# -*- coding: utf-8 -*-
"""
Created on Wed Nov 15 08:18:35 2017

@author: Miguel Almeida
"""
import numpy as np
from sklearn.tree import DecisionTreeClassifier
import matplotlib.pyplot as plt

def computeError(hyp, hyps_ws, features, classes):
    net_pred = np.zeros(size)
    preds = 0
    for ix in range(len(hyps)):
        pred_n = hyps[ix].predict(features)
        preds = preds+pred_n*hyps_ws[ix]
    
    net_pred[preds<0]=-1
    net_pred[preds>=0]=1
    errors = np.sum((net_pred != classes).astype(int))
    return errors


matrix = np.loadtxt("T6-data.txt", delimiter ="\t")

    #shuffle
np.random.shuffle(matrix)

    #normalizar os dados
means = np.mean(matrix[:,:-1], axis=0)
devs = np.std(matrix[:,:-1], axis=0)
matrix[:,:2] = (matrix[:,:-1]-means)/devs

    #split data do train
classes = matrix[:,-1]*2-1
features = matrix[:,:-1]

weigths = np.ones([features.shape[0]])
size = features.shape[0]
hyps = []
hyps_ws = []

for ix in range(size):
    weigths[ix]= 1/float(size)
    
max_hyp= 20
nerrs = []
    #Computar o erro
for ix in range(max_hyp):
    stump = DecisionTreeClassifier(max_depth=1)
    stump.fit(features,classes,sample_weight=weigths)
    pred = stump.predict(features)
    errs = (pred != classes).astype(int)
    err = np.sum(errs*weigths)
    alpha = np.log((1-err)/err)
    weigths = weigths*np.exp(alpha*errs)
    weigths = weigths/np.sum(weigths)
    hyps.append(stump)
    hyps_ws.append(alpha)
    nerrs.append(computeError(hyps, hyps_ws, features, classes))
    
    
fig = plt.plot(range(1, max_hyp +1), nerrs, color="red", linewidth=1.0)