def run(analyzer, ngrams, lowercase, stop_words, max_df, min_df, norm, use_idf,
smooth_idf, sublinear_tf):
def train_model(X_train, y_train, stat_model):
#stat_model.optimize_hyperparameters(X_train, y_train, folds=5)
# params = stat_model.best_params
params = {'kernel': 'rbf', 'C': 10, 'probability': True, 'gamma': 0.1}
#params = {'kernel': 'rbf', 'C': 1, 'probability': True, 'gamma': 0.001} #with scale
#params = {'alpha': alpha}
#print params
stat_model.train(X_train, y_train, params)
def test_model(stat_model, X_test, y_test, fscore_list):
y_true, y_pred = y_test, stat_model.predict(X_test)
fscore_list.append(accuracy_score(y_true, y_pred))
print(fscore_list[-1])
# load data
# data
data = News()
#data = Spam()
data_train_x, data_train_y = data.get_train()
data_test_x, data_test_y = data.get_test()
#print newsgroups_train
featurizer = TFIDF(analyzer, ngrams, lowercase, stop_words, max_df, min_df,
norm, use_idf, smooth_idf, sublinear_tf)
(x, y, X_test, y_test) = featurizer.featurize(data_train_x, data_train_y,
data_test_x, data_test_y)
#incremental training
stat_model = SVC_Model()
#stat_model = NaiveBayes()
X_train = None
y_train = []
accuracy_list = [0]
start_time = time.time()
#print x.shape[0]
# first model
train_model(x, y, stat_model)
test_model(stat_model, X_test, y_test, accuracy_list)
#print featurizer.pipeline.get_params()
print("--- %s seconds ---" % (time.time() - start_time))
return stat_model.model.score(X_test, y_test)
y_true, y_pred = y_test, stat_model.predict(X_test)
fscore_list.append(accuracy_score(y_true, y_pred))
print(fscore_list[-1])
# data
data = News()
#data = Spam()
data_train_x, data_train_y = data.get_train()
data_test_x, data_test_y = data.get_test()
random_seed = 42
# specify TF-IDF
featurizer = TFIDF()
(x, y, X_test, y_test) = featurizer.featurize(data_train_x, data_train_y,
data_test_x, data_test_y)
print x.shape[0]
'''
y_encoder = LabelBinarizer()
y_encoder.fit(y)
y_one_hot = y_encoder.transform(y)
print y_one_hot.shape
print x.shape
all_data = hstack((x, y_one_hot))
print all_data.shape
'''
use_idf true
'''
#print newsgroups_train
analyzer = 'char_wb'
lowercase = True
max_df = 1.0
min_df = 0.01
ngrams = 2
norm = "l2"
smooth_idf = True
stop_words = None
sublinear_tf = False
use_idf = True
featurizer = TFIDF(analyzer, ngrams, lowercase, stop_words, max_df, min_df,
norm, use_idf, smooth_idf, sublinear_tf)
(x, y, X_test, y_test) = featurizer.featurize(data_train_x, data_train_y,
data_test_x, data_test_y)
all_list = []
time_list_all = []
sum_uncertainty_all = []
#incremental training
for run in range(1):
stat_model = SVC_Model()
random_seed += 1