#!/usr/bin/env python
import numpy as np
import pandas as pd

from common import describe_data, test_env

from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import confusion_matrix
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.naive_bayes import MultinomialNB
from sklearn.ensemble import RandomForestClassifier


def read_data(file):
    """Return pandas dataFrame read from csv file"""
    try:
        return pd.read_excel(file)
    except FileNotFoundError:
        exit('ERROR: ' + file + ' not found')


def preprocess_data(df, verbose=False):
    y_column = 'In university after 4 semesters'

    # Features can be excluded by adding column name to list
    drop_columns = []

    categorical_columns = [
        'Faculty',
        'Paid tuition',
        'Study load',
        'Previous school level',
        'Previous school study language',
        'Recognition',
        'Study language',
        'Foreign student'
    ]

    # Handle dependent variable
    if verbose:
        print('Missing y values: ', df[y_column].isna().sum())

    y = df[y_column].values
    # Encode y. Naive solution
    y = np.where(y == 'No', 0, y)
    y = np.where(y == 'Yes', 1, y)
    y = y.astype(float)

    # Drop also dependent variable variable column to leave only features
    drop_columns.append(y_column)
    df = df.drop(labels=drop_columns, axis=1)

    # Remove drop columns for categorical columns just in case
    categorical_columns = [
        i for i in categorical_columns if i not in drop_columns]

    # STUDENT SHALL ENCODE CATEGORICAL FEATURES
    # fillna - NA/NaN values using the specified method.
    for i in categorical_columns:
        df[i] = df[i].fillna(value='Missing')

    dum = pd.get_dummies(df, prefix_sep=":", columns=categorical_columns)
    df = dum.drop(
        columns=[
            'Paid tuition:No',
            'Foreign student:No',
            'Previous school study language:Not known',
            'Study load:Partial',
            'Faculty:School of Engineering',
            'Recognition:Missing',
            'Study language:Estonian'])
    df = df.fillna(value=0)

    # Handle missing data. At this point only exam points should be missing
    # It seems to be easier to fill whole data frame as only particular columns
    if verbose:
        describe_data.print_nan_counts(df)

    # STUDENT SHALL HANDLE MISSING VALUES

    if verbose:
        describe_data.print_nan_counts(df)

    # Return features data frame and dependent variable
    return df, y


# STUDENT SHALL CREATE FUNCTIONS FOR LOGISTIC REGRESSION CLASSIFIER, KNN
# CLASSIFIER, SVM CLASSIFIER, NAIVE BAYES CLASSIFIER, DECISION TREE
# CLASSIFIER AND RANDOM FOREST CLASSIFIER

def random_forest_classifier(train_X, train_y, test_X, test_y):
    rnd_clf = RandomForestClassifier(n_estimators=100)
    rnd_clf.fit(train_X, train_y)
    pred_y = rnd_clf.predict(test_X)
    matrix = confusion_matrix(test_y, pred_y)
    accuracy = accuracy_score(test_y, pred_y)
    print("\n# Random forest classifier test data")
    print_confusion_matrix(matrix, accuracy)


def decision_tree_classifier(train_X, train_y, test_X, test_y):
    tree_clf = DecisionTreeClassifier()
    tree_clf.fit(train_X, train_y)
    pred_y = tree_clf.predict(test_X)
    matrix = confusion_matrix(test_y, pred_y)
    accuracy = accuracy_score(test_y, pred_y)
    print("\n# Decision tree classifier test data")
    print_confusion_matrix(matrix, accuracy)


def naive_bayes_classifier(train_X, train_y, test_X, test_y):
    naive_clf = MultinomialNB()
    naive_clf.fit(train_X, train_y)
    pred_y = naive_clf.predict(test_X)
    matrix = confusion_matrix(test_y, pred_y)
    accuracy = accuracy_score(test_y, pred_y)
    print("\n# Naive bayes test data")
    print_confusion_matrix(matrix, accuracy)


def svm_classifier(train_X, train_y, test_X, test_y):
    svm_clf = SVC(kernel='sigmoid', random_state=0,
                  gamma=.01, C=1, probability=True)
    svm_clf.fit(train_X, train_y)
    pred_y = svm_clf.predict(test_X)
    matrix = confusion_matrix(test_y, pred_y)
    accuracy = accuracy_score(test_y, pred_y)
    print("\n# SVC regression test data")
    print_confusion_matrix(matrix, accuracy)


def print_confusion_matrix(matrix, accuracy):
    print("\nConfusion matrix: \n     pred:No pred:yes")

    print("true:No   " + str(matrix[0][0]) + "  " + str(matrix[0][1]))
    print("true:Yes  " + str(matrix[1][0]) + "  " + str(matrix[1][1]))

    print("\nAccuracy: " + str(accuracy * 100) + "%")


def logistic_regression_classifier(train_X, train_y, test_X, test_y):
    logreg_clf = LogisticRegression(
        solver='newton-cg',
        multi_class='multinomial')
    logreg_clf.fit(train_X, train_y)

    pred_y = logreg_clf.predict(test_X)
    matrix = confusion_matrix(test_y, pred_y)
    print("\n# Logistic regression test data")
    accuracy = accuracy_score(test_y, pred_y)
    print_confusion_matrix(matrix, accuracy)


def knn_classifier(train_X, train_y, test_X, test_y):
    knn_clf = KNeighborsClassifier(n_neighbors=5)
    knn_clf.fit(train_X, train_y)
    pred_y = knn_clf.predict(test_X)

    print("\n# KNN test data")
    matrix = confusion_matrix(test_y, pred_y)
    accuracy = accuracy_score(test_y, pred_y)
    print_confusion_matrix(matrix, accuracy)


if __name__ == '__main__':
    modules = ['numpy', 'pandas', 'sklearn']
    test_env.versions(modules)

    students = read_data('data/students.xlsx')

    # STUDENT SHALL CALL PRINT_OVERVIEW AND PRINT_CATEGORICAL FUNCTIONS WITH
    # FILE NAME AS ARGUMENT
    describe_data.print_overview(
        students, file='results/students_overview.txt')
    describe_data.print_categorical(
        students, file='results/students_categorical_data.txt')

    students_X, students_y = preprocess_data(students)

    # data for methods
    # test size 0.25 as default will be used.

    train_X, test_X, train_y, test_y = train_test_split(
        students_X, students_y, test_size=None, train_size=None, random_state=60)

    # STUDENT SHALL CALL CREATED CLASSIFIERS FUNCTIONS
    logistic_regression_classifier(train_X, train_y, test_X, test_y)
    knn_classifier(train_X, train_y, test_X, test_y)
    svm_classifier(train_X, train_y, test_X, test_y)
    naive_bayes_classifier(train_X, train_y, test_X, test_y)
    decision_tree_classifier(train_X, train_y, test_X, test_y)

    print('Done')