholder, batch_dim_axis, batch_dim_) dim_tags = list(self.dim_tags) if dim_tag: assert dim_tag.is_batch_dim() assert (dim_tag.batch == batch) if batch: assert ((dim_tag.dimension == batch.static_dim) or (dim_tag.dimension is None)) elif batch: dim_tag = batch.batch_dim_tag else: dim_tag = Dim(kind=Dim.Types.Batch, description='batch', dimension=(batch.static_dim if batch else None), batch=batch) dim_tags.insert(batch_dim_axis, dim_tag) data_opts['dims'] = dim_tags if batch: data_opts['batch'] = batch data_opts['beam'] = batch.beam other_special_axes = self.get_special_axes_dict(counted_with_batch_dim=True, only_available=True) for (k, a) in other_special_axes.items(): data_opts[k] = (a if (a < batch_dim_axis) else (a + 1)) return _t.Tensor(placeholder=placeholder, **data_opts) def copy_add_spatial_dim(self, spatial_dim_axis=None, dim=1, auto_time_dim_axis=True) -> _t.Tensor: if (dim is None): assert (not self.placeholder) dim_tag = Dim(description='added_spatial', dimension=dim, kind=Dim.Types.Spatial) if (spatial_dim_axis is None): spatial_dim_axis = self.get_default_new_axis_for_dim_tag(dim_tag) v = self.copy_add_dim_by_tag(dim_tag, unbroadcast=True, axis=spatial_dim_axis) if (auto_time_dim_axis and (self.time_dim_axis is None)): v.time_dim_axis = spatial_dim_axis return v def copy_add_feature_dim(self, axis=None) -> _t.Tensor: if self.sparse: return self.copy_add_spatial_dim(spatial_dim_axis=axis) dim_tag = Dim(description='feature1', dimension=1, kind=Dim.Types.Feature) if (axis is None): axis = self.get_default_new_axis_for_dim_tag(dim_tag) v = self.copy_add_dim_by_tag(dim_tag, axis=axis) if (v.feature_dim_axis_or_unspecified is not NotSpecified): v.feature_dim_axis = NotSpecified if (axis < 0): axis += v.batch_ndim assert (axis >= 0) assert (0 <= axis < v.batch_ndim) if (v.feature_dim_axis != axis): v.feature_dim_axis = axis return v def get_default_new_axis_for_dim_tag(self, dim_tag: Dim) -> int: if dim_tag.is_batch_dim(): return 0 if (dim_tag.is_feature_dim() and (not self.sparse)): if (self.feature_dim_axis is not None): return (self.feature_dim_axis + 1) else: return self.batch_ndim if (dim_tag.is_dynamic() and self.get_dynamic_axes()): return (self.get_dynamic_axes()[(- 1)] + 1) if (dim_tag.is_spatial_dim() and self.get_spatial_batch_axes()): return (self.get_spatial_batch_axes()[(- 1)] + 1) elif (dim_tag.is_spatial_dim() and (self.feature_dim_axis is not None)): return self.feature_dim_axis else: return self.batch_ndim def copy_add_dim_by_tag(self, dim_tag, unbroadcast=False, axis=None) -> _t.Tensor: assert dim_tag.can_be_used_as_dim() if (axis is None): axis = self.get_default_new_axis_for_dim_tag(dim_tag=dim_tag) if (axis < 0): axis += (self.batch_ndim + 1) assert (0 <= axis <= self.batch_ndim) if dim_tag.is_batch_dim(): if unbroadcast: return self.copy_add_batch_dim(batch_dim_axis=axis, batch=dim_tag.batch, dim_tag=dim_tag) else: if (dim_tag.batch or self.batch): from returnn.tf.util.data import BatchInfo batch_info = BatchInfo.make_global_broadcast_batch_info() else: batch_info = None if (dim_tag and (dim_tag.dimension == 1) and (dim_tag.batch == batch_info)): pass else: dim_tag = Dim(kind=Dim.Types.Batch, description='batch-broadcast', dimension=1, batch=batch_info, auto_generated=True) return self.copy_add_batch_dim(batch_dim_axis=axis, batch=batch_info, dim_tag=dim_tag) data_opts = self.get_kwargs() if (self.sparse and dim_tag.is_feature_dim()): dim_tag = dim_tag.copy(same_as_self=True, kind=Dim.Types.Spatial) if ((not unbroadcast) and (dim_tag.dimension != 1)): dim_tag = Dim(kind=dim_tag.kind, description=('%s_dummy_dim1' % (dim_tag.description or 'unnamed')), dimension=1, auto_generated=True) data_opts['dims'] = ((self._dims[:axis] + (dim_tag,)) + self._dims[axis:]) other_special_axes = self.get_special_axes_dict(counted_with_batch_dim=True, only_available=True) for (k, a) in other_special_axes.items(): data_opts[k] = (a if (a < axis) else (a + 1)) if (dim_tag.is_feature_dim() and (self.feature_dim_axis is None)): data_opts.pop('feature_dim_axis', None) if (dim_tag.is_spatial_dim() and (self.time_dim_axis is None)): data_opts.pop('time_dim_axis', None) if (self.placeholder is not None): backend = self._raw_backend placeholder = backend.expand_dims_raw(self.placeholder, axis) if ((dim_tag.dimension is None) or (dim_tag.dimension > 1)): placeholder = backend.expand_raw(placeholder, axis, dim_tag.get_dim_value()) data_opts['placeholder'] = placeholder return _t.Tensor(**data_opts) def copy_split_feature_dim(self, new_feature_dim) -> _t.Tensor: assert (not self.sparse) assert (self.feature_dim_axis is not None) assert (self.dim is not None) assert ((self.dim % new_feature_dim) == 0), 'must be a multiple of the input feature dim' feature_dim_rem = (self.dim // new_feature_dim) new_feature_dim_axis = (self.feature_dim_axis + 1) data_opts = self.get_kwargs(include_special_axes=False) dim_tag_split_rem = Dim(kind=Dim.Types.Spatial, description=('feature_split_rem_%i' % feature_dim_rem), auto_generated=True, dimension=feature_dim_rem) dim_tag_new = Dim(kind=self.dim_tags[self.feature_dim_axis].kind, description=('feature_split_new_%i' % new_feature_dim), auto_generated=True, dimension=new_feature_dim) dim_tags = ((self.dim_tags[:self.feature_dim_axis] + (dim_tag_split_rem, dim_tag_new)) + self.dim_tags[(self.feature_dim_axis + 1):]) data_opts['dims'] = dim_tags other_special_axes = self.get_special_axes_dict(counted_with_batch_dim=True, only_available=True) other_special_axes.pop('feature_dim_axis', None) for (k, a) in other_special_axes.items(): data_opts[k] = (a if (a < new_feature_dim_axis) else (a + 1)) if (self.placeholder is not None): backend = self._raw_backend backend.set_known_shape_raw(self.placeholder, self.batch_shape) old_shape = backend.get_shape_tuple_raw(self.placeholder) new_shape = ((old_shape[:self.feature_dim_axis] + (feature_dim_rem, new_feature_dim)) + old_shape[(self.feature_dim_axis + 1):]) data_opts['placeholder'] = backend.reshape_raw(self.placeholder, new_shape) return _t.Tensor(**data_opts) def copy_extend_batch(self, batch) -> _t.Tensor: assert self.have_batch_axis() assert self.batch, ('%s: batch unset' % self) data = self.copy() batch = batch.copy_set_beam(data.beam) if (data.batch.beam != data.beam): data.batch = data.batch.copy_set_beam(data.beam) if (data.batch == batch): return data data.batch = batch self._adapt_batch_consistent_dim_tags() if (self.placeholder is not None): assert self._raw_backend.is_tensorflow assert set(self.batch.virtual_dims).issubset(batch.virtual_dims) import tensorflow as tf from returnn.tf.util.basic import get_shape from returnn.util.basic import ensure_list_of_type from returnn.tf.util.data import BatchInfo with tf.name_scope('copy_extend_batch'): axis = self.batch_dim_axis x = self.placeholder shape = get_shape(x) old_dims = ensure_list_of_type(self.batch.virtual_dims, BatchInfo.FixedDim) new_dims = ensure_list_of_type(batch.virtual_dims, BatchInfo.FixedDim) batch_broadcast_shape = [] ndim_batch_split = ((self.batch_ndim - 1) + len(new_dims)) tiles = ([1] * ndim_batch_split) old_idx = 0 for (new_idx, new_dim) in enumerate(new_dims): old_dim = (old_dims[old_idx] if (old_idx < len(old_dims)) else None) if (old_dim == new_dim): batch_broadcast_shape.append(old_dim.size) old_idx += 1 else: batch_broadcast_shape.append(1) tiles[(axis + new_idx)] = new_dim.size assert (old_idx == len(old_dims)) shape_batch_split = ((shape[:axis] + batch_broadcast_shape) + shape[(axis + 1):]) x = tf.reshape(x, shape_batch_split) x = tf.tile(x, tiles) shape = ((shape[:axis] + [batch.dim]) + shape[(axis + 1):]) x = tf.reshape(x, shape) data.placeholder = x return data def copy_compatible_to(self: Tensor, data: Tensor, add_dims=True, unbroadcast=False, except_feature=False, except_axis=None, check_sparse=True, check_dtype=True) -> Tensor: assert ((not check_sparse) or (self.sparse == data.sparse)) assert ((not check_dtype) or (self.dtype == data.dtype)) v = self.copy() if (v.have_batch_axis() and data.have_batch_axis() and v.batch and data.batch and (v.batch != data.batch)): v = v.copy_extend_batch(data.batch) v.sparse_dim = data.sparse_dim if ((v.batch_dim_axis is not None) and (data.batch_dim_axis is None)): raise ValueError(('copy_compatible_to: self %r has batch-dim, but target data %r has not' % (self, data))) if (data.batch_ndim < v.batch_ndim): raise ValueError(('copy_compatible_to: self %r already has more dims than target data %r' % (self, data))) is_equal_opts = dict(allow_same_feature_dim=True, allow_same_spatial_dim=True, treat_feature_as_spatial=True, ignore_feature_dim=True) mapped_axes = data.find_matching_dim_map(v, list(range(v.batch_ndim)), is_equal_opts) assert (len(mapped_axes) == v.batch_ndim) except_axis_int = (data.get_axis_from_description(except_axis, allow_int=True) if (except_axis is not None) else None) for target_axis in range(data.batch_ndim): new_v_axis = min(target_axis, v.batch_ndim) if (target_axis not in mapped_axes.values()): if (not add_dims): raise ValueError(('%s.copy_compatible_to(%s) not allowed, axis %i (%s) not in source' % (self, data, target_axis, data.dim_tags[target_axis]))) unbroadcast_axis = (unbroadcast and (not (except_feature and (data.feature_dim_axis == target_axis))) and (not ((except_axis_int is not None) and (except_axis_int == target_axis)))) v = v.copy_add_dim_by_tag(data.get_dim_tag(target_axis), axis=new_v_axis, unbroadcast=unbroadcast_axis) mapped_axes = {(v_ax + (1 if (v_ax >= new_v_axis) else 0)): trg_ax for (v_ax, trg_ax) in mapped_axes.items()} mapped_axes[new_v_axis] = target_axis else: matching_v_axes = [v_ax for (v_ax, trg_ax) in mapped_axes.items() if (trg_ax == target_axis)] assert (len(matching_v_axes) == 1) matching_v_axis = matching_v_axes[0] if (target_axis != matching_v_axis): v = v.copy_swap_axes(matching_v_axis, new_v_axis) (mapped_axes[matching_v_axis], mapped_axes[new_v_axis]) = (mapped_axes[new_v_axis], mapped_axes[matching_v_axis]) assert (v.batch_ndim == data.batch_ndim) assert all(((mapped_axes[ax] == ax) for ax in range(v.batch_ndim))) if (self.version == 1): assert (v.batch_dim_axis == data.batch_dim_axis) if (v.time_dim_axis != data.time_dim_axis): v.time_dim_axis = NotSpecified if (v.time_dim_axis != data.time_dim_axis): v.time_dim_axis = data.time_dim_axis if (v.feature_dim_axis != data.feature_dim_axis): v.feature_dim_axis = NotSpecified if (v.feature_dim_axis != data.feature_dim_axis): v.feature_dim_axis = data.feature_dim_axis elif (v.feature_dim_axis != data.feature_dim_axis): v.feature_dim_axis = data.feature_dim_axis if self.sparse: v.feature_dim_axis = NotSpecified v.sparse_dim = self.sparse_dim v.sanity_check() return v def get_out_permutation_to_dims(self, dims: Sequence[Dim]) -> List[int]: out_permutation: List[int] = [] count = 0 taken = ([False] * len(self._dims)) for dim in dims: candidates: List[int] = [] for j in range(len(self._dims)): if taken[j]: continue if (dim is self._dims[j]): candidates = [j] break if (dim == self._dims[j]): candidates.append(j) if (not candidates): out_permutation.append((- 1)) elif (len(candidates) == 1): out_permutation.append(candidates[0]) taken[candidates[0]] = True count += 1 else: max_match_priority_idx = None max_match_priority = None count_same_match_priority = 0 for j in range(len(candidates)): dim = self._dims[candidates[j]] match_priority = dim.match_priority if ((j > 0) and (match_priority == max_match_priority)): count_same_match_priority += 1 if ((j == 0) or (match_priority > max_match_priority)): max_match_priority = match_priority count_same_match_priority = 1 max_match_priority_idx = j assert (count_same_match_priority >= 1) if (count_same_match_priority > 1): raise ValueError(f'{self}: dim {dim} is ambiguous, from tensor dims {self._dims} and raw_tensor shape {dims}') out_permutation.append(candidates[max_match_priority_idx]) taken[candidates[max_match_priority_idx]] = True count += 1 assert (count == len(self._dims)) assert (len(out_permutation) == len(dims)) return out_permutation def copy_compatible_to_dims(self: _t.Tensor, dims: Sequence[Dim]) -> _t.Tensor: out_permutation = self.get_out_permutation_to_dims(dims) if (out_permutation == list(range(len(self._dims)))): return self.copy() return self._copy_compatible_to_dims_with_perm(dims, out_permutation) def _copy_compatible_to_dims_with_perm(self, dims: Sequence[Dim], out_permutation: Sequence[int]): raw_tensor = self._raw_tensor if (raw_tensor is not None): backend = self._raw_backend raw_shape = backend.get_shape_tuple_raw(raw_tensor) raw_tensor = backend.transpose_raw(raw_tensor, [p for p in out_permutation if (p >= 0)]) raw_tensor = backend.reshape_raw(raw_tensor, [(raw_shape[p] if (p >= 0) else 1) for p in out_permutation]) out_dims = [(dims[i] if (p >= 0) else Dim(kind=dims[i].kind, description=('%s_bc_dim1' % (dims[i].description or 'unnamed')), dimension=1, auto_generated=True)) for (i, p) in enumerate(out_permutation)] kwargs = self.get_kwargs(include_special_axes=False) kwargs['dims'] = out_dims kwargs['raw_tensor'] = raw_tensor res = _t.Tensor(**kwargs) if (self.version <= 1): if (self.time_dim_axis is None): if (res.time_dim_axis is not None): res.time_dim_axis = None else: axis = out_permutation.index(self.time_dim_axis) assert (axis >= 0) if (res.time_dim_axis != axis): res.time_dim_axis = axis if (self.feature_dim_axis is None): if (res.feature_dim_axis is not None): res.feature_dim_axis = None else: axis = out_permutation.index(self.feature_dim_axis) assert (axis >= 0) if (res.feature_dim_axis != axis): res.feature_dim_axis = axis return res def copy_compatible_to_dims_raw(self: _t.Tensor, dims: Sequence[Dim]) -> _t.RawTensorType: raw_tensor = self._raw_tensor assert (raw_tensor is not None), f'{self} copy_compatible_to_dims_raw: no raw tensor' out_permutation = self.get_out_permutation_to_dims(dims) if (out_permutation == list(range(len(self._dims)))): return raw_tensor backend = self._raw_backend raw_shape = backend.get_shape_tuple_raw(raw_tensor) raw_tensor = backend.transpose_raw(raw_tensor, [p for p in out_permutation if (p >= 0)]) raw_tensor = backend.reshape_raw(raw_tensor, [(raw_shape[p] if (p >= 0) else 1) for p in out_permutation]) return raw_tensor def copy_time_flattened(self) -> _t.Tensor: assert (self.batch_dim_axis is not None) assert (self.time_dim_axis is not None) data_opts = self.get_kwargs(include_special_axes=False) if (self.placeholder is not None): data_opts['placeholder'] = self.get_placeholder_time_flattened() dim_tag = self.dim_tags[self.time_dim_axis] dim_tag = Dim(kind=Dim.Types.Spatial, description=('%s_flattened' % (dim_tag.description or 'unnamed')), auto_generated=True, dimension=None) data_opts['dims'] = ((dim_tag,) + tuple((tag for (i, tag) in enumerate(self.dim_tags) if (i not in (self.batch_dim_axis, self.time_dim_axis))))) data_opts['time_dim_axis'] = None data_opts.pop('feature_dim_axis', None) return _t.Tensor(**data_opts) def copy_extend_with_beam(self, beam) -> Tensor: data = self.copy() if (data.beam and (data.beam == beam)): return data assert (data.beam is None), ('incompatible beam (%r vs %r)' % (data.beam, beam)) if (beam is None): return data data.beam = beam assert data.batch data.batch = data.batch.copy_set_beam(beam) if (data.placeholder is not None): assert data._raw_backend.is_tensorflow import tensorflow as tf from returnn.tf.util.basic import get_valid_scope_name_from_str, same_control_flow_ctx, tile_transposed with tf.name_scope(('%s_data_extend_with_beam' % get_valid_scope_name_from_str(self.name))): with same_control_flow_ctx(data.placeholder): data.placeholder = tile_transposed(data.placeholder, axis=data.batch_dim_axis, multiples=beam.beam_size) setattr(data.placeholder, '_RETURNN_beam_expanded_base_data', self) data._adapt_batch_consistent_dim_tags() return data def copy_merge_into_batch(self, axes) -> Tensor: assert self.batch assert (self.batch_dim_axis in axes) assert (sorted(set(axes)) == sorted(axes)) min_axis = min(axes) axes = list(axes) data = self.copy() if (axes != list(range(min_axis, (min_axis + len(axes))))): rem_axes_start = list(range(min_axis)) rem_axes_end = [a for a in range(min_axis, self.batch_ndim) if (a not in axes)] data = data.copy_transpose(((rem_axes_start + axes) + rem_axes_end)) axes = list(range(min_axis, (min_axis + len(axes)))) assert (data.batch_dim_axis in axes) tensor = data.placeholder batch = data.batch data = data.copy_template() batch_idx = 0 for axis in axes: if (axis == data.batch_dim_axis): batch_idx = len(batch.virtual_dims) continue batch = batch.copy_extend_with_padded_or_fixed_dim_tag(dim_tag=data.dim_tags[axis], new_dim_idx=batch_idx) batch_idx += 1 for axis in reversed(sorted(axes)): if (axis != data.batch_dim_axis): data = data.copy_template_excluding_axis(axis) data.batch = batch if (tensor is not None): assert self._raw_backend.is_tensorflow import tensorflow as tf from returnn.tf.util.basic import get_shape shape = get_shape(tensor) tensor = tf.reshape(tensor, ((shape[:min_axis] + [(- 1)]) + shape[(min_axis + len(axes)):])) data.placeholder = tensor return data def copy_squeeze_axes(self, axes) -> Tensor: assert isinstance(axes, (list, tuple)) assert all(((self.batch_shape[axis] == 1) for axis in axes)) assert all(((0 <= axis < self.batch_ndim) for axis in axes)) if (not axes): return self.copy() data_opts = self.get_kwargs(include_special_axes=False) if (self._raw_tensor is not None): backend = self._raw_backend data_opts['raw_tensor'] = backend.squeeze_raw(self._raw_tensor, axes) data_opts['dims'] = [tag for (i, tag) in enumerate(self._dims) if (i not in axes)] if ((self.time_dim_axis is not None) and (self.time_dim_axis_or_unspecified is not NotSpecified)): if (self.time_dim_axis not in axes): data_opts['time_dim_axis'] = (self.time_dim_axis - len([axis for axis in axes if (axis < self.time_dim_axis)])) if ((self.feature_dim_axis is not None) and (self.feature_dim_axis_or_unspecified is not NotSpecified)): if (self.feature_dim_axis not in axes): data_opts['feature_dim_axis'] = (self.feature_dim_axis - len([axis for axis in axes if (axis < self.feature_dim_axis)])) return _t.Tensor(**data_opts) def copy_template(self, name=None, *, dtype=None) -> _t.Tensor: kwargs = self.get_kwargs() if name: kwargs['name'] = name if dtype: kwargs['dtype'] = dtype return _t.Tensor(**kwargs) def copy_template_dense(self, name=None, dtype=None) -> Tensor: out = self.copy_template(name=name) if out.sparse: feat_dim = out.sparse_dim out.sparse = False out.dtype = 'float32' out = out.copy_add_dim_by_tag(dim_tag=feat_dim, unbroadcast=True, axis=(- 1)) out.feature_dim_axis = NotSpecified assert (out.feature_dim_axis == (out.batch_ndim - 1)) if dtype: out.dtype = dtype return out def copy_template_excluding_axis(self, exclude_axis, name=None) -> _t.Tensor: kwargs = self.get_kwargs(include_special_axes=False) if (exclude_axis < 0): exclude_axis += self.batch_ndim assert (exclude_axis >= 0) assert (0 <= exclude_axis < self.batch_ndim) if (exclude_axis == self.feature_dim_axis): kwargs.pop('dim', None) other_special_axes = self.get_special_axes_dict(counted_with_batch_dim=True, only_available=True) for (axis_name, axis) in other_special_axes.items(): if (axis == exclude_axis): continue kwargs[axis_name] = (axis if (axis < exclude_axis) else (axis - 1)) new_dim_tags = list(self.dim_tags) del new_dim_tags[exclude_axis] kwargs['dims'] = new_dim_tags if name: kwargs['name'] = name return _t.Tensor(**kwargs) def copy_template_excluding_spatial_dim(self, spatial_axis_num, name=None) -> Tensor: spatial_axes = self.get_spatial_batch_axes() if (spatial_axis_num < 0): spatial_axis_num += len(spatial_axes) assert (spatial_axis_num >= 0) assert (0 <= spatial_axis_num < len(spatial_axes)) axis_to_exclude = spatial_axes[spatial_axis_num] return self.copy_template_excluding_axis(exclude_axis=axis_to_exclude, name=name) def copy_template_excluding_time_dim(self, name=None) -> _t.Tensor: assert (self.time_dim_axis is not None) return self.copy_template_excluding_axis(exclude_axis=self.time_dim_axis, name=name) def copy_template_adding_time_dim(self, name=None, time_dim_axis=0) -> _t.Tensor: if (time_dim_axis < 0): time_dim_axis += (self.batch_ndim + 1) assert (time_dim_axis >= 0) assert (0 <= time_dim_axis <= self.batch_ndim) kwargs = self.get_kwargs(include_special_axes=False) dim_tag = Dim(kind=Dim.Types.Time, description='unknown_time', dimension=None, auto_generated=True) dim_tags = ((self.dim_tags[:time_dim_axis] + (dim_tag,)) + self.dim_tags[time_dim_axis:]) kwargs['dims'] = dim_tags other_special_axes = self.get_special_axes_dict(counted_with_batch_dim=True, only_available=True) other_special_axes.pop('time_dim_axis', None) for (axis_name, axis) in other_special_axes.items(): kwargs[axis_name] = (axis if (axis < time_dim_axis) else (axis + 1)) kwargs['time_dim_axis'] = time_dim_axis if name: kwargs['name'] = name return _t.Tensor(**kwargs) def copy_template_replace_dim_tag(self, axis, new_dim_tag, name=None) -> _t.Tensor: assert new_dim_tag.can_be_used_as_dim() if (axis < 0): assert ((axis + self.batch_ndim) >= 0) axis += self.batch_ndim assert (0 <= axis < self.batch_ndim) opts = self.get_kwargs() if self.dim_tags[axis].is_batch_dim(): opts.pop('batch', None) if new_dim_tag.is_batch_dim(): if (self.time_dim_axis == axis): opts.pop('time_dim_axis', None) if (self.feature_dim_axis == axis): opts.pop('feature_dim_axis', None) dim_tags = ((self.dim_tags[:axis] + (new_dim_tag,)) + self.dim_tags[(axis + 1):]) opts['dims'] = dim_tags if (self.feature_dim_axis_or_unspecified is not NotSpecified): if ((self.feature_dim_axis == axis) and self.dim_tags[axis].is_feature_dim() and (not new_dim_tag.is_feature_dim())): opts['feature_dim_axis'] = None if name: opts['name'] = name return _t.Tensor(**opts) def copy_template_replace_dim(self, axis, new_dim, new_size=None) -> _t.Tensor: dim_tag = self.dim_tags[axis] if dim_tag.is_batch_dim(): assert (new_dim is None) return self.copy_template() dim_tag = Dim(kind=dim_tag.kind, description=('%s_replaced' % (dim_tag.description or 'unnamed')), auto_generated=True, dimension=new_dim, dyn_size=new_size) return self.copy_template_replace_dim_tag(axis=axis, new_dim_tag=dim_tag) def copy_template_new_dim_tags(self, new_dim_tags, name=None, keep_special_axes=False) -> _t.Tensor: if keep_special_axes: assert (len(new_dim_tags) == self.batch_ndim) opts = self.get_kwargs(include_special_axes=keep_special_axes) opts['dims'] = new_dim_tags if name: opts['name'] = name return _t.Tensor(**opts) def copy_template_set_ctx(self, ctx) -> Tensor: kwargs = self.get_kwargs() kwargs['control_flow_ctx'] = ctx return _t.Tensor(**kwargs) def copy_template_unpack_batch(self) -> Tensor: assert self.have_batch_axis() assert self.batch, ('%s: batch unset' % self) data = self.copy() kwargs = self.get_kwargs(include_special_axes=False) from returnn.tf.util.data import BatchInfo dim_tags = [] for dim_tag in data.dim_tags: if (dim_tag.is_batch_dim() and dim_tag.batch and (len(dim_tag.batch.virtual_dims) > 0)): batch = dim_tag.batch new_batch_dim_tag = None for virtual_dim in batch.virtual_dims: if isinstance(virtual_dim, BatchInfo.PackedDim): dim_tags.append(virtual_dim.dim_tag) batch = batch.copy_remove_dim(virtual_dim) elif isinstance(virtual_dim, BatchInfo.GlobalBatchDim): assert (not new_batch_dim_tag) if ((batch is None) or batch.is_global_batch()): new_batch_dim_tag = batch_dim else: new_batch_dim_tag = Dim(kind=Dim.Types.Batch, description=dim_tag.description, dimension=None) dim_tags.append(new_batch_dim_tag) assert new_batch_dim_tag, ('%s: batch info %r invalid' % (self, batch)) new_batch_dim_tag.batch = batch kwargs['batch'] = batch else: dim_tags.append(dim_tag) kwargs['dims'] = dim_tags return _t.Tensor(**kwargs) def _get_variable_dim_pattern(self): return tuple([(dim is None) for dim in self.shape]) def _get_var_len_axes(self): return [i for (i, d) in enumerate(self._get_variable_dim_pattern()) if d] def matches_var_dim_pattern(self, other: _t.Tensor) -> bool: if (self.time_dim_axis_excluding_batch != other.time_dim_axis_excluding_batch): return False return (self._get_var_len_axes() == other._get_var_len_axes()) def dim_tags(self) -> Tuple[(Dim, ...)]: return self._dims def shape(self) -> Tuple[(Optional[int], ...)]: return tuple((tag.dimension for tag in self._dims if (not tag.is_batch_dim()))) def shape(self, shape): if (tuple(shape) == self.shape): return raise Exception(('%s: setting the shape is not allowed (new shape %s)' % (self, shape))) def batch_shape(self) -> Tuple[(Optional[int], ...)]: return tuple((tag.dimension for tag in self.dim_tags)) def get_batch_shape(self, batch_dim): if (self.batch_dim_axis is not None): return ((self.shape[:self.batch_dim_axis] + (batch_dim,)) + self.shape[self.batch_dim_axis:]) return self.shape def get_dynamic_batch_shape(self): return [self.get_dim(axis) for axis in range(self.batch_ndim)] def have_varying_shape_in_ctx(self): return any((tag.control_flow_ctx for tag in self.dim_tags)) def size_placeholder(self: Tensor): return _SizePlaceholderProxy(self) _placeholder.setter def size_placeholder(self, sizes): if (sizes is None): return for (axis_wo_b, size) in sizes.items(): self.set_dynamic_size(axis=self.get_batch_axis(axis_wo_b), sizes=size) def shape_dense(self): if self.sparse: return (self.shape + (self.dim,)) return self.shape def batch_shape_dense(self): if self.sparse: return (self.batch_shape + (self.dim,)) return self.batch_shape def dim_tags_sparse(self): if (self.sparse or (not self.have_feature_axis())): return self.dim_tags return (self.dim_tags[:self.feature_dim_axis] + self.dim_tags[(self.feature_dim_axis + 1):]) def dim_tags_set_implicit_only_wrapped(self): self_dim_tags = set(self.dim_tags) dims = set() if (self.sparse_dim and (self.sparse_dim not in self_dim_tags)): dims.add(_m.ImplicitSparseDim(self.sparse_dim)) for dim in self.dim_tags: if dim.dyn_size_ext: for dim_ in dim.dyn_size_ext.dim_tags: if (dim_ not in self_dim_tags): dims.add(_m.ImplicitDynSizeDim(dim_)) return dims def dim_tags_set_implicit_only(self): return set((dim.tag for dim in self.dim_tags_set_implicit_only_wrapped)) def dim_tags_set_implicit(self): dims = set(self.dim_tags) dims.update(self.dim_tags_set_implicit_only) return dims def remaining_dims(self: _t.Tensor, remove: Optional[Union[(Dim, Sequence[Dim])]]=None) -> List[Dim]: batch_dims = list(self._dims) if (not remove): pass elif isinstance(remove, Dim): batch_dims.remove(remove) else: for remove_ in remove: batch_dims.remove(remove_) return batch_dims def ndim(self): return len(self.shape) def ndim_dense(self): if self.sparse: return (self.ndim + 1) return self.ndim def batch_ndim(self): return len(self._dims) def batch_ndim_dense(self): if self.sparse: return (self.batch_ndim + 1) return self.batch_ndim def is_time_major(self): return (self.time_dim_axis == 0) def is_batch_major(self): return (self.batch_dim_axis == 0) def is_batch_feature_major(self): return ((self.batch_dim_axis == 0) and (self.feature_dim_axis == 1)) def batch_dim_axis(self): return _batch_dim_axis_from_dim_tags_tuple(self._dims) _dim_axis.setter def batch_dim_axis(self, axis): if (axis == self.batch_dim_axis): return raise Exception(('%s: cannot set batch_dim_axis = %s' % (self, axis))) def _default_feature_dim_axis(self): if (self.version >= 2): return None return _default_feature_dim_axis(batch_dim_axis=self.batch_dim_axis, time_dim_axis=self.time_dim_axis, batch_shape=self.batch_shape, sparse=self.sparse) def feature_dim_axis(self): if (self._feature_dim_axis is not NotSpecified): return self._feature_dim_axis return self._default_feature_dim_axis() _dim_axis.setter def feature_dim_axis(self: _t.Tensor, value): assert ((value is NotSpecified) or (value is None) or isinstance(value, int)) if (self.feature_dim_axis_or_unspecified == value): return if ((self.version >= 2) and (value is NotSpecified)): value = None if isinstance(value, int): assert (0 <= value < self.batch_ndim) self._feature_dim_axis = value def feature_dim_axis_or_unspecified(self): return self._feature_dim_axis def time_dim_axis(self) -> Optional[int]: if (self.version >= 2): return None if (self.time_dim_axis_or_unspecified is not NotSpecified): return self.time_dim_axis_or_unspecified return _default_time_dim_axis_dim_tags(self.dim_tags) _dim_axis.setter def time_dim_axis(self: _t.Tensor, value): assert ((value is NotSpecified) or (value is None) or isinstance(value, int)) if (self.time_dim_axis_or_unspecified == value): return if ((self.version >= 2) and (value in (None, NotSpecified))): return assert (self.version == 1), 'time_dim_axis is deprecated' if isinstance(value, int): assert (0 <= value < self.batch_ndim) self._make_extra().time_dim_axis = value def time_dim_axis_or_unspecified(self): if (self.version >= 2): return NotSpecified if (not self._extra): return NotSpecified return self._extra.time_dim_axis def time_dim_axis_excluding_batch(self): if (self.time_dim_axis is None): return None return self.get_batch_axis_excluding_batch(self.time_dim_axis) def placeholder(self): return self._raw_tensor def placeholder(self: _t.Tensor, value: Optional[_t.RawTensorType]): self.raw_tensor = value def batch(self): if (not self._extra): return None return self._extra.batch def batch(self, batch): if batch: assert (batch.beam == self.beam) if (self.batch == batch): return self._make_extra().batch = batch self._adapt_batch_consistent_dim_tags() def beam(self): if (not self._extra): return None if self._extra.beam: return self._extra.beam if self._extra.batch: return self._extra.batch.beam return None def beam(self, beam): if (self.beam == beam): return self._make_extra().beam = beam if self._extra.batch: self._extra.batch = self.batch.copy_set_beam(beam=beam) self._adapt_batch_consistent_dim_tags() def dim(self): tag = self.feature_dim_or_sparse_dim if tag: return tag.dimension return None def dim(self, dim): assert (dim == self.dim) def feature_dim_or_sparse_dim(self: Tensor): if self.sparse_dim: return self.sparse_dim feature_dim_axis = self.feature_dim_axis if (feature_dim_axis is not None): return self._dims[feature_dim_axis] return None def sparse(self): return (self.sparse_dim is not None) def sparse(self, sparse): if (self.sparse == sparse): return if (not sparse): self.sparse_dim = None return raise Exception(('%s: setting sparse=True not supported anymore. set sparse_dim instead' % self)) def vocab(self): if self.sparse_dim: return self.sparse_dim.vocab if self.have_feature_axis(): return self.dim_tags[self.feature_dim_axis].vocab return None def vocab(self, vocab): raise Exception(('%s: setting vocab not supported anymore. set sparse_dim instead' % self)) def time_dimension(self) -> Union[(int, _t.RawTensorType)]: assert (self.time_dim_axis is not None) return self.get_dim(self.time_dim_axis) def get_dim(self, axis: int) -> Union[(int, _t.RawTensorType)]: if (self.batch_shape[axis] is not None): return self.batch_shape[axis] assert (self._raw_tensor is not None) backend = self._raw_backend return backend.get_shape_tuple_raw(self._raw_tensor)[axis] def get_placeholder_as_time_major(self): assert (self.placeholder is not None) return self.copy_as_time_major().placeholder def get_placeholder_as_batch_major(self): assert (self.placeholder is not None) return self.copy_as_batch_major().placeholder def get_placeholder_with_runtime_sanity_checks(self): assert (self._raw_tensor is not None) backend = self._raw_backend return backend.identity_with_control_dependencies_raw(self._raw_tensor, [self.get_runtime_sanity_check_op()]) def get_placeholder_time_flattened(self): assert self._raw_backend.is_tensorflow, 'get_placeholder_time_flattened only implemented for TF yet' from returnn.tf.util.basic import flatten_with_seq_len_mask, get_shape import tensorflow as tf assert (self.placeholder is not None) assert self.have_time_axis() assert (0 in [self.time_dim_axis, self.batch_dim_axis]) time_dim = self.get_time_dim_tag() if time_dim.need_masking(): assert (time_dim.dyn_size_ext.dims == (self.get_batch_dim_tag(),)) return flatten_with_seq_len_mask(self.placeholder, time_dim.dyn_size, batch_dim_axis=self.batch_dim_axis, time_dim_axis=self.time_dim_axis) else: x = tf.transpose(self.placeholder, ([self.batch_dim_axis, self.time_dim_axis] + [i for i in range(self.batch_ndim) if (i not in [self.batch_dim_axis, self.time_dim_axis])])) shape = get_shape(x) return tf.reshape(x, ([(shape[0] * shape[1])] + shape[2:])) def get_placeholder_flattened(self, keepdims=False): assert (self.placeholder is not None) assert self._raw_backend.is_tensorflow, 'get_placeholder_flattened only implemented for TF yet' import tensorflow as tf x = self.placeholder orig_dyn_axes = (self.get_spatial_batch_axes() + [self.batch_dim_axis]) dyn_axes = list(orig_dyn_axes) if (dyn_axes == [self.batch_dim_axis]): return x assert (0 in dyn_axes), 'would need some transpose, not supported at the moment' assert (len(dyn_axes) > 1) orig_num_dyn_axes = len(dyn_axes) ndim = len(self.batch_shape) if self.have_time_axis(): x = self.get_placeholder_time_flattened() removed_axis = max(self.time_dim_axis, self.batch_dim_axis) dyn_axes.remove(removed_axis) dyn_axes = [(i if (i < removed_axis) else (i - 1)) for i in dyn_axes] ndim -= 1 if (len(dyn_axes) > 1): shape = tf.shape(x) x = tf.reshape(x, ([tf.reduce_prod([shape[i] for i in dyn_axes])] + [shape[i] for i in range(ndim) if (i not in dyn_axes)])) dyn_axes = [0] assert (dyn_axes == [0]) if (keepdims and (orig_num_dyn_axes >= 2)): for i in orig_dyn_axes: if (i not in dyn_axes): x = tf.expand_dims(x, axis=i) x.set_shape(([None] * self.batch_ndim)) return x def get_axes(self, exclude_time=False, exclude_batch=False, exclude_feature=False): axes = list(range(len(self.batch_shape))) if (exclude_time and (self.time_dim_axis is not None)): axes.pop(axes.index(self.time_dim_axis)) if (exclude_batch and (self.batch_dim_axis is not None)): axes.pop(axes.index(self.batch_dim_axis)) if (exclude_feature and (self.feature_dim_axis is not None)): axes.pop(axes.index(self.feature_dim_axis)) return axes def _verify_axis_int_from_description(cls, allow_int=NotSpecified): msg = 'Do not specify axis as int but as str or Dim instead.' if (allow_int is NotSpecified): from returnn.util import BehaviorVersion BehaviorVersion.require(condition=False, message=msg, version=5) if allow_int: return raise Exception(msg) def _verify_axis_order_dependent(cls): from returnn.util import BehaviorVersion BehaviorVersion.require(condition=False, message='Do not specify axis or axes in a way that depends on the order of the axes.', version=7) def _make_valid_int_axis(self, axis): if (axis < 0): assert ((axis + self.batch_ndim) >= 0) axis += self.batch_ndim assert (axis < self.batch_ndim) return axis def get_axes_from_description(self, axes, allow_int=NotSpecified): if ((axes is None) or (isinstance(axes, str) and (axes == ''))): return [] if isinstance(axes, Dim): dims = [i for (i, tag) in enumerate(self.dim_tags) if (tag == axes)] if (len(dims) > 1): max_match_priority = max((self.dim_tags[i].match_priority for i in dims)) dims = [i for i in dims if (self.dim_tags[i].match_priority == max_match_priority)] assert (len(dims) <= 1), ('%s: matching dim %s must be unique, use `match_priority` to resolve the matching order of ambiguous dimensions' % (self, axes)) return dims if isinstance(axes, int): self._verify_axis_int_from_description(allow_int=allow_int) return [self._make_valid_int_axis(axes)] assert isinstance(axes, (str, int, list, tuple, Sequence)) if isinstance(axes, str): import re axes = axes.lower() if (axes in ['b', 'batch']): assert (self.batch_dim_axis is not None) return [self.batch_dim_axis] elif (axes == 'spatial'): return self.get_spatial_batch_axes() elif re.match('(s|spatial):-?\\d+$', axes): self._verify_axis_order_dependent() s = int(axes.split(':')[1]) spatial_axes = self.get_spatial_batch_axes() if (s < 0): s += len(spatial_axes) assert (s < len(spatial_axes)), ('%s get_axes_from_description: %r invalid' % (self, axes)) return [spatial_axes[s]] elif (axes in ['dyn', 'dynamic']): return self.get_dynamic_axes() elif re.match('(d|dyn|dynamic):-?\\d+$', axes): self._verify_axis_order_dependent() s = int(axes.split(':')[1]) dyn_axes = self.get_dynamic_axes() if (s < 0): s += len(dyn_axes) assert (0 <= s < len(dyn_axes)), ('%s get_axes_from_description: %r invalid' % (self, axes)) return [dyn_axes[s]] elif (axes == 'spatial_except_time'): axes = self.get_spatial_batch_axes() assert (self.time_dim_axis is not None) axes.remove(self.time_dim_axis) return axes elif (axes in ['t', 'time']): assert (self.time_dim_axis is not None) return [self.time_dim_axis] elif (axes == 't?'): return ([self.time_dim_axis] if (self.time_dim_axis is not None) else []) elif (axes == 'except_time'): axes = list(range(self.batch_ndim)) axes.remove(self.batch_dim_axis) if (self.time_dim_axis is not None): axes.remove(self.time_dim_axis) return axes elif (axes == 'except_batch'): axes = list(range(self.batch_ndim)) axes.remove(self.batch_dim_axis) return axes elif re.match('(except_batch):-?\\d+$', axes): self._verify_axis_order_dependent() s = int(axes.split(':')[1]) non_batch_axes = list(range(self.batch_ndim)) if (self.batch_dim_axis is not None): non_batch_axes.remove(self.batch_dim_axis) if (s < 0): s += len(non_batch_axes) assert (0 <= s < len(non_batch_axes)), ('%s get_axes_from_description: %r invalid' % (self, axes)) return [non_batch_axes[s]] elif (axes == '*'): return list(range(self.batch_ndim)) elif (axes == 'static'): return self.get_static_axes() elif re.match('(static):-?\\d+$', axes): self._verify_axis_order_dependent() s = int(axes.split(':')[1]) static_axes = self.get_static_axes() if (s < 0): s += len(static_axes) assert (0 <= s < len(static_axes)), ('%s get_axes_from_description: %r invalid' % (self, axes)) return [static_axes[s]] elif re.match('(dim):\\d+$', axes): s = int(axes.split(':')[1]) dims = [a for a in range(self.batch_ndim) if (self.batch_shape[a] == s)] assert dims, ("%s get_axes_from_description: 'dim:%i' not found" % (self, s)) assert (len(dims) == 1), ("%s get_axes_from_description: 'dim:%i' only allowed when unique" % (self, s)) return dims elif (axes in ['f', 'feature', 'non_spatial']): return self.get_feature_batch_axes() elif all([(a in 'btf') for a in axes]): return self.get_axes_from_description(list(axes)) elif axes.startswith('stag:'): return [self.get_axis_by_tag_name(axes[len('stag:'):], spatial_only=True)] elif axes.startswith('stag-single:'): (_, idx_s, name) = axes.split(':', 2) idx = int(idx_s) return [self.get_axes_by_tag_name(name, spatial_only=True)[idx]] elif axes.startswith('tag:'): return [self.get_axis_by_tag_name(axes[len('tag:'):])] raise Exception(('invalid axis mode %r' % axes)) assert isinstance(axes, (tuple, list, Sequence)), ('invalid axes %r' % axes) flat_axes = [] for i in axes: if isinstance(i, int): self._verify_axis_int_from_description(allow_int=allow_int) flat_axes.append(self._make_valid_int_axis(i)) else: assert isinstance(i, (str, tuple, list, Dim)) flat_axes += self.get_axes_from_description(i, allow_int=allow_int) res = [] for i in flat_axes: if (i not in res): res.append(i) return res def get_dim_tag_from_description(self, axis): axis_int = self.get_axis_from_description(axis, allow_int=False) return self.dim_tags[axis_int] def get_axis_from_description(self, axis, allow_int=NotSpecified): if isinstance(axis, Dim): res_idx: Optional[int] = None res_tag: Optional[Dim] = None for (i, tag) in enumerate(self._dims): tag: Dim if ((tag is axis) or (tag == axis)): if ((res_tag is None) or (res_tag.match_priority < tag.match_priority)): res_idx = i res_tag = tag continue if (res_tag.match_priority > tag.match_priority): continue raise Exception(f'{self}: get_axis_from_description({axis}) not unique. use match_priority to resolve ambiguity') if (res_idx is None): raise Exception(f'{self}: get_axis_from_description({axis}) not found') return res_idx axes = self.get_axes_from_description(axis, allow_int=allow_int) assert axes, ('%s: %r axis not found' % (self, axis)) assert (len(axes) == 1), ('%r: %r is not a unique axis but %r' % (self, axis, axes)) return axes[0] def get_description_from_axis(self, axis): assert (0 <= axis < self.batch_ndim) if (axis == self.batch_dim_axis): return 'B' dim_tag = self.dim_tags[axis] matching_tags = [i for (i, tag) in enumerate(self.dim_tags) if (tag == dim_tag)] if (dim_tag.dyn_size_ext and (len(matching_tags) == 1)): return dim_tag if (axis == self.time_dim_axis): return 'T' if (axis == self.feature_dim_axis): return 'F' if (len(matching_tags) == 1): return dim_tag name = dim_tag.description matching_axes = self.get_axes_by_tag_name(name, spatial_only=True) assert (axis in matching_axes) return ('stag-single:%i:%s' % ((matching_axes.index(axis) - len(matching_axes)), name)) def has_axis(self, axis): axes = self.get_axes_from_description(axis, allow_int=False) return (len(axes) > 0) def get_axes_by_tag_name(self, name, spatial_only=False): dim_tags = self.get_batch_shape_dim_tags() matching_dim_tags = [(axis, tag) for (axis, tag) in enumerate(dim_tags) if ((name.lower() in tag.description.lower()) or (name.lower() in tag.get_same_base().description.lower()))] if spatial_only: spatial_axes = self.get_spatial_batch_axes() matching_dim_tags = [(axis, tag) for (axis, tag) in matching_dim_tags if ((axis in spatial_axes) or tag.is_spatial_dim())] return [ax for (ax, _) in matching_dim_tags] def get_axis_by_tag_name(self, name, spatial_only=False): matching_dim_tags = self.get_axes_by_tag_name(name, spatial_only) assert (len(matching_dim_tags) > 0), ('%r: no %stag found with name %r' % (self, ('spatial ' if spatial_only else ''), name)) assert (len(matching_dim_tags) == 1), ('%r: tag name %r is not unique in dim tags %r' % (self, name, self.get_batch_shape_dim_tags())) return matching_dim_tags[0] def get_batch_axis_excluding_batch(self, axis): return _get_axis_wo_b(axis, batch_dim_axis=self.batch_dim_axis, batch_ndim=self.batch_ndim) def have_dim_tag(self, tag, include_implicit=True, unique=False): dims = list(self.dim_tags) if include_implicit: dims.extend(self.dim_tags_set_implicit_only) matching_dims = [dim for dim in dims if (dim == tag)] return ((len(matching_dims) == 1) if unique else (len(matching_dims) >= 1)) def get_batch_axis(self, axis): return _get_axis_wb(axis, batch_dim_axis=self.batch_dim_axis) def have_batch_axis(self): return (self.batch_dim_axis is not None) def have_time_axis(self): return (self.time_dim_axis is not None) def have_feature_axis(self): return (self.feature_dim_axis is not None) def is_time_axis_dynamic(self): assert (self.time_dim_axis is not None) if (self.placeholder is None): return (self.batch_shape[self.time_dim_axis_excluding_batch] is None) if (self.time_dim_axis_excluding_batch in self.size_placeholder): return True assert isinstance(self.shape[self.time_dim_axis_excluding_batch], int), ('%s: dynamic time axis dim (None) (axis %i) but size_placeholder %r misses information' % (self, self.time_dim_axis, self.size_placeholder)) return False def is_axis_dynamic(self, axis): if (axis == self.batch_dim_axis): return False return (self.batch_shape[axis] is None) def has_dynamic_size(self, axis): return (self.dim_tags[axis].dyn_size is not None) def get_dynamic_size(self, axis): tag = self.dim_tags[axis] assert (tag.dyn_size is not None), ('%s: axis %i has no dyn size' % (self, axis)) return tag.dyn_size def set_dynamic_size(self, axis, sizes): if (getattr(sizes, '_RETURNN_dyn_size_beam', NotSpecified) is NotSpecified): sizes._RETURNN_dyn_size_beam = self.beam if (self.beam and (getattr(sizes, '_RETURNN_dyn_size_beam', None) != self.beam)): tag = Dim.get_tag_from_size_tensor(sizes) assert (tag and self.batch) tag = tag.get_for_batch_ctx(batch=self.batch, ctx=self.control_flow_ctx) assert (tag.dyn_size is not None) sizes = tag.dyn_size sizes_tag = Dim.get_tag_from_size_tensor(sizes) if sizes_tag: assert sizes_tag.is_same_size_tensor(sizes) tag = self._dims[axis] assert tag.is_dynamic() if tag.is_same_size_tensor(sizes): pass elif (tag.dyn_size is None): if sizes_tag: assert sizes_tag.is_same_size_tensor(sizes) tag = sizes_tag else: tag = tag.set_tag_on_size_tensor(sizes, batch=self.batch) else: assert sizes_tag, ('%s: assign dyn sizes %s without defined dim tag' % (self, sizes)) tag = sizes_tag if self.batch: tag = tag.get_for_batch_ctx(batch=self.batch, ctx=self.control_flow_ctx) if (tag is not self._dims[axis]): self._dims = ((self._dims[:axis] + (tag,)) + self._dims[(axis + 1):]) if (tag.dyn_size is None): tag.dyn_size = sizes def get_dynamic_axes(self): return [axis for (axis, dim) in enumerate(self.batch_shape) if ((axis != self.batch_dim_axis) and (dim is None))] def get_static_axes(self): return [axis for (axis, dim) in enumerate(self.batch_shape) if ((axis != self.batch_dim_axis) and (dim is not None))] def mark_same_time(self, tags, must_match=False): if isinstance(tags, Dim): tags = {tags} assert all((isinstance(tag, Dim) for tag in tags)) for (axis, dim_tag) in enumerate(self.dim_tags): if (dim_tag in tags): self.time_dim_axis = axis return True if must_match: raise Exception(('%s mark_same_time: %s not found' % (self, tags))) return False def is_same_time_dim(self, other: Tensor) -> bool: assert self.have_time_axis() if (not other.have_time_axis()): return False tag_self = self.get_dim_tag(self.time_dim_axis) tag_other = other.get_dim_tag(other.time_dim_axis) return (tag_self == tag_other) def get_sequence_lengths(self) -> _t.RawTensorType: assert (self.time_dim_axis is not None) dim = self._dims[self.time_dim_axis] assert isinstance(dim, Dim) if dim.dyn_size_ext: if (dim.dyn_size_ext.raw_tensor is None): dim.complete_dyn_size() assert (dim.dyn_size_ext.raw_tensor is not None) return dim.dyn_size_ext.raw_tensor assert (self.batch_shape[self.time_dim_axis] is not None) assert (self.batch_dim_axis is not None) batch_dim_ = self._dims[self.batch_dim_axis] assert isinstance(batch_dim_, Dim) if (batch_dim_.dyn_size_ext and (batch_dim_.dyn_size_ext.raw_tensor is not None)): backend = batch_dim_.dyn_size_ext._raw_backend return backend.fill_raw([batch_dim_.dyn_size_ext.raw_tensor], dim.size) import tensorflow as tf return tf.fill([self.get_batch_dim()], dim.size) def get_sequence_mask(self): from returnn.frontend._backend import get_backend_by_raw_tensor_type assert (self.time_dim_axis is not None) assert (self.batch_dim_axis is not None) dyn_seq_len = self.get_sequence_lengths() backend = get_backend_by_raw_tensor_type(type(dyn_seq_len)) if self.is_time_major: assert (self.batch_dim_axis == 1) return backend.sequence_mask_raw(dyn_seq_len, batch_major=False) else: assert (self.batch_dim_axis == 0) assert (self.time_dim_axis == 1) return backend.sequence_mask_raw(dyn_seq_len, batch_major=True) def get_sequence_mask_broadcast(self: Tensor, axis=None) -> _t.RawTensorType: if isinstance(axis, Dim): axis = self.get_axis_from_description(axis) if (axis is None): assert (self.time_dim_axis is not None) axis = self.time_dim_axis if (axis < 0): assert ((axis + self.batch_ndim) > 0) axis += self.batch_ndim assert (0 <= axis < self.batch_ndim) assert (axis != self.batch_dim_axis) tag: Dim = self.dim_tags[axis] assert (tag.dyn_size_ext and (tag.dyn_size_ext.raw_tensor is not None)) backend = tag.dyn_size_ext._raw_backend assert set(tag.dyn_size_ext.dim_tags).issubset(self.dim_tags) with backend.name_scope_raw('get_sequence_mask_broadcast'): if (backend.have_sequence_mask_raw() and tag.dyn_size_ext.have_batch_axis() and (tag.dyn_size_ext.batch_ndim == 1)): size = tag.dyn_size seq_mask = backend.sequence_mask_raw(size, batch_major=(axis >= self.batch_dim_axis)) shape = ([1] * self.batch_ndim) shape[self.batch_dim_axis] = self.get_batch_dim() shape[axis] = tag.get_dim_value() seq_mask = backend.reshape_raw(seq_mask, shape) assert (seq_mask.get_shape().ndims == self.batch_ndim) else: seq_mask = self.get_sequence_mask_tensor(axis).copy_compatible_to_dims_raw(self.dims) return seq_mask def get_sequence_mask_tensor(self: Tensor, axis: int) -> Tensor: if (axis < 0): assert ((axis + self.batch_ndim) > 0) axis += self.batch_ndim assert (0 <= axis < self.batch_ndim) assert (axis != self.batch_dim_axis) tag: Dim = self.dim_tags[axis] return tag.get_mask(dim_order=self.dims, device=self.device) def get_sequence_lengths_broadcast(self, axis=None): if (axis is None): assert (self.time_dim_axis is not None) axis = self.time_dim_axis if (axis < 0): assert ((axis + self.batch_ndim) > 0) axis += self.batch_ndim assert (0 <= axis < self.batch_ndim) assert (axis != self.batch_dim_axis) tag = self.dim_tags[axis] assert tag.dyn_size_ext return tag.dyn_size_ext.copy_compatible_to(self, check_dtype=False, check_sparse=False).placeholder def num_elements(self: Tensor) -> Union[(int, Tensor)]: import returnn.frontend as rf return rf.num_elements_of_shape(self.dims) def copy_masked(self: Tensor, mask_value: Union[(Tensor, float, int, _t.RawTensorType)], *, dims: Optional[Sequence[Union[(Dim, int)]]]=None, allow_int: bool=NotSpecified) -> Tensor: assert (self.raw_tensor is not None) if (dims is None): axes = range(self.batch_ndim) else: axes = [self.get_axis_from_description(dim, allow_int=allow_int) for dim in dims] assert (len(set(axes)) == len(dims)), f'{self} copy_masked, dims {dims} not unique, axes {axes}' axes_ = [] for axis in axes: tag: Dim = self.dims[axis] if (not tag.need_masking()): continue if set(tag.dyn_size_ext.dim_tags).issubset(self.dim_tags): axes_.append(axis) axes = axes_ if (not axes): return self.copy() use_padding_info = False tf_util = None if self._raw_backend.is_tensorflow: import returnn.tf.util.basic as tf_util use_padding_info = isinstance(mask_value, (int, float)) if use_padding_info: d = tf_util.get_padding_info_dict_ref(self.raw_tensor) existing_pad_values = [d.get(self.dim_tags[axis]) for axis in axes] if (set(existing_pad_values) == {mask_value}): return self.copy() import returnn.frontend as rf mask = None for axis in axes: mask_ = self._dims[axis].get_mask(dim_order=self.dims, device=self.device) mask = (rf.logical_and(mask, mask_) if (mask is not None) else mask_) assert isinstance(mask, _t.Tensor) res = rf.where(mask, self, mask_value) if use_padding_info: d = tf_util.get_padding_info_dict_ref(res.raw_tensor) d.clear() d.update({self.dim_tags[axis]: mask_value for axis in axes}) return res def get_batch_dim(self) -> Union[(_t.RawTensorType, int)]: assert (self.batch_dim_axis is not None) if self.batch: if self.beam: assert (self.batch.beam == self.beam) dim = self.batch.dim if (not isinstance(dim, int)): batch_dim_ = self.dim_tags[self.batch_dim_axis] batch_dim_.set_tag_on_size_tensor(dim, batch=self.batch) return dim from returnn.tf.layers.base import LayerBase batch = LayerBase.get_recent_layer().get_batch_info() batch = batch.copy_set_beam(self.beam) return batch.dim def get_batch_dim_tag(self): assert self.have_batch_axis() return self.dim_tags[self.batch_dim_axis] def get_static_batch_dim(self): if self.batch: return self.batch.static_dim if self.have_batch_axis(): return self.get_batch_dim_tag().dimension return None def get_spatial_batch_axes(self): return [axis for axis in range(self.batch_ndim) if ((axis != self.batch_dim_axis) and ((axis != self.feature_dim_axis) or (axis == self.time_dim_axis) or (self.batch_shape[axis] is None)))] def get_spatial_axes(self): return [self.get_batch_axis_excluding_batch(axis) for axis in self.get_spatial_batch_axes()] def get_feature_batch_axes(self): if (self.feature_dim_axis is not None): return [self.feature_dim_axis] return [] def get_feature_axes(self): return [self.get_batch_axis_excluding_batch(axis) for axis in self.get_feature_batch_axes()] SpecialAxesNames = ('time_dim_axis', 'feature_dim_axis') def get_special_axes_dict(self, counted_with_batch_dim=True, only_available=False): axes = list(self.SpecialAxesNames) d = {k: getattr(self, k) for k in axes} if (not counted_with_batch_dim): d = {k: (self.get_batch_axis_excluding_batch(v) if (v is not None) else None) for (k, v) in d.items()} if only_available: d = {k: v for (k, v) in d.items() if (v is not None)} if (self.feature_dim_axis_or_unspecified is NotSpecified): d.pop('feature_dim_axis', None) return d def get_bc_spatial_batch_shape(self): dyn_axes = self.get_spatial_batch_axes() if (self.batch_dim_axis is not None): dyn_axes += [self.batch_dim_axis] return tuple([(1 if (axis in dyn_axes) else dim) for (axis, dim) in enumerate(self.batch_shape)]) def get_bc_shape(self, opts=None): if (opts is None): opts = {} default_axes_map = dict(enumerate(self.get_bc_spatial_batch_shape())) axes_map = {} for (key, value) in opts.items(): assert (value in ((- 1), 1, 'x', None)), ('%r get_bc_shape: invalid value in opts %r' % (self, opts)) if (value == 'x'): value = 1 if (value == (- 1)): value = None key_axes = self.get_axes_from_description(key) for key_axis in key_axes: assert (key_axis not in axes_map), ('%r get_bc_shape: axis %i is defined multiple times in opts %r' % (self, key_axis, opts)) assert (0 <= key_axis < self.batch_ndim), ('%r get_bc_shape: invalid axis %i in opts %r' % (self, key_axis, opts)) (axes_map if (key != '*') else default_axes_map)[key_axis] = (self.batch_shape[key_axis] if (value is None) else value) remaining_axes = sorted(set(range(self.batch_ndim)).difference(axes_map.keys())) for axis in remaining_axes: axes_map[axis] = default_axes_map[axis] assert (sorted(axes_map.keys()) == list(range(self.batch_ndim))) return tuple([axes_map[i] for i in range(self.batch_ndim)]) def get_scope_name(self): if (self.placeholder is not None): return os.path.dirname(self.placeholder.name) if self.size_placeholder: for (i, v) in sorted(self.size_placeholder.items()): if (v is not None): return os.path.dirname(v.name) return None def get_full_name(self): scope_name = self.get_scope_name() if scope_name: return ('%s/%s' % (scope_name, self.name)) return self.name def get_dim_tag(self, axis): return self._dims[axis] def get_time_dim_tag(self): assert (self.time_dim_axis is not None) return self.get_dim_tag(self.time_dim_axis) def get_dyn_size_tags(self): return [dim_tag for dim_tag in self._dims if dim_tag.is_dynamic_seq_length()] def get_size_dim_tag(self, number): dyn_size_tags = self.get_dyn_size_tags() return dyn_size_tags[number] def get_batch_shape_dim_tags(self): return self.dim_tags def get_common_data(cls, sources: List[Tensor], ignore_feature_dim=False, allow_broadcast_all_sources=NotSpecified, name=None) -> Optional[Tensor]: from returnn.util import BehaviorVersion if (not sources): return None assert sources if (len(sources) == 1): return sources[0].copy_template() max_ndim = max([s.batch_ndim for s in sources]) if any((src.batch for src in sources)): from returnn.tf.util.data import BatchInfo common_batch = BatchInfo.get_common_batch_info([src.batch for src in sources if src.batch]) else: common_batch = None common = [s for s in sources if (s.batch_ndim == max_ndim)][0] common = common.copy_template(name=name) common.beam = None if common_batch: common.batch = common_batch.copy_set_beam(None) if any([s.beam for s in sources]): from returnn.tf.util.data import SearchBeam common.beam = SearchBeam.get_combined_beam(*[s.beam for s in sources]) is_equal_opts = dict(ignore_feature_dim=ignore_feature_dim, treat_feature_as_spatial=True, allow_same_spatial_dim=True, undefined_matches=True, derived_matches=True) if (BehaviorVersion.get() < 11): is_equal_opts['broadcast_matches'] = True (all_dim_tags, tags_dict) = Dim.get_all_dimension_tags(sources, is_equal_opts=is_equal_opts) for dim_tag in all_dim_tags: common_tag = Dim.get_existing_tag_from_collection(dim_tag, common.dim_tags, is_equal_opts=is_equal_opts) if common_tag: if (dim_tag != common_tag): axis = common.dim_tags.index(common_tag) common = common.copy_template_replace_dim_tag(axis=axis, new_dim_tag=dim_tag) else: axis = common.get_default_new_axis_for_dim_tag(dim_tag) common = common.copy_add_dim_by_tag(dim_tag, unbroadcast=True, axis=axis) if all(((s.batch_ndim < common.batch_ndim) for s in sources)): from returnn.util.basic import validate_broadcast_all_sources validate_broadcast_all_sources(allow_broadcast_all_sources=allow_broadcast_all_sources, inputs=sources, common=common) return common def find_matching_dims(self: Tensor, dim_tag: Dim, is_equal_opts) -> List[int]: return [axis for axis in range(self.batch_ndim) if self.get_dim_tag(axis).is_equal(dim_tag, **is_equal_opts)] def find_matching_dim_map(self: Tensor, other: Tensor, other_axes, is_equal_opts=None) -> Dict[(int, int)]: if (is_equal_opts is None): is_equal_opts = dict(allow_same_feature_dim=True, allow_same_spatial_dim=True, treat_feature_as_spatial=True) def map_other_axis_to_self(other_axis: int, taken_self_axes: Set[int]) -> int: other_axis_dim_tag = other.dims[other_axis] is_equal_opts_ = None matching = None for opt in [{}, is_equal_opts, 'broadcast_matches', 'unknown_spatial_matches']: if isinstance(opt, dict): is_equal_opts_ = opt.copy() elif isinstance(opt, str): if (opt in is_equal_opts_): continue is_equal_opts_[opt] = True matching = [self_axis for self_axis in self.find_matching_dims(other_axis_dim_tag, is_equal_opts_) if (self_axis not in taken_self_axes)] if (opt == 'unknown_spatial_matches'): assert (len(matching) <= 1), ('cannot match axes %s from %s to %s, failed at other %s, not unique after %s' % (other_axes, other, self, other_axis, opt)) if matching: break assert matching, ('cannot match the axes %s from %s to %s. Failing at axis %s, tag %s' % (other_axes, other, self, other_axis, other.dim_tags[other_axis])) if (len(matching) == 1): return matching[0] max_match_priority = max((dim.match_priority for dim in self.dims)) return max(matching, key=(lambda ax: ((max_match_priority + 1) if (self.dims[ax] is other_axis_dim_tag) else self.dims[ax].match_priority))) other_to_self_mapping = {} for axis in other_axes: other_to_self_mapping[axis] = map_other_axis_to_self(axis, set(other_to_self_mapping.values())) assert (len(other_to_self_mapping) == len(other_axes)), 'other_axes may not contain duplicates' return other_to_self_mapping def is_valid_in_current_graph(self: _t.Tensor) -> bool: if (self._raw_tensor is None): return True return self._raw_backend.is_valid_in_current_graph(self) def mark_as_loss(self: Tensor, name: str, *, scale: Optional[float]=1.0, as_error: bool=False, use_normalized_loss: bool=False, use_flatten_frames: bool=True, custom_inv_norm_factor: Optional[Tensor]=None) -> None: import returnn.frontend as rf rf.get_run_ctx().mark_as_loss(loss=self, name=name, scale=scale, as_error=as_error, use_normalized_loss=use_normalized_loss, use_flatten_frames=use_flatten_frames, custom_inv_norm_factor=custom_inv_norm_factor) def mark_as_output(self: Tensor, name: str, *, shape: Optional[Sequence[Dim]]=None) -> None: import returnn.frontend as rf rf.get_run_ctx().mark_as_output(self, name=name, dims=shape) def mark_as_default_output(self: Tensor, *, shape: Optional[Sequence[Dim]]=None) -> None: import returnn.frontend as rf rf.get_run_ctx().mark_as_default_output(self, shape=shape)
def procedure(xloader, teacher, network, criterion, scheduler, optimizer, mode, config, extra_info, print_freq, logger): (data_time, batch_time, losses, top1, top5) = (AverageMeter(), AverageMeter(), AverageMeter(), AverageMeter(), AverageMeter()) (Ttop1, Ttop5) = (AverageMeter(), AverageMeter()) if (mode == 'train'): network.train() elif (mode == 'valid'): network.eval() else: raise ValueError('The mode is not right : {:}'.format(mode)) teacher.eval() logger.log('[{:5s}] config :: auxiliary={:}, KD :: [alpha={:.2f}, temperature={:.2f}]'.format(mode, (config.auxiliary if hasattr(config, 'auxiliary') else (- 1)), config.KD_alpha, config.KD_temperature)) end = time.time() for (i, (inputs, targets)) in enumerate(xloader): if (mode == 'train'): scheduler.update(None, ((1.0 * i) / len(xloader))) data_time.update((time.time() - end)) targets = targets.cuda(non_blocking=True) if (mode == 'train'): optimizer.zero_grad() (student_f, logits) = network(inputs) if isinstance(logits, list): assert (len(logits) == 2), 'logits must has {:} items instead of {:}'.format(2, len(logits)) (logits, logits_aux) = logits else: (logits, logits_aux) = (logits, None) with torch.no_grad(): (teacher_f, teacher_logits) = teacher(inputs) loss = loss_KD_fn(criterion, logits, teacher_logits, student_f, teacher_f, targets, config.KD_alpha, config.KD_temperature) if ((config is not None) and hasattr(config, 'auxiliary') and (config.auxiliary > 0)): loss_aux = criterion(logits_aux, targets) loss += (config.auxiliary * loss_aux) if (mode == 'train'): loss.backward() optimizer.step() (sprec1, sprec5) = obtain_accuracy(logits.data, targets.data, topk=(1, 5)) losses.update(loss.item(), inputs.size(0)) top1.update(sprec1.item(), inputs.size(0)) top5.update(sprec5.item(), inputs.size(0)) (tprec1, tprec5) = obtain_accuracy(teacher_logits.data, targets.data, topk=(1, 5)) Ttop1.update(tprec1.item(), inputs.size(0)) Ttop5.update(tprec5.item(), inputs.size(0)) batch_time.update((time.time() - end)) end = time.time() if (((i % print_freq) == 0) or ((i + 1) == len(xloader))): Sstr = ((' {:5s} '.format(mode.upper()) + time_string()) + ' [{:}][{:03d}/{:03d}]'.format(extra_info, i, len(xloader))) if (scheduler is not None): Sstr += ' {:}'.format(scheduler.get_min_info()) Tstr = 'Time {batch_time.val:.2f} ({batch_time.avg:.2f}) Data {data_time.val:.2f} ({data_time.avg:.2f})'.format(batch_time=batch_time, data_time=data_time) Lstr = 'Loss {loss.val:.3f} ({loss.avg:.3f}) {top1.val:.2f} ({top1.avg:.2f}) {top5.val:.2f} ({top5.avg:.2f})'.format(loss=losses, top1=top1, top5=top5) Lstr += ' Teacher : ={:.2f}, ={:.2f}'.format(Ttop1.avg, Ttop5.avg) Istr = 'Size={:}'.format(list(inputs.size())) logger.log(((((((Sstr + ' ') + Tstr) + ' ') + Lstr) + ' ') + Istr)) logger.log(' **{:5s}** accuracy drop :: ={:.2f}, ={:.2f}'.format(mode.upper(), (Ttop1.avg - top1.avg), (Ttop5.avg - top5.avg))) logger.log(' **{mode:5s}** {top1.avg:.2f} {top5.avg:.2f} {error1:.2f} {error5:.2f} Loss:{loss:.3f}'.format(mode=mode.upper(), top1=top1, top5=top5, error1=(100 - top1.avg), error5=(100 - top5.avg), loss=losses.avg)) return (losses.avg, top1.avg, top5.avg)
def test_encoder(): img_feat = torch.randn(4, 36, 2048) seq_size = 20 ques = torch.randperm(seq_size).view(1, seq_size) ques = ques.unsqueeze(1).repeat(4, 10, 1) ques_len = torch.LongTensor([6, 5, 4, 3]).unsqueeze(1).repeat(1, 10)
def test_unmatched_lengths_2d_np_array(): y_true = np.array([[1, 2, 3], [1, 2, 4], [1, 5, 6], [1, 5, 8]]) y_pred = np.array([[1, 2, 3], [1, 2, 4]]) with pytest.raises(AssertionError): precision(y_true, y_pred)
class CarlaEngine(): def __init__(self, config, traffic_manager, carla_observers): self._running = False self.config = config self._traffic_manager = traffic_manager self._carla_process = None self._carla_simulation = None self._carla_sensors = [] self._carla_actors = {} self._carla_observers = carla_observers self.world = None def reset(self, scenario): if (not self._running): def is_used(port): return (port in [conn.laddr.port for conn in psutil.net_connections()]) server_port = random.randint(15000, 32000) while (is_used(server_port) or is_used((server_port + 1))): server_port += 2 carla_executable = self.config.simulation.carla_path if self.config.rendering.human_mode: carla_start_cmd = [carla_executable, '-windowed', '-ResX={}'.format(600), '-ResY={}'.format(600)] else: carla_start_cmd = [carla_executable, '-RenderOffScreen'] carla_start_cmd += [f'--carla-rpc-port={server_port}', '-quality-level=Epic'] if (os.name != 'nt'): self._carla_process = subprocess.Popen(carla_start_cmd, shell=True, preexec_fn=os.setsid, stdout=open(os.devnull, 'w')) else: self._carla_process = subprocess.Popen(carla_start_cmd) time.sleep(10) self._carla_simulation = CarlaSimulation('127.0.0.1', server_port, self.config.simulation.dt) BridgeHelper.blueprint_library = self._carla_simulation.world.get_blueprint_library() self._running = True for x in self._carla_sensors: x.destroy() for (k, v) in self._carla_actors.items(): self._carla_simulation.destroy_actor(v) self._carla_simulation.tick() self._carla_actors = {} self._carla_sensors = [] BridgeHelper.offset = scenario.sumo_net.getLocationOffset() if os.path.exists(scenario.xodr_path): with open(scenario.xodr_path, encoding='utf-8') as od_file: try: data = od_file.read() except OSError: print('file could not be read.') sys.exit() vertex_distance = 2.0 max_road_length = 500.0 wall_height = 0.0 extra_width = 1.0 self.world = self._carla_simulation.client.generate_opendrive_world(data, carla.OpendriveGenerationParameters(vertex_distance=vertex_distance, max_road_length=max_road_length, wall_height=wall_height, additional_width=extra_width, smooth_junctions=True, enable_mesh_visibility=True)) else: raise ValueError settings = self._carla_simulation.world.get_settings() settings.synchronous_mode = True settings.fixed_delta_seconds = self.config.simulation.dt settings.substepping = False self._carla_simulation.world.apply_settings(settings) def close(self): if self._running: self._carla_simulation.close() del self._carla_simulation if (os.name != 'nt'): os.killpg(os.getpgid(self._carla_process.pid), signal.SIGTERM) else: subprocess.call(['taskkill', '/F', '/T', '/PID', str(self._carla_process.pid)]) self._carla_simulation = None self._carla_process = None self._carla_actors = {} self._carla_sensors = [] self._running = False logging.info('Closed Carla instance...') def simulationStep(self): spawned_actors = [x for x in self._traffic_manager.actor_ids if (x not in self._carla_actors.keys())] destroyed_actors = [x for x in self._carla_actors.keys() if (x not in self._traffic_manager.actor_ids)] for sumo_actor_id in spawned_actors: sumo_actor = self._get_sumo_actor(sumo_actor_id) if (sumo_actor_id == self.config.simulation.egoID): carla_blueprint = self._carla_simulation.world.get_blueprint_library().find('vehicle.audi.a2') else: carla_blueprint = BridgeHelper.get_carla_blueprint(sumo_actor, False) if (carla_blueprint is not None): carla_transform = BridgeHelper.get_carla_transform(sumo_actor.transform, sumo_actor.extent) carla_actor_id = self._carla_simulation.spawn_actor(carla_blueprint, carla_transform) if (carla_actor_id != INVALID_ACTOR_ID): self._carla_actors[sumo_actor_id] = carla_actor_id else: print('No blueprint') continue for sumo_actor_id in destroyed_actors: if (sumo_actor_id in self._carla_actors): self._carla_simulation.destroy_actor(self._carla_actors.pop(sumo_actor_id)) for sumo_actor_id in self._carla_actors: sumo_actor = self._get_sumo_actor(sumo_actor_id) carla_actor_id = self._carla_actors[sumo_actor_id] carla_actor = self._carla_simulation.get_actor(carla_actor_id) carla_transform = BridgeHelper.get_carla_transform(sumo_actor.transform, sumo_actor.extent) carla_lights = BridgeHelper.get_carla_lights_state(carla_actor.get_light_state(), sumo_actor.signals) self._carla_simulation.synchronize_vehicle(carla_actor_id, carla_transform, carla_lights) if (self.config.simulation.egoID in spawned_actors): for observer in self._carla_observers: sensor = self._setup_observer(observer) self._carla_sensors.append(sensor) elif (self.config.simulation.egoID in destroyed_actors): for x in self._carla_sensors: x.destroy() self._carla_sensors = [] try: self._carla_simulation.tick() except Exception: logging.info('Error ticking Carla instance...') self._running = False return True return False def _get_sumo_actor(self, actor_id): actor_state = self._traffic_manager.get_actor_state(actor_id).to_sumo() type_id = actor_state.veh_type vclass = SumoActorClass('passenger') location = actor_state.location rotation = actor_state.rotation transform = carla.Transform(carla.Location(location[0], location[1], location[2]), carla.Rotation(rotation[0], rotation[1], rotation[2])) signals = actor_state.signals extent = carla.Vector3D(*actor_state.extent) return SumoActor(type_id, vclass, transform, signals, extent, color) def _setup_observer(self, observer: CarlaCameraObserver): if (observer.sensor_type == CarlaSensor.RGBCamera): bp_name = 'sensor.camera.rgb' elif (observer.sensor_type == CarlaSensor.DepthCamera): bp_name = 'sensor.camera.depth' else: raise NotImplementedError camera_bp = self._carla_simulation.world.get_blueprint_library().find(bp_name) camera_bp.set_attribute('image_size_x', str(observer.width)) camera_bp.set_attribute('image_size_y', str(observer.height)) camera_bp.set_attribute('sensor_tick', '0.0') for (k, v) in observer.attributes.items(): camera_bp.set_attribute(k, str(v)) carla_actor = self._carla_simulation.get_actor(self._carla_actors[self.config.simulation.egoID]) loc = observer.location rot = observer.rotation camera_transform = carla.Transform(carla.Location(loc[0], loc[1], loc[2]), carla.Rotation(rot[0], rot[1], rot[2])) camera = self._carla_simulation.world.spawn_actor(camera_bp, camera_transform, attach_to=carla_actor, attachment_type=carla.AttachmentType.Rigid) camera.listen(observer.listen) return camera
class TestSamplerDeterministic(unittest.TestCase): def test_to_iterable(self): sampler = TrainingSampler(100, seed=10) dataset = DatasetFromList(list(range(100))) dataset = ToIterableDataset(dataset, sampler) data_loader = data.DataLoader(dataset, num_workers=0, collate_fn=operator.itemgetter(0)) output = list(itertools.islice(data_loader, 100)) self.assertEqual(set(output), set(range(100))) data_loader = data.DataLoader(dataset, num_workers=2, collate_fn=operator.itemgetter(0), worker_init_fn=worker_init_reset_seed) output = list(itertools.islice(data_loader, 100)) self.assertEqual(set(output), set(range(100))) def test_training_sampler_seed(self): seed_all_rng(42) sampler = TrainingSampler(30) data = list(itertools.islice(sampler, 65)) seed_all_rng(42) sampler = TrainingSampler(30) seed_all_rng(999) data2 = list(itertools.islice(sampler, 65)) self.assertEqual(data, data2)
class GELU_SENet(nn.Module): def __init__(self, block, num_blocks, num_classes=10): super(GELU_SENet, self).__init__() self.in_planes = 64 self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False) self.bn1 = nn.BatchNorm2d(64) self.layer1 = self._make_layer(block, 64, num_blocks[0], stride=1) self.layer2 = self._make_layer(block, 128, num_blocks[1], stride=2) self.layer3 = self._make_layer(block, 256, num_blocks[2], stride=2) self.layer4 = self._make_layer(block, 512, num_blocks[3], stride=2) self.linear = nn.Linear(512, num_classes) def _make_layer(self, block, planes, num_blocks, stride): strides = ([stride] + ([1] * (num_blocks - 1))) layers = [] for stride in strides: layers.append(block(self.in_planes, planes, stride)) self.in_planes = planes return nn.Sequential(*layers) def forward(self, x): out = F.gelu(self.bn1(self.conv1(x))) out = self.layer1(out) out = self.layer2(out) out = self.layer3(out) out = self.layer4(out) out = F.avg_pool2d(out, 4) out = out.view(out.size(0), (- 1)) out = self.linear(out) return out
.openapi_version('3.0') .operations('success') def test_process_call_kwargs(testdir, cli, cli_args, mocker, app_type): module = testdir.make_importable_pyfile(hook='\nimport schemathesis\nimport requests\n\\ndef process_call_kwargs(context, case, kwargs):\n if case.app is not None:\n kwargs["follow_redirects"] = False\n else:\n kwargs["allow_redirects"] = False\n ') if (app_type == 'real'): spy = mocker.spy(requests.Session, 'request') else: spy = mocker.spy(werkzeug.Client, 'open') result = cli.main('run', *cli_args, hooks=module.purebasename) assert (result.exit_code == ExitCode.OK), result.stdout if (app_type == 'real'): assert (spy.call_args[1]['allow_redirects'] is False) else: assert (spy.call_args[1]['follow_redirects'] is False)
(('%s.visualize_utils.mmcv' % __name__)) def test_show_pred_gt(mock_mmcv): preds = [[0, 0, 1, 0, 1, 1, 0, 1]] gts = [[0, 0, 1, 0, 1, 1, 0, 1]] show = True win_name = 'test' wait_time = 0 out_file = tempfile.NamedTemporaryFile().name with pytest.raises(AssertionError): visualize_utils.show_pred_gt(np.array([]), gts) with pytest.raises(AssertionError): visualize_utils.show_pred_gt(preds, np.array([])) visualize_utils.show_pred_gt(preds, gts, show, win_name, wait_time, out_file) mock_mmcv.imshow.assert_called_once() mock_mmcv.imwrite.assert_called_once()
class ResNet_context(nn.Module): def __init__(self, num_classes, disable_self_attn, pretrained): self.num_ch_enc = np.array([128, 256, 512, 1024, 2048]) self.disable_self_attn = disable_self_attn super(ResNet_context, self).__init__() self.basedir = os.path.dirname(os.path.abspath(__file__)) self.resnet_model = ResNet(Bottleneck, [3, 4, 23, 3]) if pretrained: pretrained_weights = torch.load(os.path.join(self.basedir, '../splits/resnet101-imagenet.pth')) model_dict = self.resnet_model.state_dict() self.resnet_model.load_state_dict({k: v for (k, v) in pretrained_weights.items() if (k in model_dict)}) self.context = nn.Sequential(nn.Conv2d(2048, 512, kernel_size=3, stride=1, padding=1), ABN_module(512), ASP_OC_Module(512, 256, disable_self_attn=self.disable_self_attn)) self.cls = nn.Conv2d(512, num_classes, kernel_size=1, stride=1, padding=0, bias=True) self.dsn = nn.Sequential(nn.Conv2d(1024, 512, kernel_size=3, stride=1, padding=1), ABN_module(512), nn.Dropout2d(0.1), nn.Conv2d(512, num_classes, kernel_size=1, stride=1, padding=0, bias=True)) def forward(self, x): all_features = self.resnet_model(x) all_features[(- 1)] = self.context(all_features[(- 1)]) all_features[(- 1)] = self.cls(all_features[(- 1)]) return all_features
def level__Tornaria(self): return self.base_ring()(((abs(self.disc()) / self.omega()) / (self.content() ** self.dim())))
def dump_raw_data(contents, file_path): with open(file_path, 'w') as ouf: writer = csv.writer(ouf, delimiter='\t', quotechar='"') for line in contents: writer.writerow(line)
def get_server_partition_dataset(data_path, data_name, part_id): part_name = os.path.join(data_name, ('partition_' + str(part_id))) path = os.path.join(data_path, part_name) if (not os.path.exists(path)): print('Partition file not found.') exit() train_path = os.path.join(path, 'train.txt') local2global_path = os.path.join(path, 'local_to_global.txt') relation_path = os.path.join(path, 'relation_count.txt') dataset = PartitionKGDataset(relation_path, train_path, local2global_path, read_triple=False) n_entities = _file_line(os.path.join(path, 'partition_book.txt')) local_to_global = [] with open(local2global_path) as f: for line in f: local_to_global.append(int(line)) global_to_local = ([0] * n_entities) for i in range(len(local_to_global)): global_id = local_to_global[i] global_to_local[global_id] = i local_to_global = None return (global_to_local, dataset)
def advect(): for i in range(n_tracer): p = tracer[i] v1 = compute_u_full(p) v2 = compute_u_full((p + ((v1 * dt) * 0.5))) v3 = compute_u_full((p + ((v2 * dt) * 0.75))) tracer[i] += (((((2 / 9) * v1) + ((1 / 3) * v2)) + ((4 / 9) * v3)) * dt)
def test_superb_er(): with tempfile.TemporaryDirectory() as tempdir: with pseudo_audio([10, 2, 1, 8, 5]) as (wav_paths, num_samples): class TestER(SuperbER): def default_config(self) -> dict: config = super().default_config() config['prepare_data'] = {} return config def prepare_data(self, prepare_data: dict, target_dir: str, cache_dir: str, get_path_only: bool=False): ids = [Path(path).stem for path in wav_paths] labels = ['a', 'b', 'a', 'c', 'd'] start_secs = [0.0, 0.1, 0.2, None, 0.0] end_secs = [5.2, 1.0, 0.3, None, 4.9] df = pd.DataFrame(data={'id': ids, 'wav_path': wav_paths, 'label': labels, 'start_sec': start_secs, 'end_sec': end_secs}) train_csv = (target_dir / 'train.csv') valid_csv = (target_dir / 'valid.csv') test_csv = (target_dir / 'test.csv') df.to_csv(train_csv) df.to_csv(valid_csv) df.to_csv(test_csv) return (train_csv, valid_csv, [test_csv]) problem = TestER() config = problem.default_config() config['target_dir'] = tempdir config['device'] = 'cpu' config['train']['total_steps'] = 4 config['train']['log_step'] = 1 config['train']['eval_step'] = 2 config['train']['save_step'] = 2 config['eval_batch'] = 2 config['build_upstream']['name'] = 'fbank' problem.run(**config)
class DenseLayer(nn.Module): def __init__(self, num_input_features, growth_rate, bn_size, norm_layer=BatchNormAct2d, drop_rate=0.0, memory_efficient=False): super(DenseLayer, self).__init__() (self.add_module('norm1', norm_layer(num_input_features)),) (self.add_module('conv1', nn.Conv2d(num_input_features, (bn_size * growth_rate), kernel_size=1, stride=1, bias=False)),) (self.add_module('norm2', norm_layer((bn_size * growth_rate))),) (self.add_module('conv2', nn.Conv2d((bn_size * growth_rate), growth_rate, kernel_size=3, stride=1, padding=1, bias=False)),) self.drop_rate = float(drop_rate) self.memory_efficient = memory_efficient def bottleneck_fn(self, xs): concated_features = torch.cat(xs, 1) bottleneck_output = self.conv1(self.norm1(concated_features)) return bottleneck_output def any_requires_grad(self, x): for tensor in x: if tensor.requires_grad: return True return False .unused def call_checkpoint_bottleneck(self, x): def closure(*xs): return self.bottleneck_fn(xs) return cp.checkpoint(closure, *x) ._overload_method def forward(self, x): pass ._overload_method def forward(self, x): pass def forward(self, x): if isinstance(x, torch.Tensor): prev_features = [x] else: prev_features = x if (self.memory_efficient and self.any_requires_grad(prev_features)): if torch.jit.is_scripting(): raise Exception('Memory Efficient not supported in JIT') bottleneck_output = self.call_checkpoint_bottleneck(prev_features) else: bottleneck_output = self.bottleneck_fn(prev_features) new_features = self.conv2(self.norm2(bottleneck_output)) if (self.drop_rate > 0): new_features = F.dropout(new_features, p=self.drop_rate, training=self.training) return new_features
.gpu def test_pythonmode(): def runs_on_gpu(a: (dace.float64[20] StorageType.GPU_Global), b: (dace.float64[20] StorageType.GPU_Global)): for i in (dace.map[0:20] ScheduleType.GPU_Device): b[i] = (a[i] + 1.0) gpu_a = cupy.random.rand(20) gpu_b = cupy.random.rand(20) runs_on_gpu(gpu_a, gpu_b) assert cupy.allclose(gpu_b, (gpu_a + 1))
def extend_phyche_index(original_index, extend_index): if (0 == len(extend_index)): return original_index for key in list(original_index.keys()): original_index[key].extend(extend_index[key]) return original_index
class ResnetBlock(nn.Module): def __init__(self, dim, kernel_size=1, padding_type='zero', norm_layer=nn.BatchNorm2d, use_dropout=False, use_bias=True, act=None): super(ResnetBlock, self).__init__() self.conv_block = self.build_conv_block(dim, kernel_size, padding_type, norm_layer, use_dropout, use_bias, act) def build_conv_block(self, dim, kernel_size, padding_type, norm_layer, use_dropout, use_bias, act=nn.GELU()): conv_block = [] p = 0 if (padding_type == 'reflect'): conv_block += [nn.ReflectionPad2d(1)] elif (padding_type == 'replicate'): conv_block += [nn.ReplicationPad2d(1)] elif (padding_type == 'zero'): p = 0 else: raise NotImplementedError(('padding [%s] is not implemented' % padding_type)) conv_block += [nn.Conv2d(dim, dim, kernel_size=kernel_size, padding=p, bias=use_bias)] if norm_layer: conv_block += [norm_layer(dim, momentum=0.1)] if act: conv_block += [act] if use_dropout: conv_block += [nn.Dropout(0.5)] p = 0 if (padding_type == 'reflect'): conv_block += [nn.ReflectionPad2d(1)] elif (padding_type == 'replicate'): conv_block += [nn.ReplicationPad2d(1)] elif (padding_type == 'zero'): p = 0 else: raise NotImplementedError(('padding [%s] is not implemented' % padding_type)) conv_block += [nn.Conv2d(dim, dim, kernel_size=kernel_size, padding=p, bias=use_bias)] if norm_layer: conv_block += [norm_layer(dim, momentum=0.1)] return nn.Sequential(*conv_block) def forward(self, x): out = (x + self.conv_block(x)) return out
def bert_large_uncased_whole_word_maskings_384_4p_bw12_async_pipedream(): return dict(model_type='bert_squad', model_name_or_path='bert-large-uncased-whole-word-masking', do_lower_case=True, output_past=False, stateless_tied=False, explicitly_set_dict={'precompute_attention_mask': True, 'return_dict': False}, do_resize_token_embedding=False)
class FrameStack(gym.Wrapper): def __init__(self, env, n_frames): super().__init__(env) self.n_frames = n_frames self.frames = deque([], maxlen=n_frames) shape = ((n_frames,) + env.observation_space.shape) self.observation_space = gym.spaces.Box(low=np.min(env.observation_space.low), high=np.max(env.observation_space.high), shape=shape, dtype=env.observation_space.dtype) def reset(self): obs = self.env.reset() for _ in range(self.n_frames): self.frames.append(obs) return self._get_ob() def step(self, action): (obs, reward, done, info) = self.env.step(action) self.frames.append(obs) return (self._get_ob(), reward, done, info) def _get_ob(self): return np.stack(self.frames, axis=0)