X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=escape.py;h=f51863b0c42e9c4c0d38fa334296c6ebfca9ed04;hb=19ec7f3e4030ddece2647983dcf1bed5eb0d9544;hp=43843f0bfbea776410b42cccaf56c534a673e150;hpb=9eadd2cd6913a0de53b4b0f526157497e8d14381;p=picoclvr.git diff --git a/escape.py b/escape.py index 43843f0..f51863b 100755 --- a/escape.py +++ b/escape.py @@ -94,7 +94,7 @@ def generate_episodes(nb, height=6, width=6, T=10, nb_walls=3): ) hit = (hit > 0).long() - assert hit.min() == 0 and hit.max() <= 1 + # assert hit.min() == 0 and hit.max() <= 1 rewards[:, t + 1] = -hit + (1 - hit) * agent[:, t + 1, -1, -1] @@ -133,27 +133,27 @@ def episodes2seq(states, actions, rewards, lookahead_delta=None): r = rewards[:, :, None] rewards = (r + 1) + first_rewards_code - assert ( - states.min() >= first_state_code - and states.max() < first_state_code + nb_state_codes - ) - assert ( - actions.min() >= first_actions_code - and actions.max() < first_actions_code + nb_actions_codes - ) - assert ( - rewards.min() >= first_rewards_code - and rewards.max() < first_rewards_code + nb_rewards_codes - ) + # assert ( + # states.min() >= first_state_code + # and states.max() < first_state_code + nb_state_codes + # ) + # assert ( + # actions.min() >= first_actions_code + # and actions.max() < first_actions_code + nb_actions_codes + # ) + # assert ( + # rewards.min() >= first_rewards_code + # and rewards.max() < first_rewards_code + nb_rewards_codes + # ) if lookahead_delta is None: return torch.cat([states, actions, rewards], dim=2).flatten(1) else: - assert ( - lookahead_rewards.min() >= first_lookahead_rewards_code - and lookahead_rewards.max() - < first_lookahead_rewards_code + nb_lookahead_rewards_codes - ) + # assert ( + # lookahead_rewards.min() >= first_lookahead_rewards_code + # and lookahead_rewards.max() + # < first_lookahead_rewards_code + nb_lookahead_rewards_codes + # ) return torch.cat([states, actions, rewards, lookahead_rewards], dim=2).flatten( 1 )