('clf', LogisticRegression())
# ('clf', RandomForestClassifier())
# ('clf', GradientBoostingClassifier())
# ('clf', RandomForestClassifier())
])
params = {
'clf__penalty': ('l1', 'l2'), # Logistic
'clf__C': (10, 1, 0.1, 0.01, 0.001),
# 'clf__n_neighbors': (3, 5, 7, 9, 10, 15, 20, 50, 100),
# 'clf__leaf_size': (10, 20, 30, 50, 100),
# 'clf__p': (2, 3, 5),
# 'clf__kernel': ('rbf', 'linear'), # SVM
# 'clf__gamma': (0.1, 0.01, 0.001, 0.0001), # SVM
# 'clf__p': (1, 2), # 1: mahnatan, 2: eucledian # k-NN
# 'clf__n_neighbors': (3, 4, 5, 6, 7, 8), # k-NN
# 'clf__learning_rate': (0.1, 0.01, 0.001), # Gradient Boosting
# 'clf__n_estimators': (100, 300, 600), # Gradient Boosting, Random Forest
# 'clf__alpha': (0.5, 1.0), # MultinomialNB
# 'clf__max_depth': [2, 5, None], # Random Forest
}
x_train_validation = all_features[all_features['Article'].isin(
train_ids + validation_ids)].set_index('Article')
grid_results = run_grid_search(X=x_train_validation,
y=dl_obj.y_train_validation,
pipeline=vect_based_pipeline,
parameters=params,
scoring='accuracy')
# ('clf', KNeighborsClassifier())
# ('clf', GradientBoostingClassifier())
# ('clf', RandomForestClassifier())
],
memory=memory)
params = {
'embedding_feat__embedding_type': ['tfidf', 'tf'], # embedding
'embedding_feat__embedding_dimensions': [
50,
], # embedding 100, 200, 300
'clf__penalty': ('l1', 'l2'), # Logistic
# 'clf__kernel': ('rbf', 'linear'), # SVM
# 'clf__gamma': (0.1, 0.01, 0.001, 0.0001), # SVM
# 'clf__p': (1, 2), # 1: mahnatan, 2: eucledian # k-NN
# 'clf__n_neighbors': (3, 4, 5, 6, 7, 8), # k-NN
# 'clf__learning_rate': (0.1, 0.01, 0.001), # Gradient Boosting
# 'clf__n_estimators': (100, 300, 600), # Gradient Boosting, Random Forest
# 'clf__alpha': (0.5, 1.0), # MultinomialNB
# 'clf__max_depth': [10, 50, 100, None], # Random Forest
}
grid_results = run_grid_search(X=X_train,
y=y_train,
pipeline=final_pipeline,
parameters=params,
scoring='accuracy')
# Delete the temporary cache before exiting
rmtree(cachedir)
])
params = {
'union__title__tfidf__ngram_range': [
(1, 1),
],
'union__abstract_bow__tfidf__ngram_range': [
(1, 1),
],
'union__abstract_bow__best__n_components': [50, 100, 150],
'clf__penalty': ('l1', 'l2') # Logistic
}
grid_search_obj = run_grid_search(X=train_data['csv_df'],
y=train_data['labels'],
pipeline=pipeline,
parameters=params,
scoring='accuracy')
y_pred = grid_search_obj.predict_proba(test_data['csv_df'])
# Write predictions to a file
with open(os.path.join(DATA_DIR, 'sample_submission_bow.csv'),
'w') as csvfile:
writer = csv.writer(csvfile, delimiter=',')
lst = grid_search_obj.classes_.tolist()
lst.insert(0, "Article")
writer.writerow(lst)
for i, test_id in enumerate(test_ids):
lst = y_pred[i, :].tolist()