def extract_episodes(self):
collected = collect_episodes(
ReinforcementAI(self.judge), episode_cnt, episode_length)
episodes = []
for c in collected:
obs = torch.cat(
tuple(Translator.encode_board(b).unsqueeze(0)
for b in c['boards']), 0)
idx = torch.tensor(
tuple(Translator.encode_move_idx(m) for m in c['moves']),
dtype=torch.long)
prob = self.judge.forward(obs)[0][
torch.arange(idx.size()[0], dtype=torch.long), idx]
ext = c['extrinsic']
episodes += [
{'obs': obs, 'idx': idx, 'prob': prob, 'ext': ext,
'over': c['boards'][-1].is_game_over()}]
return episodes
