import numpy as np
import pandas as pd
from sklearn.linear_model import SGDClassifier
from sklearn.grid_search import GridSearchCV
from sklearn.cross_validation import cross_val_score
import sys

# check for valid sys.argv (must be 2 files)
if len(sys.argv) < 3:
    sys.stdout.write("You must set train and test files as arguments!")
    sys.exit(0)

# read train data
with open(sys.argv[1], 'r') as train_file:
    train_df = pd.read_csv(train_file, sep='\t')
features_cols = ['x' + str(i).zfill(3) for i in range(30)]
X_train = np.array(train_df[features_cols]) #features
y_train = np.array(train_df['y'])

# build a model
n_samples = len(train_df.index)
n_iter = np.ceil((10**6)/n_samples)
clf = SGDClassifier(loss='log', penalty='none', fit_intercept=False, n_iter=n_iter)

#clf.fit(X, Y)

# TODO:
# Параметры подбирай через GridSearch, scoring function — f1_weigted,
# проверку делай через cross_val_score с cv=10.

param_grid = {
    #'alpha': [0.001, 0.0001, 0.00001, 0.000001],
}
grid = GridSearchCV(clf, param_grid, cv=10, scoring='f1_weighted')
print(grid)
grid.fit(X_train, y_train)

best_params = grid.best_estimator_.get_params()
print('best params:', best_params)

# read test data
with open(sys.argv[2], 'r') as test_file:
    test_df = pd.read_csv(test_file, sep='\t')
X_test = test_df[features_cols]
y_test = np.array(test_df['y'])

predict_Y = grid.predict(X_test)
print('test predicted:', predict_Y)
print('test real:', y_test)
error = np.mean(predict_Y != y_test)
print('test error:', error)

predict_train_Y = grid.predict(X_train)
print('train predicted:', predict_train_Y)
print('train real:', y_train)
train_error = np.mean(predict_train_Y != y_train)
print('train error:', train_error)

clf = SGDClassifier(**best_params)
scores = cross_val_score(clf, X_train, y_train, cv=10)
print('Best %s: %0.3f (+/- %0.2f)' % \
    ('f1_weighted', scores.mean(), scores.std() / 2))