python
import numpy as np
from pylearn2.models import mlp
from pylearn2.training_algorithms import sgd
from pylearn2.termination_criteria import EpochCounter
from pylearn2.datasets import mnist
from pylearn2.train import Train
from pylearn2.training_callbacks import MonitorBasedSaveBest
train_set = mnist.MNIST(which_set='train', start=0, stop=50000)
valid_set = mnist.MNIST(which_set='train', start=50000, stop=60000)
test_set = mnist.MNIST(which_set='test')
layers = [mlp.Sigmoid(layer_name='h1', dim=100, irange=0.1),
mlp.Softmax(layer_name='y', n_classes=10, irange=0.1)]
model = mlp.MLP(layers, nvis=784)
algorithm = sgd.SGD(learning_rate=0.01, cost=mlp.Costs.MeanSquaredReconstructionError())
termination_criterion = EpochCounter(max_epochs=10)
callbacks = [MonitorBasedSaveBest(channel_name='valid_y_misclass', save_path='best_model.pkl')]
trainer = Train(model=model,
dataset=train_set,
algorithm=algorithm,
extensions=callbacks,
termination_criterion=termination_criterion)
trainer.main_loop()
test_error = trainer.evaluate(test_set)
print("Test Error: {:.2f}%".format(test_error * 100))
