In [1]:

    

import numpy as np
from sklearn.datasets import fetch_mldata
import pandas as pd
%matplotlib notebook


    

In [2]:

    

mnist = fetch_mldata("MNIST original")
X_ = mnist.data / 255.0
y_ = mnist.target


    

In [3]:

    

print(X_.shape, y_.shape)


    

    




(70000, 784) (70000,)

In [4]:

    

rndperm = np.random.permutation(X_.shape[0])


    

In [5]:

    

N_ = X_.shape[0]
X_train = X_[rndperm][:int(0.7*N_)]
y_train = y_[rndperm][:int(0.7*N_)].reshape(-1,1)

X_cv = X_[rndperm][int(0.7*N_):int(0.85*N_)]
y_cv = y_[rndperm][int(0.7*N_):int(0.85*N_)].reshape(-1,1)

X_test = X_[rndperm][int(0.85*N_):N_]
y_test = y_[rndperm][int(0.85*N_):N_].reshape(-1,1)

print("Train: ", np.shape(X_train), " ", np.shape(y_train))
print("Cross-Validation: ", np.shape(X_cv), " ", np.shape(y_cv))
print("Test: ", np.shape(X_test), " ", np.shape(y_test))


    

    




Train:  (49000, 784)   (49000, 1)
Cross-Validation:  (10500, 784)   (10500, 1)
Test:  (10500, 784)   (10500, 1)

In [6]:

    

import matplotlib.pyplot as plt
%matplotlib inline
# Plot the graph
plt.gray()
fig = plt.figure( figsize=(16,7) )
for i in range(0,30):
    ax = fig.add_subplot(3,10,i+1)
    ax.matshow(X_train[i].reshape((28,28)).astype(float))
    ax.axis("off")
plt.show()


    

    






<matplotlib.figure.Figure at 0x10160ee48>






    

In [11]:

    

from scipy import stats
class kNNModel:
    def __init__(self, X, y, k=5):
        self.k = k
        self.X = X
        self.y = y
        
    def distanceMatrix(self, T):
        M, _ = np.shape(self.X)
        L, _ = np.shape(T)
        X2 = np.sum(self.X**2, axis=1).reshape(-1,1)
        T2 = np.sum(T**2, axis=1).reshape(-1,1)
        TX = T.dot(self.X.T)
        dm = T2 - 2*TX + X2.T
        return dm
    
    def predict(self, T, dm=None, k=None):
        if dm is None:
            dm = self.distanceMatrix(T)
        if k is None:
            k = self.k
        indices = np.argsort(dm, axis=1)[:,:k]
        return stats.mode(self.y[indices], axis=1).mode.reshape(-1,1)
    
    def accuracy(self, T, l, dm=None, k=None):
        p = self.predict(T, dm, k)
        return np.mean(p==l)


    

In [18]:

    

model = kNNModel(X_train, y_train, 6)
accu = model.accuracy(X_cv[:5], y_cv[:5], None, 2)
print(accu)


    

    




1.0

In [20]:

    

accuL = []
Ks = list(range(1,10))
dm = model.distanceMatrix(X_cv[:1000])
for k in Ks:
    print("k =", k)
    accu = model.accuracy(X_cv[:1000], y_cv[:1000], dm, k)
    accuL.append(accu)
print(accuL)


    

    




k = 1
k = 2
k = 3
k = 4
k = 5
k = 6
k = 7
k = 8
k = 9
[0.96399999999999997, 0.96699999999999997, 0.97099999999999997, 0.97099999999999997, 0.97099999999999997, 0.96999999999999997, 0.96799999999999997, 0.96799999999999997, 0.96499999999999997]

In [21]:

    

%matplotlib inline
plt.plot(Ks, accuL)
plt.show()