TensorFlow的可训练重量sequence_loss_by_example()

时间:2016-03-11 02:05:05

标签: python tensorflow

我希望在seq2seq.sequence_loss_by_example()中有一个可训练的重量,例如

w = tf.get_variable("w", [batch_size*num_steps])
loss = seq2seq.sequence_loss_by_example([logits_1],
            [tf.reshape(self._targets, [-1])],
            w,vocab_size_all)

但是,运行此代码会出现以下错误:

seq2seq.py, line 654, in sequence_loss_by_example
if len(targets) != len(logits) or len(weights) != len(logits):

根据seq2seq.py中此函数的docstring:

weights: list of 1D batch-sized float-Tensors of the same length as logits.

它需要" Tensor",但我想传递tf.Variable。有没有办法在这个功能中有可训练的权重?

1 个答案:

答案 0 :(得分:2)

在TensorFlow中,tf.Variable可以在tf.Tensor(具有相同元素类型和形状)的任何地方使用。

因此,您要定义一个可训练的权重,您可以将tf.Variable个对象列表作为weights参数传递给seq2seq.sequence_loss_by_example()。例如,您可以执行以下操作:

# Defines a list of `num_steps` variables, each 1-D with length `batch_size`.
weights = [tf.get_variable("w", [batch_size]) for _ in range(num_steps)]

loss = seq2seq.sequence_loss_by_example([logits_1, ..., logits_n],
                                        [targets_1, ..., targets_n],
                                        weights,
                                        vocab_size_all)