lamorgia_classifier.py

import datetime
import pandas as pd
from sklearn.ensemble import RandomForestClassifier, AdaBoostClassifier
from sklearn.metrics import f1_score, recall_score, precision_score
from sklearn.model_selection import StratifiedKFold, cross_val_predict, train_test_split
from sklearn.tree import DecisionTreeClassifier


def classifier(time_freq):

    computed_data = pd.read_csv('labeled_features/features_' + time_freq + '.csv.gz', parse_dates=['date'])

    features = ['std_rush_order',
                'avg_rush_order',
                'std_trades',
                'std_volume',
                'avg_volume',
                'std_price',
                'avg_price',
                'avg_price_max',
                'hour_sin',
                'hour_cos',
                'minute_sin',
                'minute_cos']

    X = computed_data[features]
    Y = computed_data['gt'].astype(int).values.ravel()

    clf = RandomForestClassifier(n_estimators=200, max_depth=5, random_state=1)
    #clf = AdaBoostClassifier(DecisionTreeClassifier(max_depth=5), n_estimators=100, learning_rate=0.5, random_state=0)

    cv_list = [5]
    processes = 7

    for n_fold in cv_list:
        #print('Processing: {} folds - time freq {}'.format(n_fold, time_freq))
        #y_pred = cross_val_predict(clf, X, Y.ravel(), cv=StratifiedKFold(n_splits=n_fold), n_jobs=processes)
        print(f'Fitting model for {time_freq}.')
        X_train, X_test, Y_train, Y_test = train_test_split(X, Y.ravel(), shuffle=False, train_size=0.8)
        clf.fit(X_train, Y_train)
        y_pred = clf.predict(X_test)
        print('Recall: {}'.format(recall_score(Y_test, y_pred)))
        print('Precision: {}'.format(precision_score(Y_test, y_pred)))
        print('F1 score: {}'.format(f1_score(Y_test, y_pred)))




if __name__ == '__main__':
    start = datetime.datetime.now()
    classifier(time_freq='25S')
    classifier(time_freq='15S')
    classifier(time_freq='5S')
    print(datetime.datetime.now() - start)