knn_mnist.txt

import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
%matplotlib inline

ds = pd.read_csv('./mnist_train.csv')
print(ds.shape)

data = ds.values
print(data.shape)

y_train = data[:, 0]
X_train = data[:, 1:]

# X_train = (X_train - X_train.mean(axis=0))/(X_train.std(axis=0) + 1e-03)

print(y_train.shape, X_train.shape)

plt.figure(0)
idx = 1009
print(y_train[idx])
plt.imshow(X_train[idx].reshape((28, 28)), cmap='gray')
plt.show()

def dist(x1, x2): #function to calculate ecludian distance for x1 and x2 vector points
    return np.sqrt(((x1 - x2)**2).sum())


def knn(X_train, x, y_train, k=5):
    vals = [] #to store distance
    
    for ix in range(X_train.shape[0]):
        #compute distance of ix from all the points
        v = [dist(x, X_train[ix, :]), y_train[ix]]
        vals.append(v) #appends all distances in a list
    
    updated_vals = sorted(vals, key=lambda x: x[0]) #sort to get the minimum distance to return the index which is closest to the testing example
    pred_arr = np.asarray(updated_vals[:k])
    pred_arr = np.unique(pred_arr[:, 1], return_counts=True)
    pred = pred_arr[1].argmax()
    # return pred_arr[0][pred]
    return pred_arr, pred_arr[0][pred]

idq = int(np.random.random() * X_train.shape[0])
q = X_train[idq]

res = knn(X_train[:10001], q, y_train[:10001], k=5)
print(res)
print(y_train[idq])

plt.figure(0)
plt.imshow(q.reshape((28, 28)), cmap='gray')
plt.show()