def tryParameters(test_name,
N_hidden,
lam,
l_rate,
decay,
mom,
epochs=50,
batch_size=250):
net = Net([
BatchNorm(cifar.in_size, trainMean()),
Linear(cifar.in_size, N_hidden, lam=lam),
ReLU(N_hidden),
Linear(N_hidden, cifar.out_size, lam=lam),
Softmax(cifar.out_size)
], lam, l_rate, decay, mom)
results = net.trainMiniBatch(train, val, epochs, batch_size, shuffle=True)
print('{} Test Accuracy: {:.2f}'.format(
test_name, net.accuracy(test['one_hot'].T, test['images'].T)))
print('Final train a/c, val a/c: {:.2f}/{:.2f}, {:.2f}/{:.2f}'.format(
results['last_a_train'], results['last_c_train'],
results['last_a_val'], results['last_c_val']))
plotResults(test_name, results['a_train'], results['c_train'],
results['a_val'], results['c_val'])
#weights_plot(net, "plots/weights_vizualisation_{}.png".format(test_name), labels)
return results
def tryParameters(test_name,
lin_neurons,
with_BN,
lam,
l_rate,
decay,
mom,
epochs=50,
batch_size=250):
count = 0
layers = []
for N in lin_neurons:
not_last_layer = count < (len(lin_neurons) - 1)
layers.append(
Linear(cifar.in_size if count == 0 else lin_neurons[count - 1],
N if not_last_layer else cifar.out_size,
lam=lam))
if not_last_layer:
if with_BN:
layers.append(BatchNorm(N))
layers.append(ReLU(N))
count += 1
if len(lin_neurons) == 1 and with_BN:
layers.append(BatchNorm(cifar.out_size))
layers.append(Softmax(cifar.out_size))
# init the network
print(["{}:{},{}".format(l.name, l.in_size, l.out_size) for l in layers])
net = Net(layers, lam=lam, l_rate=l_rate, decay=0.99, mom=0.99)
results = net.trainMiniBatch(train, val, epochs, batch_size, shuffle=True)
print('{} Test Accuracy: {:.2f}'.format(
test_name, net.accuracy(test['one_hot'].T, test['images'].T)))
print('Final train a/c, val a/c: {:.2f}/{:.2f}, {:.2f}/{:.2f}'.format(
results['last_a_train'], results['last_c_train'],
results['last_a_val'], results['last_c_val']))
plotResults(test_name, results['a_train'], results['c_train'],
results['a_val'], results['c_val'])
#weights_plot(net, "plots/weights_vizualisation_{}.png".format(test_name), labels)
return results
grad = network.gradient(x_batch, t_batch)
optimizer.update(network.params, grad)
loss = network.loss(x_batch, t_batch)
train_loss_list.append(loss)
print(f'train loss: {loss}')
for key in ('W1', 'b1', 'W2', 'b2', 'W3', 'b3'):
network.params[key] -= learning_rate * grad[key]
loss = network.loss(x_batch, t_batch)
train_loss_list.append(loss)
if i % iter_per_epoch == 0:
train_acc = network.accuracy(x_train, t_train)
test_acc = network.accuracy(x_test, t_test)
train_acc_list.append(train_acc)
test_acc_list.append(test_acc)
print(train_acc, test_acc)
markers = {'train': 'o', 'test': 's'}
x = np.arange(epochs)
plt.plot(x, train_acc_list, marker='o', label='train', markevery=2)
plt.plot(x, test_acc_list, marker='s', label='test', markevery=2)
plt.xlabel("epochs")
plt.ylabel("accuracy")
plt.ylim(0, 1.0)
plt.legend(loc='lower right')
plt.show()
