import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.utils import shuffle
import seaborn as sns
from scipy.stats import norm


from scipy.stats.distributions import *
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import log_loss
from sklearn.model_selection import RandomizedSearchCV


def get_data():
    train = pd.read_csv('data/raw/train.csv')
    test = pd.read_csv('data/raw/test.csv')
    train = shuffle(train)


    # #train
    # train.SibSp = train.SibSp.map({0: 7, 1: 9, 2: 8, 3: 6, 4: 5, 5: 4, 6: 3, 7: 2, 8: 4, 9: 0})
    # train.SibSp -= 4
    # print(train.corr(method='pearson')['SibSp'])

    target_data = train['Survived'].to_numpy()
    target_data = np.expand_dims(target_data, axis=1)


    # sns.catplot(x='Cabin', y='Survived', data=train
    #train['SibSp'] = train['SibSp'].map({0: 2, 1: 0, 2: 1, 3: 3, 4: 4})

    train["Age"] = train.groupby("Pclass")["Age"].transform(lambda x: x.fillna(x.median()))
    test["Age"] = test.groupby("Pclass")["Age"].transform(lambda x: x.fillna(x.median()))
    test["Fare"] = train.groupby("Pclass")["Fare"].transform(lambda x: x.fillna(x.median()))

    train.drop(columns='Survived', inplace=True)

    data = pd.concat([train, test])

    data["Name"] = data["Name"].transform(lambda x: x.split(', ')[1].split()[0])

    data['Name'] = data['Name'].replace('the', 'Mrs.')
    data['Name'] = data['Name'].replace('Lady.', 'Mrs.')
    data['Name'] = data['Name'].replace('Mme.', 'Mrs.')
    data['Name'] = data['Name'].replace('Dona.', 'Mrs.')
    data['Name'] = data['Name'].replace('Ms.', 'Miss.')
    data['Name'] = data['Name'].replace('Don.', 'Mr.')
    data['Name'] = data['Name'].replace('Sir.', 'Mr.')
    data['Name'] = data['Name'].replace('Capt.', 'Rev.')
    data['Name'] = data['Name'].replace('Major.', 'Pers.')
    data['Name'] = data['Name'].replace('Col.', 'Pers.')
    data['Name'] = data['Name'].replace('Dr.', 'Pers.')
    data['Name'] = data['Name'].replace('Jonkheer.', 'Rev.')
    data['Name'] = data['Name'].replace('Mlle.', 'Miss.')

    data['Embarked'] = data['Embarked'].fillna('S')

    # data['Alone'] = ~((data['SibSp'] > 0) | (data['Parch'] > 0))
    # data['SibSp'] = data['SibSp'].map({0: 2, 1: 0, 2: 1, 3: 3, 4: 4, 5: 5, 6: 6, 7: 7, 8: 8, 9: 9})
    # data['Par'] = ((data['Age'] < 16) * data['Parch']).map({0: 0, 1: 2, 2: 1})
    # data['Ch'] = ((data['Age'] > 16) * data['Parch']).map({4: 0, 6: 0, 5: 1, 0: 2, 1: 3, 2: 4, 3: 5})
    #
    # data['NoCabin'] = (~data['Cabin'].isna()).astype(int)

    data = pd.get_dummies(data, columns=["Sex", "Name", "Embarked"], drop_first=True)
    data.drop(columns=['Ticket', 'Cabin'], inplace=True)

    #  BOX COX TRANSFORMATION

    from scipy.special import boxcox
    from scipy.stats import boxcox_normmax
    data['Fare'] -= data['Fare'].min() - 1.
    data['Age'] -= data['Age'].min() - 1.

    data['Fare'] = boxcox(data['Fare'], boxcox_normmax(data['Fare']))
    data['Age'] = boxcox(data['Age'], boxcox_normmax(data['Age']))

    train = data[0:891].copy()
    test = data[891:1310].copy()

    train = train - train.mean(axis=0)
    train = train / train.std(axis=0)
    test = test - test.mean(axis=0)
    test = test / test.std(axis=0)

    train_data = train.to_numpy()
    test_data = test.to_numpy()

    print(train_data.shape)
    return train_data, target_data.ravel(), test_data


def main():
    dtrain, dtarget, dtask = get_data()

    all_train = dtrain
    dtest = dtrain[791:891]
    dtrain = dtrain[0:791]

    all_target = dtarget
    dtestT = dtarget[791:891]
    dtarget = dtarget[0:791]

    params = {
        "n_estimators": rv_discrete(a=1000, b=3000),
        "max_depth": rv_discrete(a=1, b=5),
        "min_samples_split": rv_continuous(a=0.2, b=0.8),
        "min_samples_leaf": rv_continuous(a=0.2, b=0.8),
        "max_features": "string"
    }

    # n_estimators: 1000
    # min_samples_split: 0.008
    # min_samples_leaf: 0.0005
    # max_features: sqrt
    # max_depth: None
    # bootstrap: True
    # 84 / 100
    #
    # n_estimators: 1050
    # min_samples_split: 0.01
    # min_samples_leaf: 0.0008
    # max_features: auto
    # max_depth: None
    # bootstrap: True
    # 86 / 100
    #
    # n_estimators: 1050
    # min_samples_split: 0.01105
    # min_samples_leaf: 0.0008
    # max_features: sqrt
    # max_depth: None
    # bootstrap: True
    # 84 / 100
    #
    # n_estimators: 1100
    # min_samples_split: 0.011
    # min_samples_leaf: 0.00075
    # max_features: auto
    # max_depth: None
    # bootstrap: True
    # # 81 / 100

    # params = {'bootstrap': [True],
    #           'max_depth': [None],
    #           'max_features': ['auto', 'sqrt'],
    #           'min_samples_leaf': [0.0003, 0.0004, 0.0005, 0.0007, 0.00075, 0.0008, 0.00085, 0.0009],
    #           'min_samples_split': [0.01, 0.0105, 0.1107, 0.011, 0.01103, 0.01105, 0.01107, 0.012],
    #           'n_estimators': [950, 1000, 1050, 1075, 1100]
    #           }

    rfc = RandomForestClassifier(n_estimators=1050, min_samples_split=0.01105, min_samples_leaf=0.0008,
                                 max_features='sqrt', max_depth=None, bootstrap=True)

    # rscv = RandomizedSearchCV(rfc, param_distributions=params, verbose=2, n_jobs=10, n_iter=700)
    # rscv.fit(dtrain, dtarget)

    # for k, v in rscv.best_params_.items():
    #     print(k, ': ', v, sep='')
    rfc.fit(dtrain, dtarget)
    out = rfc.predict(dtest)

    print(len(out) - np.abs(out - dtestT).sum(), '/', len(out))

    # rfc.set_params(**rfc.best_params_)
    rfc.fit(all_train, all_target)
    answer = rfc.predict(dtask)

    indx = [i for i in range(892, 1310, 1)]
    df = pd.DataFrame.from_dict({'PassengerId': indx, 'Survived': answer})
    df.to_csv('data/out.csv', index=False)


main()