File size: 7,679 Bytes
7393a38 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
# coding: utf-8
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import os
import time
import collections
import numpy as np
import tensorflow as tf
from utils import dtype
def batch_indexer(datasize, batch_size):
"""Just divide the datasize into batched size"""
dataindex = np.arange(datasize).tolist()
batchindex = []
for i in range(datasize // batch_size):
batchindex.append(dataindex[i * batch_size: (i + 1) * batch_size])
if datasize % batch_size > 0:
batchindex.append(dataindex[-(datasize % batch_size):])
return batchindex
def token_indexer(dataset, token_size):
"""Divide the dataset into token-based batch"""
# assume dataset format: [(len1, len2, ..., lenN)]
dataindex = np.arange(len(dataset)).tolist()
batchindex = []
_batcher = [0.] * len(dataset[0])
_counter = 0
i = 0
while True:
if i >= len(dataset): break
# attempt put this datapoint into batch
_batcher = [max(max_l, l)
for max_l, l in zip(_batcher, dataset[i])]
_counter += 1
for l in _batcher:
if _counter * l >= token_size:
# when an extreme instance occur, handle it by making a 1-size batch
if _counter > 1:
batchindex.append(dataindex[i-_counter+1: i])
i -= 1
else:
batchindex.append(dataindex[i: i+1])
_counter = 0
_batcher = [0.] * len(dataset[0])
break
i += 1
_counter = sum([len(slice) for slice in batchindex])
if _counter != len(dataset):
batchindex.append(dataindex[_counter:])
return batchindex
def mask_scale(value, mask, scale=None):
"""Prepared for masked softmax"""
if scale is None:
scale = dtype.inf()
return value + (1. - mask) * (-scale)
def valid_apply_dropout(x, dropout):
"""To check whether the dropout value is valid, apply if valid"""
if dropout is not None and 0. <= dropout <= 1.:
return tf.nn.dropout(x, 1. - dropout)
return x
def layer_dropout(dropped, no_dropped, dropout_rate):
"""Layer Dropout"""
pred = tf.random_uniform([]) < dropout_rate
return tf.cond(pred, lambda: dropped, lambda: no_dropped)
def label_smooth(labels, vocab_size, factor=0.1):
"""Smooth the gold label distribution"""
if 0. < factor < 1.:
n = tf.cast(vocab_size - 1, tf.float32)
p = 1. - factor
q = factor / n
t = tf.one_hot(tf.cast(tf.reshape(labels, [-1]), tf.int32),
depth=vocab_size, on_value=p, off_value=q)
normalizing = -(p * tf.log(p) + n * q * tf.log(q + 1e-20))
else:
t = tf.one_hot(tf.cast(tf.reshape(labels, [-1]), tf.int32),
depth=vocab_size)
normalizing = 0.
return t, normalizing
def closing_dropout(params):
"""Removing all dropouts"""
for k, v in params.values().items():
if 'dropout' in k:
setattr(params, k, 0.0)
# consider closing label smoothing
if 'label_smoothing' in k:
setattr(params, k, 0.0)
return params
def dict_update(d, u):
"""Recursive update dictionary"""
for k, v in u.items():
if isinstance(v, collections.Mapping):
d[k] = dict_update(d.get(k, {}), v)
else:
d[k] = v
return d
def shape_list(x):
# Copied from Tensor2Tensor
"""Return list of dims, statically where possible."""
x = tf.convert_to_tensor(x)
# If unknown rank, return dynamic shape
if x.get_shape().dims is None:
return tf.shape(x)
static = x.get_shape().as_list()
shape = tf.shape(x)
ret = []
for i in range(len(static)):
dim = static[i]
if dim is None:
dim = shape[i]
ret.append(dim)
return ret
def get_shape_invariants(tensor):
# Copied from Tensor2Tensor
"""Returns the shape of the tensor but sets middle dims to None."""
shape = tensor.shape.as_list()
for i in range(1, len(shape) - 1):
shape[i] = None
return tf.TensorShape(shape)
def merge_neighbor_dims(x, axis=0):
"""Merge neighbor dimension of x, start by axis"""
if len(x.get_shape().as_list()) < axis + 2:
return x
shape = shape_list(x)
shape[axis] *= shape[axis+1]
shape.pop(axis+1)
return tf.reshape(x, shape)
def unmerge_neighbor_dims(x, depth, axis=0):
"""Inverse of merge_neighbor_dims, axis by depth"""
if len(x.get_shape().as_list()) < axis + 1:
return x
shape = shape_list(x)
width = shape[axis] // depth
new_shape = shape[:axis] + [depth, width] + shape[axis+1:]
return tf.reshape(x, new_shape)
def expand_tile_dims(x, depth, axis=1):
"""Expand and Tile x on axis by depth"""
x = tf.expand_dims(x, axis=axis)
tile_dims = [1] * x.shape.ndims
tile_dims[axis] = depth
return tf.tile(x, tile_dims)
def gumbel_noise(shape, eps=None):
"""Generate gumbel noise shaped by shape"""
if eps is None:
eps = dtype.epsilon()
u = tf.random_uniform(shape, minval=0, maxval=1)
return -tf.log(-tf.log(u + eps) + eps)
def log_prob_from_logits(logits):
"""Probability from un-nomalized logits"""
return logits - tf.reduce_logsumexp(logits, axis=-1, keepdims=True)
def batch_coordinates(batch_size, beam_size):
"""Batch coordinate indices under beam_size"""
batch_pos = tf.range(batch_size * beam_size) // beam_size
batch_pos = tf.reshape(batch_pos, [batch_size, beam_size])
return batch_pos
def variable_printer():
"""Print parameters"""
all_weights = {v.name: v for v in tf.trainable_variables()}
total_size = 0
for v_name in sorted(list(all_weights)):
v = all_weights[v_name]
tf.logging.info("%s\tshape %s", v.name[:-2].ljust(80),
str(v.shape).ljust(20))
v_size = np.prod(np.array(v.shape.as_list())).tolist()
total_size += v_size
tf.logging.info("Total trainable variables size: %d", total_size)
def uniform_splits(total_size, num_shards):
"""Split the total_size into uniform num_shards lists"""
size_per_shards = total_size // num_shards
splits = [size_per_shards] * (num_shards - 1) + \
[total_size - (num_shards - 1) * size_per_shards]
return splits
def fetch_valid_ref_files(path):
"""Extracting valid reference files according to MT convention"""
path = os.path.abspath(path)
if tf.gfile.Exists(path):
return [path]
if not tf.gfile.Exists(path + ".ref0"):
tf.logging.warn("Invalid Reference Format {}".format(path))
return None
num = 0
files = []
while True:
file_path = path + ".ref%s" % num
if tf.gfile.Exists(file_path):
files.append(file_path)
else:
break
num += 1
return files
def get_session(gpus):
"""Config session with GPUS"""
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
if len(gpus) > 0:
device_str = ",".join([str(i) for i in gpus])
sess_config.gpu_options.visible_device_list = device_str
sess = tf.Session(config=sess_config)
return sess
def flatten_list(values):
"""Flatten a list"""
return [v for value in values for v in value]
def time_str(t=None):
"""String format of the time long data"""
if t is None:
t = time.time()
ts = time.strftime("[%Y-%m-%d %H:%M:%S]", time.localtime(t))
return ts
|