w2 = np.load('w2')
w3 = np.load('w3')
w4 = np.load('w4')
b1 = np.load('b1')
b2 = np.load('b2')
b3 = np.load('b3')
b4 = np.load('b4')
else:
w1, b1 = init_weights(784, 256)
w2, b2 = init_weights(256, 256)
w3, b3 = init_weights(256, 256)
w4, b4 = init_weights(256, 10)
t = time.time()
w1, w2, w3, w4, b1, b2, b3, b4, losses = minibatch_gd(
30, w1, w2, w3, w4, b1, b2, b3, b4, x_train, y_train, 10)
print("runtime: ", time.time() - t)
np.save('w1', w1)
np.save('w2', w2)
np.save('w3', w3)
np.save('w4', w4)
np.save('b1', b1)
np.save('b2', b2)
np.save('b3', b3)
np.save('b4', b4)
avg_class_rate, class_rate_per_class, y_pred = test_nn(
w1, w2, w3, w4, b1, b2, b3, b4, x_test, y_test, 10)
w3 = np.load('w3')
w4 = np.load('w4')
b1 = np.load('b1')
b2 = np.load('b2')
b3 = np.load('b3')
b4 = np.load('b4')
else:
w1, b1 = init_weights(784, 256)
w2, b2 = init_weights(256, 256)
w3, b3 = init_weights(256, 256)
w4, b4 = init_weights(256, 10)
num_epochs = 50
start_time = time.time()
w1, w2, w3, w4, b1, b2, b3, b4, losses = minibatch_gd(num_epochs, w1, w2, w3, w4, b1, b2, b3, b4, x_train, y_train, 10)
end_time = time.time()
np.save('w1', w1)
np.save('w2', w2)
np.save('w3', w3)
np.save('w4', w4)
np.save('b1', b1)
np.save('b2', b2)
np.save('b3', b3)
np.save('b4', b4)
plt.plot(np.arange(num_epochs), losses)
plt.xlabel("Epochs")
plt.ylabel("Losses")