def choice_1_COL_RANGE_FIRST(dict_nouns): def _reset_params(): Parameters.PARAMETERS_CHOICE_0_1.CHOICE_1_COL_RANGE_FIRST = 1.01 def _change_params_for_step(cur_step,alpha): _reset_params() Parameters.PARAMETERS_CHOICE_0_1.CHOICE_1_COL_RANGE_FIRST += cur_step*alpha _reset_params() _alpha = 0.02 _max_step = 10 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step, _alpha) _reset_params() return (_best_result,_best_params)
def nbest_similarity(dict_nouns): def _reset_params(): Parameters.PARAMETERS.N_BEST_CALCULATE_SIMILARITY = 1 def _change_params_for_step(cur_step, alpha): _reset_params() Parameters.PARAMETERS.N_BEST_CALCULATE_SIMILARITY += cur_step*alpha _reset_params() _alpha = 1 _max_step = 9 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step,_alpha) _reset_params() return (_best_result,_best_params)
def jaccard_weight(dict_nouns): def _reset_params(): Parameters.PARAMETERS.JACCARD_WEIGHT = 0.01 def _change_params_for_step(cur_step, alpha): _reset_params() Parameters.PARAMETERS.JACCARD_WEIGHT += cur_step*alpha _reset_params() _alpha = 0.02 _max_step = 20 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step,_alpha) _reset_params() return (_best_result,_best_params)
def feature_POS(dict_nouns): def _reset_params(): Parameters.PARAMETERS.POS_FEATURE_n = 1 Parameters.PARAMETERS.POS_FEATURE_v = 0 def _change_params_for_step(cur_step): _reset_params() if cur_step == 1: Parameters.PARAMETERS.POS_FEATURE_v = 1 _reset_params() _max_step = 1 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step) _reset_params() return (_best_result,_best_params)
def feature_dict(dict_nouns): def _reset_params(): Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_sd = 1 Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_d = 1 Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_xh = 1 Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_x = 1 def _change_params_for_step(cur_step): _reset_params() if cur_step == 1: Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_xh = 0 elif cur_step == 2: Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_x = 0 elif cur_step == 3: Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_xh = 0 Parameters.PARAMETERS.DICT_OX_FEATURE_RELATION_x = 0 _reset_params() _max_step = 3 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step) _reset_params() return (_best_result,_best_params)
def feature_wn(dict_nouns): def _reset_params(): Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_definition = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 0 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 0 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_part_meronyms = 0 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_member_holonyms = 0 def _change_params_for_step(cur_step): _reset_params() if cur_step == 1: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 elif cur_step == 2: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 1 elif cur_step == 3: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_part_meronyms = 1 elif cur_step == 4: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_member_holonyms = 1 elif cur_step == 5: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 1 elif cur_step == 6: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_part_meronyms = 1 elif cur_step == 7: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_member_holonyms = 1 elif cur_step == 8: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_part_meronyms = 1 elif cur_step == 9: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_member_holonyms = 1 elif cur_step == 10: Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hypernyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_hyponyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_part_meronyms = 1 Parameters.PARAMETERS.DICT_WN_FEATURE_RELATION_member_holonyms = 1 _reset_params() _max_step = 10 _current_step = 0 _best_result = 0 _best_params = Parameters.get_current_params() while _current_step <= _max_step: (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() _current_step += 1 _change_params_for_step(_current_step) _reset_params() return (_best_result,_best_params)
def append_result_to_file(precision,recall,accuracy): params_value = Parameters.get_current_params() params_value.append(precision) params_value.append(recall) params_value.append(accuracy) append_params_and_result_to_file(params_value)
def train_with_random(dict_nouns): _best_result = 0; _best_params = Parameters.get_current_params() for i in range(0,1000): # random_all_feature Parameters.random_params_values() # calculate_result -> current best (precision, recall, accuracy) = SimilarityWordnetOxford.similarityWords(dict_nouns) if accuracy > _best_result: _best_result = accuracy _best_params = Parameters.get_current_params() (best_result_loop, best_params_loop) = train_from_base(dict_nouns) if best_result_loop>= _best_result: _best_result = best_result_loop _best_params = best_params_loop Parameters.set_params_from_arr(_best_params) WriteParametersAndResult.append_params_and_result_to_file(_best_params)
def train_from_base(dict_nouns): # run this feature in range -> choice best result lower = -1 previous_random = -1; best_result_loop = 0 best_params_loop = Parameters.get_current_params() while lower < 7: chosen_feature = random.randint(0,6) while chosen_feature == previous_random: chosen_feature = random.randint(0,6) previous_random = chosen_feature best_result = 0 best_params = [] if lower >= 0: chosen_feature = lower if chosen_feature == 0: (best_result, best_params) = jaccard_weight(dict_nouns) elif chosen_feature == 1: (best_result, best_params) = choice_1_COL_RANGE_FIRST(dict_nouns) elif chosen_feature == 2: (best_result, best_params) = choice_N_N_RANGE_FIRST(dict_nouns) elif chosen_feature == 3: (best_result, best_params) = feature_wn(dict_nouns) elif chosen_feature == 4: (best_result, best_params) = feature_dict(dict_nouns) elif chosen_feature == 5: (best_result, best_params) = feature_POS(dict_nouns) else: (best_result, best_params) = nbest_similarity(dict_nouns) # compare with _best if best_result >= best_result_loop: best_result_loop = best_result best_params_loop = best_params lower = -1 else: lower += 1 Parameters.set_params_from_arr(best_params_loop) return (best_result_loop, best_params_loop)