Tensorflow 中 RNN softmax输出层采样 tf.multinomial(logits, num_samples)

Tensorflow 中，想要使用sequence to sequence 模型，在RNN的输出端采样(sampling)，可以在softmax层之后，做简单的log p 再用tf.multinomial()来实现：

tf.multinomial(logits, num_samples, seed=None, name=None)

例如：

# samples has shape [1, 5], where each value is either 0 or 1 with equal
# probability.
samples = tf.multinomial(tf.log([[10., 10.]]), 5)

Args:

• logits: 2-D Tensor with shape [batch_size, num_classes]. Each slice [i, :] represents the unnormalized log probabilities for all classes.
• num_samples: 0-D. Number of independent samples to draw for each row slice.
• seed: A Python integer. Used to create a random seed for the distribution. See tf.set_random_seed for behavior.
• name: Optional name for the operation.

Returns:

The drawn samples of shape [batch_size, num_samples].

Defined in tensorflow/python/ops/random_ops.py.