#!/usr/bin/python
# -*- coding: utf-8 -*-
from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import confusion_matrix
from numpy import genfromtxt,asarray 
from imageio import imread
# Trainingdaten einlesen
 
trainingdata = genfromtxt('C:/temp/digits/digitsdata.csv', delimiter=',')
 
# Testdaten einlesen
 
testdata = genfromtxt('C:/temp/digits/digitsdatatest.csv', delimiter=',')
 
# Trainingsfeatures und -klassen trennen.
# Kolonne 0:255: Features, Kolonne 256 Klasse
# Fuer Ekrlaerung der Syntax siehe: https://docs.scipy.org/doc/numpy-1.13.0/reference/arrays.indexing.html
train_digits_features = trainingdata[:, :-1]
train_digits_class = trainingdata[:, -1]
 
# Testfeatures und -klassen trennen.
test_digits_features = testdata[:, :-1]
test_digits_class = testdata[:, -1]
 
# kNN initilisieren mit 5 Nachbarn
knn = KNeighborsClassifier(n_neighbors=5)
 
# kNN traininieren
knn.fit(train_digits_features, train_digits_class)
 
# Vorhersage
predicted_class = knn.predict(test_digits_features)
 
# Wahre Test digits ausgeben
print(test_digits_class)
 
# Predicted Test digits ausgeben
print(predicted_class)
 
## Vergleichen und Klassifikationsguete berechnen.
## Differenz muss = 0 sein wenn richtig klassifziert. Boolsche Variablen koennen summiert werden.

print sum(predicted_class - test_digits_class == 0) / float(len(predicted_class))

## Konfusionsmatrix berechnen 
mat_knn = confusion_matrix(test_digits_class, predicted_class)
print(mat_knn)

# Aquivalent der Funktion getPixelListFromFilePath
def getPixelArrayFromFilePath(filepath):
    img = imread(filepath)
    return(asarray(((img.flatten()/float(255)*2-1).reshape(1,-1))))



getPixelArrayFromFilePath("C:/temp/digits/test/testimg_181710187299_0_.gif")
print(knn.predict(getPixelArrayFromFilePath("C:/temp/digits/test/testimg_181710187299_0_.gif")))