def main():
X, y, train_X, train_y, test_X, test_y = ml_utils.load_data(my_dir, 0)
estimator = RandomForestClassifier(n_estimators=40 , max_depth=5)
estimator.fit(X, y)
keys_model = model.Model(estimator, my_dir)
keys_model.save(os.path.join(my_dir, "keys_model.pkl"))
def main(data, train, test=None):
classes_to_score = [0, 1, 2, 3]
train_text, train_labels, score_text, score_ids = load_data(
data, train, test, classes_to_score, scoring=True)
train_text, train_labels = fix_data(train_text, train_labels)
score(train_text, train_labels, score_text, score_ids)
def main(data, train, test=None):
classes_to_score = [0, 1, 2, 3, 4]
train_text, train_labels, score_text, score_ids, timestamps = load_data(
data, train, test, id_key='__index__', text_key='text', scoring=True)
train_text, train_labels = fix_data(train_text, train_labels)
score(train_text, train_labels, score_text, score_ids, timestamps,
classes_to_score)
def main(data, train_data, test_data):
train_text, train_labels, test_text, test_labels = load_data(
data,
train_data,
test_data,
text_key='article',
id_key=['source', 'source_index'])
analyze(train_text, train_labels, test_text, test_labels, threshold=.6)
def main(data, train_data, test_data):
train_text, train_labels, test_text, test_labels = load_data(
data,
train_data,
test_data,
text_key='article',
id_key=['source', 'source_index'])
train(train_text, train_labels, test_text, test_labels)
def main(data, train_data, test_data):
classes_to_analyze = [0, 1, 2, 3, 4]
train_text, train_labels, test_text, test_labels = load_data(
data, train_data, test_data, id_key='__index__', text_key='text')
train_text, train_labels = fix_data(train_text, train_labels)
test_text, test_labels = fix_data(test_text, test_labels)
analyze(train_text, train_labels, test_text, test_labels,
classes_to_analyze)
def main(data, train_data, test_data):
CLASSES_TO_TRAIN = [0, 1, 2, 3, 4]
# load the data
train_text, train_labels, test_text, test_labels = load_data(
data, train_data, test_data, id_key='__index__', text_key='text')
# fix the data
train_text, train_labels = fix_data(train_text, train_labels)
test_text, test_labels = fix_data(test_text, test_labels)
# train on the data
train(train_text, train_labels, test_text, test_labels, CLASSES_TO_TRAIN)
def main():
if len(sys.argv) < 2:
print("Usage: {} <task directory>".format(sys.argv[0]))
return
task_dir = sys.argv[1]
X, y, train_X, train_y, test_X, test_y = ml_utils.load_data(task_dir, TEST_FRACTION)
curr_results_dir = os.path.join(task_dir, "results")
# # Boosted Decision Tree
# num_estimators_param = Parameter("Max Number of Estimators", "n_estimators", range(10, 111, 10))
# base_estimators = [DecisionTreeClassifier(max_depth=d) for d in range(2, 6)]
# estimator_depth_param = Parameter("Max Tree Depth", "base_estimator", base_estimators, range(1, 6))
# boosted_algo = Algorithm("Boosted Decision Tree", [num_estimators_param, estimator_depth_param], AdaBoostClassifier())
# analyze_algorithm(boosted_algo, curr_results_dir, train_X, train_y, X, y)
# print()
# Random Forest
num_estimators_param = Parameter("Number of Estimators", "n_estimators", range(1, 121, 10))
max_depth = Parameter("Max Tree Depth", "max_depth", range(1, 7))
rf_algo = Algorithm("Random Forest", [num_estimators_param, max_depth], RandomForestClassifier())
analyze_algorithm(rf_algo, curr_results_dir, train_X, train_y, X, y)
print()
# Naiive Bayes
bayes_algo = Algorithm("Naiive Bayes", [], GaussianNB())
analyze_algorithm(bayes_algo, curr_results_dir, train_X, train_y, X, y)
print()
# Support Vector Machine (Linear)
c_param = Parameter("C Value", "C", [1e-2, 1e-1, 1e0, 1e1, 1e2], log_scale=True)
linear_svm_algo = Algorithm("Linear Support Vector Machine", [c_param], svm.SVC(kernel="linear"))
analyze_algorithm(linear_svm_algo, curr_results_dir, train_X, train_y, X, y)
print()
# Linear Discriminant Analysis
lda_algo = Algorithm("Linear Discriminant Analysis", [], LinearDiscriminantAnalysis())
analyze_algorithm(lda_algo, curr_results_dir, train_X, train_y, X, y)
print()
# Neural Network
hidden_layer_param = Parameter("Number of Hidden Layers", "hidden_layer_sizes", vary_num_hidden_layers(30, 5), range(1, 6))
alpha_param = Parameter("Regularization Strength (alpha)", "alpha", [1e0, 1e-2, 1e-4, 1e-6, 1e-8], log_scale=True)
nn_algo = Algorithm("Neural Network", [hidden_layer_param, alpha_param], MLPClassifier(max_iter=5000))
analyze_algorithm(nn_algo, curr_results_dir, train_X, train_y, X, y)
print()
def train(model_class, domain=None):
cache_path = './cache'
if os.path.exists(
os.path.join(cache_path,
"{}_boosted_tree.pkl".format(target_year))):
df = pickle.load(
open(
os.path.join(cache_path,
"{}_boosted_tree.pkl".format(target_year)), 'rb'))
else:
df = load_data(target_year)
pickle.dump(
df,
open(
os.path.join(cache_path,
"{}_boosted_tree.pkl".format(target_year)), 'wb'))
if domain is not None:
print("Find important feature in {}".format(domain))
domain2category = pickle.load(open("domain2category.pkl", 'rb'))
df = df.loc[df['main_category'].isin(domain2category[domain])]
selected_feature = const_selected_feature
X = df[selected_feature].values
bde = joblib.load('category_encoder.pkl')
# bde = ce.BinaryEncoder(cols=categorical_feature, return_df=False)
venue_embedding = bde.transform(df[['venue']]).values
# pca = PCA(n_components=32)
# venue_embedding = pca.fit_transform(venue_embedding)
X = np.hstack((X, venue_embedding))
venue_embedding_size = venue_embedding.shape[1]
print(venue_embedding_size)
for i in range(venue_embedding_size):
selected_feature.append('venue_' + str(i))
nlp_pipeline = joblib.load('nlp_pipeline.pkl')
embeddings = nlp_pipeline.transform(df['summary'].values)
embeddings_size = embeddings.shape[1]
print(embeddings_size)
for i in range(embeddings_size):
selected_feature.append('title_' + str(i))
X = np.hstack((X, embeddings))
y = df['citationCount'].values
y = np.clip(y, 0, 23)
kf = KFold(n_splits=5)
accuracies = []
print(model_class)
importance_stats = {}
for train_index, test_index in tqdm(kf.split(y)):
model = model_class(verbose=0)
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
model.fit(X_train, y_train)
y_pred = model.predict(X_test)
mse_score = mse(y_pred, y_test)
accuracies.append(mse_score)
feature_importance = model.get_feature_importance(prettified=True)
for feature_id, score in feature_importance:
if selected_feature[int(feature_id)] not in importance_stats:
importance_stats[selected_feature[int(feature_id)]] = score
else:
importance_stats[selected_feature[int(feature_id)]] += score
limit = 0
for key, value in importance_stats.items():
print(key, " score: ", value)
limit += 1
if limit >= 30:
break
print("MSE: {} ({})".format(np.mean(accuracies), np.std(accuracies)))
print("Sample : ")
print(y_pred[:10])
print(y_test[:10])