def model_train_predict(essays_TD, essays_VD, extractor_names, cost_function_name, ngrams, stemmed, beta, max_epochs):
extractors = get_functions_by_name(extractor_names, all_extractor_fns)
# get single cost function
cost_fn = get_functions_by_name([cost_function_name], all_cost_functions)[0]
assert cost_fn is not None, "Cost function look up failed"
# Ensure all extractors located
assert len(extractors) == len(extractor_names), "number of extractor functions does not match the number of names"
template_feature_extractor = NonLocalTemplateFeatureExtractor(extractors=extractors)
if stemmed:
ngram_extractor = NgramExtractorStemmed(max_ngram_len=ngrams)
else:
ngram_extractor = NgramExtractor(max_ngram_len=ngrams)
parse_model = SearnModelTemplateFeatures(feature_extractor=template_feature_extractor,
cost_function=cost_fn,
min_feature_freq=MIN_FEAT_FREQ,
ngram_extractor=ngram_extractor, cr_tags=cr_tags,
base_learner_fact=BASE_LEARNER_FACT,
beta=beta,
# log_fn=lambda s: print(s))
log_fn=lambda s: None)
parse_model.train(essays_TD, max_epochs=max_epochs)
num_feats = template_feature_extractor.num_features()
essay_td_ys_bycode = parse_model.get_label_data_essay_level(essays_TD)
essay_vd_ys_bycode = parse_model.get_label_data_essay_level(essays_VD)
essay_td_pred_ys_bycode = parse_model.predict_essay_level(essays_TD)
essay_vd_pred_ys_bycode = parse_model.predict_essay_level(essays_VD)
return num_feats, essay_td_ys_bycode, essay_vd_ys_bycode, essay_td_pred_ys_bycode, essay_vd_pred_ys_bycode
def model_train_predict(essays_TD, essays_VD, extractor_names, cost_function_name, ngrams, stemmed, beta):
extractors = get_functions_by_name(extractor_names, all_extractor_fns)
# get single cost function
cost_fn = get_functions_by_name([cost_function_name], all_cost_functions)[0]
assert cost_fn is not None, "Cost function look up failed"
# Ensure all extractors located
assert len(extractors) == len(extractor_names), "number of extractor functions does not match the number of names"
template_feature_extractor = NonLocalTemplateFeatureExtractor(extractors=extractors)
if stemmed:
ngram_extractor = NgramExtractorStemmed(max_ngram_len=ngrams)
else:
ngram_extractor = NgramExtractor(max_ngram_len=ngrams)
parse_model = SearnModelTemplateFeatures(feature_extractor=template_feature_extractor,
cost_function=cost_fn,
min_feature_freq=MIN_FEAT_FREQ,
ngram_extractor=ngram_extractor, cr_tags=cr_tags,
base_learner_fact=BASE_LEARNER_FACT,
beta=beta,
# log_fn=lambda s: print(s))
log_fn=lambda s: None)
parse_model.train(essays_TD, MAX_EPOCHS)
num_feats = template_feature_extractor.num_features()
sent_td_ys_bycode = parse_model.get_label_data(essays_TD)
sent_vd_ys_bycode = parse_model.get_label_data(essays_VD)
sent_td_pred_ys_bycode = parse_model.predict(essays_TD)
sent_vd_pred_ys_bycode = parse_model.predict(essays_VD)
return num_feats, sent_td_ys_bycode, sent_vd_ys_bycode, sent_td_pred_ys_bycode, sent_vd_pred_ys_bycode
for i, (essays_TD, essays_VD) in enumerate(folds):
print("\nCV % i" % i)
# parse_model = SearnModel(feat_extractor, cr_tags, base_learner_fact=LogisticRegression, beta_decay_fn=lambda beta: beta - 0.1)
# parse_model = SearnModelXgBoost(feat_extractor, cr_tags, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelCla(feat_extractor, cr_tags, base_learner_fact=CostSensitiveLogisticRegression, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags, base_learner_fact=LogisticRegression, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags,
# base_learner_fact=lambda : RandomForestClassifier(n_jobs=1, max_depth=25), beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags,
# base_learner_fact=lambda : GradientBoostingClassifier(max_depth=3, max_features="log2"),
# beta_decay_fn=lambda beta: beta - 0.3, sparse=False)
parse_model = SearnModelTemplateFeatures(feature_extractor=template_feature_extractor,
cost_function=cost_fn,
min_feature_freq=MIN_FEAT_FREQ,
ngram_extractor=ngram_extractor,
cr_tags=cr_tags,
base_learner_fact=LogisticRegression,
beta=BETA
)
parse_model.train(essays_TD, MAX_EPOCHS)
sent_td_ys_bycode = parse_model.get_label_data(essays_TD)
sent_vd_ys_bycode = parse_model.get_label_data(essays_VD)
sent_td_pred_ys_bycode = parse_model.predict(essays_TD)
sent_vd_pred_ys_bycode = parse_model.predict(essays_VD)
merge_dictionaries(sent_td_ys_bycode, cv_sent_td_ys_by_tag)
merge_dictionaries(sent_vd_ys_bycode, cv_sent_vd_ys_by_tag)
merge_dictionaries(sent_td_pred_ys_bycode, cv_sent_td_predictions_by_tag)
merge_dictionaries(sent_vd_pred_ys_bycode, cv_sent_vd_predictions_by_tag)
for i, (essays_TD, essays_VD) in enumerate(folds):
print("\nCV % i" % i)
# parse_model = SearnModel(feat_extractor, cr_tags, base_learner_fact=LogisticRegression, beta_decay_fn=lambda beta: beta - 0.1)
# parse_model = SearnModelXgBoost(feat_extractor, cr_tags, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelCla(feat_extractor, cr_tags, base_learner_fact=CostSensitiveLogisticRegression, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags, base_learner_fact=LogisticRegression, beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags,
# base_learner_fact=lambda : RandomForestClassifier(n_jobs=1, max_depth=25), beta_decay_fn=lambda beta: beta - 0.3)
# parse_model = SearnModelSklearnWeighted(feat_extractor, cr_tags,
# base_learner_fact=lambda : GradientBoostingClassifier(max_depth=3, max_features="log2"),
# beta_decay_fn=lambda beta: beta - 0.3, sparse=False)
parse_model = SearnModelTemplateFeatures(
feature_extractor=template_feature_extractor,
cost_function=cost_fn,
min_feature_freq=MIN_FEAT_FREQ,
ngram_extractor=ngram_extractor,
cr_tags=cr_tags,
base_learner_fact=LogisticRegression,
beta=BETA)
parse_model.train(essays_TD, MAX_EPOCHS)
sent_td_ys_bycode = parse_model.get_label_data(essays_TD)
sent_vd_ys_bycode = parse_model.get_label_data(essays_VD)
sent_td_pred_ys_bycode = parse_model.predict(essays_TD)
sent_vd_pred_ys_bycode = parse_model.predict(essays_VD)
merge_dictionaries(sent_td_ys_bycode, cv_sent_td_ys_by_tag)
merge_dictionaries(sent_vd_ys_bycode, cv_sent_vd_ys_by_tag)
merge_dictionaries(sent_td_pred_ys_bycode, cv_sent_td_predictions_by_tag)
merge_dictionaries(sent_vd_pred_ys_bycode, cv_sent_vd_predictions_by_tag)