import numpy as np
import pickle
import gzip
from sklearn.svm import LinearSVC, SVC
from time import time
from sklearn import metrics
def read_mnist(mnist_file):
"""
Reads MNIST data.
Parameters
----------
mnist_file : string
The name of the MNIST file (e.g., 'mnist.pkl.gz').
Returns
-------
(train_X, train_Y, val_X, val_Y, test_X, test_Y) : tuple
train_X : numpy array, shape (N=50000, d=784)
Input vectors of the training set.
train_Y: numpy array, shape (N=50000)
Outputs of the training set.
val_X : numpy array, shape (N=10000, d=784)
Input vectors of the validation set.
val_Y: numpy array, shape (N=10000)
Outputs of the validation set.
test_X : numpy array, shape (N=10000, d=784)
Input vectors of the test set.
test_Y: numpy array, shape (N=10000)
Outputs of the test set.
"""
f = gzip.open(mnist_file, 'rb')
train_data, val_data, test_data = pickle.load(f, encoding='latin1')
f.close()
train_X, train_Y = train_data
val_X, val_Y = val_data
test_X, test_Y = test_data
return train_X, train_Y, val_X, val_Y, test_X, test_Y
# get datasets
train_X, train_Y, val_X, val_Y, test_X, test_Y = read_mnist('mnist.pkl.gz')
# Use SVM with Linear_Kernel
def SVM_Linear_Kernel(C):
linear_svm = LinearSVC(C=C)
linear_svm_time = time()
linear_svm.fit(train_X, train_Y)
linear_svm_time = time() - linear_svm_time
# Validation
linear_svm_time_for_validation = time()
linear_svm_score = linear_svm.score(val_X, val_Y)
linear_svm_time_for_validation = linear_svm_time + time() - linear_svm_time_for_validation
print("Done with Validation in %f s" % linear_svm_time_for_validation)
print("Accuracy of Validation: %0.3f" % linear_svm_score)
# Test
linear_svm_time_for_test = time()
linear_svm_score = linear_svm.score(test_X, test_Y)
linear_svm_time_for_test = linear_svm_time + time() - linear_svm_time_for_test
print("Done with Test in %f s" % linear_svm_time_for_test)
print("Accuracy of Test: %0.3f" % linear_svm_score)
SVM_Linear_Kernel(1.0)
# Use SVM With GaussianRBF Kernel
def SVM_GaussianRBF_Kernel(C, gamma):
kernel_svm = SVC(C=C, gamma=gamma)
kernel_svm_time = time()
kernel_svm.fit(train_X, train_Y)
kernel_svm_time = time() - kernel_svm_time
# Validation
kernel_svm_time_for_validation = time()
kernel_svm_score = kernel_svm.score(val_X, val_Y)
kernel_svm_time_for_validation = kernel_svm_time + time() - kernel_svm_time_for_validation
print("Done with Validation in %f s" % kernel_svm_time_for_validation)
print("Accuracy of Validation: %0.3f" % kernel_svm_score)
# Test
kernel_svm_time_for_test = time()
kernel_svm_score = kernel_svm.score(test_X, test_Y)
kernel_svm_time_for_test = kernel_svm_time + time() - kernel_svm_time_for_test
print("Done with Test in %f s" % kernel_svm_time_for_test)
print("Accuracy of Test: %0.3f" % kernel_svm_score)
SVM_GaussianRBF_Kernel(0.1, 0.9)
# Test
train_X, train_Y, val_X, val_Y, test_X, test_Y = read_mnist('mnist.pkl.gz')
print('train_X.shape =', train_X.shape)
print('train_Y.shape =', train_Y.shape)
print('val_X.shape =', val_X.shape)
print('val_Y.shape =', val_Y.shape)
print('test_X.shape =', test_X.shape)
print('test_Y.shape =', test_Y.shape)
print('\ntrain_X: min = %.3f, max = %.3f' %(train_X.min(), train_X.max()))
print('train_Y: min = %d, max = %d' %(train_Y.min(), train_Y.max()))