deepword.tools package

Submodules

deepword.tools.build_test_set module

deepword.tools.build_test_set.build_test_set(f_all_types, all_game_dir)

Random select two games for each type as test set :param f_all_types: file contains type info, one per line :param all_game_dir: dir contains all games :return: names of selected test games, without file extension

deepword.tools.clean_hs2tj module

deepword.tools.clean_hs2tj.clean_data(tp, hs)
deepword.tools.clean_hs2tj.clean_hs2tj(hash_states2tjs, tjs)

Due to code problem, some hash_states2tjs contains states that don’t have tjs companions. These states should be removed, otherwise the request for their companions would cause error.

Parameters

  • hash_states2tjs – a map from states to tjs. States are composed of observation + inventory

  • tjs –

Returns

deepword.tools.clean_hs2tj.load_snapshot(raw_tjs_path, hs2tj_path)
deepword.tools.clean_hs2tj.main(data_path, n_data)

deepword.tools.collect_game_elements module

class deepword.tools.collect_game_elements.CollectorAgent

Bases: object

act(obs, scores, dones, infos)
post_run()
pre_run()
classmethod request_infos()
class deepword.tools.collect_game_elements.FPCollector

Bases: deepword.tools.collect_game_elements.CollectorAgent

act(obs, scores, dones, infos)
post_run()
pre_run()
classmethod request_infos()
class deepword.tools.collect_game_elements.Main

Bases: object

classmethod collect_floor_plans(save_to, game_dir, f_games=None, nb_episodes=10)

Collect floor plans with games in game_dir or specified by f_games.

Parameters

  • save_to – save to path, must be a npz file, e.g. floor_plan.npz

  • game_dir –

  • f_games –

  • nb_episodes –

Returns

classmethod collect_others(game_dir, f_games=None, nb_episodes=1)
class deepword.tools.collect_game_elements.OneStepCollector

Bases: deepword.tools.collect_game_elements.CollectorAgent

act(obs, scores, dones, infos)
post_run()
pre_run()
classmethod request_infos()
deepword.tools.collect_game_elements.contain_theme_words(theme_words, actions)
deepword.tools.collect_game_elements.contain_words(sentence, words)
deepword.tools.collect_game_elements.get_room(master)
deepword.tools.collect_game_elements.run_games(agent, game_files, nb_episodes, max_steps)

deepword.tools.compare_eval_results module

deepword.tools.compare_eval_results.diff_from_keys(keys, j_result, j_result2)
deepword.tools.compare_eval_results.format_steps(mcs, max_n_vals)
deepword.tools.compare_eval_results.main(f_result, f_result2)
deepword.tools.compare_eval_results.most_common_steps(scoring_steps)

  1. find the most common scoring steps,

2. if ties, choose the one with the best score. :param scoring_steps: :return:

deepword.tools.compare_eval_results.reason_of_failure(last_commands)
deepword.tools.compare_eval_results.summary_from_keys(keys, j_result)

deepword.tools.diff_train_w_test module

deepword.tools.diff_train_w_test.diff_names(f_games, test_set)

deepword.tools.play_game module

class deepword.tools.play_game.HumanAgent

Bases: deepword.tools.collect_game_elements.CollectorAgent

act(obs, scores, dones, infos)
post_run()
pre_run()
classmethod request_infos()
deepword.tools.play_game.human_play(game_file, nb_episodes=1, max_steps=100)

deepword.tools.read_eval_failed_reason module

deepword.tools.read_eval_failed_reason.count_failed_twice(g_name, j_game)
deepword.tools.read_eval_failed_reason.main(f_result)
deepword.tools.read_eval_failed_reason.parse_one_game(g_name, j_game)
deepword.tools.read_eval_failed_reason.repeat_action_error(g_name, j_game)
deepword.tools.read_eval_failed_reason.summary_from_keys(keys, j_result)

deepword.tools.read_eval_results module

deepword.tools.read_eval_results.main(f_result, f_result2=None)
deepword.tools.read_eval_results.merge_eval_results(j_res1, j_res2=None)
deepword.tools.read_eval_results.summary_from_keys(keys, eval_results)

deepword.tools.replay_from_log module

class deepword.tools.replay_from_log.ReplayAgent(fn_game, fn_log)

Bases: deepword.tools.collect_game_elements.CollectorAgent

act(obs, scores, dones, infos)
end_episode()
classmethod extract_eval_results(fn_log)
post_run()
pre_run()
classmethod request_infos()
start_episode()
deepword.tools.replay_from_log.play(game_file, log_file, nb_episodes=2, max_steps=100)

Module contents