Tensorflow seq2seq: Tensor' object is not iterable

Question

I am using seq2seq below code, I found below error:

cell = tf.nn.rnn_cell.BasicLSTMCell(size)
a, b = tf.nn.dynamic_rnn(cell, seq_input, dtype=tf.float32)
cell_a = tf.contrib.rnn.OutputProjectionWrapper(cell, frame_dim)
dec_output= tf.contrib.legacy_seq2seq.rnn_decoder(seq_input, b, cell_a)

but I get the error:

TypeError: 'Tensor' object is not iterable.

I checked and it comes from seq2seq line.

Answer 1

Looks like seq_input is a tensor, not a list of tensors. A single tensor works fine for tf.nn.dynamic_rnn, but rnn_decoder requires unstacking the sequence into a list of tensors:

decoder_inputs: A list of 2D Tensors [batch_size x input_size].

In the source code, you can see that the implementation simply iterates over decoder_inputs in a for loop.