# -*- coding: utf-8 -*-
"""
Created on Fri Oct 23 15:11:39 2015

@author: Camil Staps, s4498062

This is Python 2 code.
"""

import matplotlib.pyplot as plt
from scipy import io as sciio
from sklearn import tree
from sklearn import cross_validation
import numpy as np

# 3.2.1
wine = sciio.loadmat('./Data/wine.mat')
data = wine['X']
clss = wine['y']
classNames = [str(n[0][0]) for n in wine['classNames']]

X_train, X_test, y_train, y_test = cross_validation.train_test_split(data, clss)

depths = range(2,21)
optimal_depth, max_score, scores = 0, 0, []
for depth in depths:
    clf = tree.DecisionTreeClassifier(max_depth=depth, criterion='gini')
    clf = clf.fit(X_train, y_train)
    score = clf.score(X_test, y_test)
    scores.append(score)
    if score > max_score:
        max_score, optimal_depth = score, depth

print(optimal_depth, max_score)
plt.plot(depths, scores, label='Holdout CV')

# 3.2.2
k = 10
depths = range(2,21)

optimal_depth, max_score, scores = 0, 0, []
kf = cross_validation.KFold(len(data), k)
for depth in depths:
    temp_scores = []
    for train, test in kf:
        X_train, X_test = [data[i] for i in train], [data[i] for i in test]
        y_train, y_test = [clss[i] for i in train], [clss[i] for i in test]

        clf = tree.DecisionTreeClassifier(max_depth=depth, criterion='gini')
        clf = clf.fit(X_train, y_train)
        score = clf.score(X_test, y_test)
        temp_scores.append(score)

    score = np.mean(temp_scores)
    scores.append(score)
    if score > max_score:
        max_score, optimal_depth = score, depth

print(optimal_depth, max_score)
plt.plot(depths, scores, label=str(k) + '-fold CV')

plt.ylabel('Classification error')
plt.xlabel('Tree depth')
plt.legend(loc=4)
plt.grid()
plt.show()