Inconsistent BatchSize

[HMJiangGatech]

airdialogue_model/self_play.py

Lines 132 to 144 in 5bdc799

	for i in tqdm(list(range(0, ceil))):
	start_ind = i * batch_size
	end_ind = min(i * batch_size + batch_size, len(selfplay_data))
	batch_data = selfplay_data[start_ind:end_ind]
	batch_kb = selfplay_kb[start_ind:end_ind]
	# we indicaet to let agent1 to talk first. Keep in mind that we will
	# swap between agent1 and agent2.
	speaker = flip % 2
	generated_data, _, summary = dialogue.talk(hparams.max_dialogue_len,
	batch_data, batch_kb, agent1,
	agent2, worker_step,
	batch_size, speaker)

In line 144, we should replace batch_size by end_ind-start_ind. Otherwise, there will be an inconsistent batchsize issue in the last iteration.

[josephch405]

This will be fixed in #6

[josephch405]

Address in #10