knn.py

import numpy as np 
import loader

train_features, train_labels, test_features, test_labels = loader.load_images('TrainImages')
train_features = np.reshape(train_features,(500, 4000))
train_labels = np.reshape(train_labels, (500, 2))
test_features = np.reshape(test_features, (550, 4000))
test_labels = np.reshape(test_labels, (550, 2))
train_data = zip(train_features, train_labels)
test_data = zip(test_features, test_labels)
train_data = list(train_data)
test_data = list(test_data)

def normalize(array):
	return [i/255 for i in array]

train_features = normalize(train_features)
test_features = normalize(test_features)

def get_index(labels):
	if labels[0] == 1: return 0 
	else: return 1

def l1_distance(img1, img2):
	return np.sum(np.abs(img1.ravel() - img2.ravel()))

def nearest_neighbor(img, data):
	all_dists = []
	labels = []
	minimum = data[0][0]
	min_distance = l1_distance(img, minimum)
	for each in data:
		l1_dist = l1_distance(img, each[0][0])
		all_dists.append(l1_dist)
		labels.append(get_index(each[1]))
		if l1_dist < min_distance:
			minimum = each[0]
	min_index = all_dists.index(min(all_dists))
	return labels[min_index], minimum

def k_mins(array, k):
	nn = nearest_neighbor(array[0], train_data)
	for i,j in train_data:
		if check_eq(nn[1], i):
			return 0

def check_eq(arr1, arr2):
	counter = 0
	for i, j in zip(arr1, arr2):
		if i == j:
			counter += 1
	if counter == len(arr1):
		return True 
	return False

def run_model(test, k):
	classification = []
	for entry in test:
		classification.append((k_mins(entry, k), entry[1]))
	return sum(i == j for i, j in classification)

k = 5
x = run_model(test_data, k)
print((x/len(test_data)) * 100)