def check_email():
# set script to run at intervals (time.sleep?) - morn, mid-day, afternoon
current_time = datetime.now()
previous_check = cpm.unpickle(os.path.join(pkl_dir, 'last_check_time.pkl'))
print previous_check
emails = cpm.get_emails('INBOX', previous_check)
if not emails: # avoid unpickling if empty inbox
return
feature_model = cpm.unpickle(os.path.join(pkl_dir, 'final_vec.pkl'))
classifier_model = cpm.unpickle(os.path.join(pkl_dir, 'final_model.pkl'))
print 'classifier model', classifier_model
for email in emails:
if email.sent_at > previous_check:
clean_b = cpm.clean_raw_txt(email.body)
clean_s = cpm.clean_raw_txt(email.subject)
print 'clean_email', clean_s, clean_b
email_features = feature_model.transform([clean_b+clean_s])
print 'email_bag', email_features.shape
classifier_result = classifier_model.predict(email_features)
eval_email(classifier_result, email)
cpm.pickle(current_time, os.path.join(pkl_dir,'last_check_time.pkl'))
def main(model_name, X_test=None, y_test=None, models=None, save=False, train_fn='train_split.pkl', model_fn='final_model.pkl'):
#model_names = ['LogisticRegression', 'MultinomialNB', 'SVC', 'RandomForest', 'GradientBoost']
if models is None:
models = [cpm.unpickle(os.path.join(pkl_dir, model_fn))]
if X_test is None:
X_train, X_test, y_train, y_test = cpm.unpickle(os.path.join(pkl_dir, train_fn))
X_test = X_test.todense()
for i, model in enumerate(models):
model_eval(model_name[i], model, X_test, y_test, save)
plot_roc_curve(model_name[:len(models)], models, X_test, y_test, save)
def main(model_names, save=False, train_fn='train_split.pkl', model_fn='final_model.pkl', new_data_split=False, random=11):
if new_data_split:
X_train, X_test, y_train, y_test = cpm.create_datasets(save=save, vectorize=True, random=random)
else:
X_train, X_test, y_train, y_test = cpm.unpickle(os.path.join(pkl_dir, train_fn))
X_train = X_train.todense()
#model_names = ['LogisticRegression', 'MultinomialNB', 'SVC', 'RandomForest', 'GradientBoost']
class_model = {
'LogisticRegression': LogisticRegression(C=10000.0 , penalty='l2', class_weight='auto', fit_intercept=True),
'MultinomialNB': MultinomialNB(alpha=0.100000, fit_prior=True),
'SVC': SVC(C=10000.0, kernel='linear', shrinking=True, probability=True, class_weight='auto'),
'RandomForest': RandomForestClassifier(n_estimators=121, criterion='entropy', bootstrap=True, oob_score=True, max_features='auto'),
'GradientBoost': GradientBoostingClassifier()
}
model_results = []
for model in model_names:
model_results.append(setup_model(model, class_model[model], X_train, y_train, save))
return model_results
def get_feature_names(vectorizer=None, vec_fn='final_vec.pkl'):
if vectorizer == None:
vectorizer = cpm.unpickle(os.path.join(pkl_dir, vec_fn))
return vectorizer.get_feature_names()