Encoder inputs are less than source_seq_len by 1

[Ammola]

Hi @una-dinosauria
I'm wondering why did you choose encoder_inputs to be less than source_seq_len by 1?
Line 453
While you choose decoder_outputs and decoder_inputs to be the same as target_seq_len.

human-motion-prediction/src/seq2seq_model.py

Lines 435 to 455 in c9a2774

	def get_batch( self, data, actions ):
	"""Get a random batch of data from the specified bucket, prepare for step.
	Args
	data: a list of sequences of size n-by-d to fit the model to.
	actions: a list of the actions we are using
	Returns
	The tuple (encoder_inputs, decoder_inputs, decoder_outputs);
	the constructed batches have the proper format to call step(...) later.
	"""
	# Select entries at random
	all_keys = list(data.keys())
	chosen_keys = np.random.choice( len(all_keys), self.batch_size )
	# How many frames in total do we need?
	total_frames = self.source_seq_len + self.target_seq_len
	encoder_inputs = np.zeros((self.batch_size, self.source_seq_len-1, self.input_size), dtype=float)
	decoder_inputs = np.zeros((self.batch_size, self.target_seq_len, self.input_size), dtype=float)
	decoder_outputs = np.zeros((self.batch_size, self.target_seq_len, self.input_size), dtype=float)

[una-dinosauria]

Because the last frame of encoder_inputs is given as a seed to the decoder.