import numpy as np
import pandas as pd
from sklearn.utils import shuffle
from sklearn.model_selection import GridSearchCV
from sklearn.decomposition import PCA
from sklearn.preprocessing import LabelEncoder , StandardScaler
from sklearn.metrics import brier_score_loss, make_scorer
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import RandomForestClassifier
import datetime as datetime
from sklearn.svm import SVC


brier = make_scorer(brier_score_loss, greater_is_better = False , needs_proba = True)



def test(year):
    print("-------------------------------------------------------")
    print(year)
    print(datetime.datetime.now())
    print()
    print()
    full_training_df = pd.read_csv("yty_train_" + str(year) + ".csv")
    full_testing_df = pd.read_csv("yty_test_" + str(year) + ".csv")
    X_test = full_testing_df.drop("y_test" , axis = 1)
    y_test = full_testing_df["y_test"]

    X_train = full_training_df.drop("y_test" , axis = 1)
    y_train = full_training_df["y_test"]


    X_train = X_train[X_train.columns & X_test.columns]
    X_test = X_test[X_train.columns & X_test.columns ]


    rf_clf = RandomForestClassifier(n_estimators = 5000 , max_depth = 35 ,
                                         n_jobs = -1)


    t1 = datetime.datetime.now()
    rf_clf.fit(X_train , y_train)
    t2 = datetime.datetime.now()
    print("time to fit:" , t2-t1)


    t1 = datetime.datetime.now()
    predicted_probs = pd.DataFrame(rf_clf.predic_proba(X_test_full)[:,1])
    t2 = datetime.datetime.now()
    print("time to test:" , t2-t1)


    print("test score:" , brier_score_loss(y_test , predicted_probs))


    predicted_probs.columns = ["predicted_prob"]


    y_test.columns = ["true_prob"]



    joined_df = pd.concat([predicted_probs , y_test], axis=1, sort=False)

    joined_df["adjusted"] = 0

    N_plus = joined["true_prob"].sum()
    N_minus = joined.shape[0]-N_plus
    beta = N_plus / N_minus
    joined_df.adjusted = (beta*joined.predicted_probs)/(beta*joined.predicted_probs-joined.predicted_probs+1)


    joined_df.to_csv(str(year) + "results_full_rf.csv" , index = False)


test(15)
test(16)
test(17)