在編寫RNN程序時，一個很常見的函數就是sequence_loss_by_example
loss = tf.contrib.legacy_seq2seq.sequence_loss_by_example(logits_list, targets_list, weights_list, average_across_timesteps)
這個函數在contrib中的legacy（遺產）中，可見這個函數不是tensorflow支持的官方函數。
import numpy as np
import tensorflow as tf
def sequence_loss_by_example(logits,
targets,
weights,
average_across_timesteps=True,
softmax_loss_function=None,
name=None):
"""Weighted cross-entropy loss for a sequence of logits (per example).
Args:
logits: List of 2D Tensors of shape [batch_size x num_decoder_symbols].
targets: List of 1D batch-sized int32 Tensors of the same length as logits.
weights: List of 1D batch-sized float-Tensors of the same length as logits.
average_across_timesteps: If set, divide the returned cost by the total
label weight.
softmax_loss_function: Function (labels, logits) -> loss-batch
to be used instead of the standard softmax (the default if this is None).
**Note that to avoid confusion, it is required for the function to accept
named arguments.**
name: Optional name for this operation, default: "sequence_loss_by_example".
Returns:
1D batch-sized float Tensor: The log-perplexity for each sequence.
Raises:
ValueError: If len(logits) is different from len(targets) or len(weights).
"""
# 此三者都是列表，長度都應該相同
if len(targets) != len(logits) or len(weights) != len(logits):
raise ValueError("Lengths of logits, weights, and targets must be the same "
"%d, %d, %d." % (len(logits), len(weights), len(targets)))
with tf.name_scope(name, "sequence_loss_by_example",
logits + targets + weights):
log_perp_list = []
# 計算每個時間片的損失
for logit, target, weight in zip(logits, targets, weights):
if softmax_loss_function is None:
# 默認使用sparse
target = tf.reshape(target, [-1])
crossent = tf.nn.sparse_softmax_cross_entropy_with_logits(
labels=target, logits=logit)
else:
crossent = softmax_loss_function(labels=target, logits=logit)
log_perp_list.append(crossent * weight)
# 把各個時間片的損失加起來
log_perps = tf.add_n(log_perp_list)
# 對各個時間片的損失求平均數
if average_across_timesteps:
total_size = tf.add_n(weights)
total_size += 1e-12 # Just to avoid division by 0 for all-0 weights.
log_perps /= total_size
return log_perps
"""
考慮many2many形式的RNN用法，每次輸入一個就會得到一個輸出
這些輸出需要計算平均損失，我們可以指定：
* 每個樣本的權重
* 每個時間片的權重
"""
sample_count = 4
target_count = 3
frame_count = 2
# 各個時間片我的答案
logits = [tf.random_uniform((sample_count, target_count)) for i in range(frame_count)]
# 各個時間片的真正答案
targets = [tf.constant(np.random.randint(0, target_count, (sample_count,))) for i in range(frame_count)]
# 每個時間片，每個樣本的權重。利用weights我們可以指定時間片權重和樣本權重
weights = [tf.ones((sample_count,), dtype=tf.float32) * (i + 1) for i in range(frame_count)]
loss1 = sequence_loss_by_example(logits, targets, weights, average_across_timesteps=True)
loss = tf.contrib.legacy_seq2seq.sequence_loss_by_example(logits, targets, weights, True)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
x, y, = sess.run([loss, loss1])
print(x)
print(y)
print(x.shape, y.shape)
這個函數非常有用，tensorflow.nn中的sparse_softmax_cross_entropy無法指定樣本的權重，這個函數可以。
使用時，只需要傳入一個時間片即可。如果各個樣本權重都為1，最后得到的結果跟sparse_softmax_cross_entropy得到的結果是一樣的。