I want to use variable length input with dynamic RNN of tensorflow, but I don't know how to padding

Question

As an example, input tensor is a batch size 3, a maximum time sequence length 4 and dimension of feature value is 2

[ [ [1, 2], [3, 4], [5, 6], [7, 8] ], [ [9, 10] ] ]

, And in order to make it an input to RNN

[ [ [1, 2], [3, 4], [5, 6], [7, 8] ], [ [9, 10], [0,0], [0,0], [0,0] ] ]

I would like to have a form like this, is there a good way to do it? Since we assume that the tensor to be input to this RNN is the feature amount extracted by another neural network before entering RNN, I would like to know a method that will not break the calculation graph.

Answer 1

我认为可能需要tf.train.batch （ https://www.tensorflow.org/api_docs/python/tf/train/batch ）并使用dynamic_pad = True 。