from data_set import DataSet
if __name__ == '__main__':
argparser = argparse.ArgumentParser(
description='Demonstrate Multilayer Perceptron')
argparser.add_argument(
'--training-epochs',
dest='epochs',
type=int,
default='1000',
help='number of epochs to run the training (default: 1000)')
dataset = DataSet()
dataset.load()
trainer = MultilayerPerceptronTrainer(
dataset, n_epochs=argparser.parse_args().epochs)
trainer.initialize()
state = trainer.start_training(patience=10000,
patience_increase=2,
improvement_threshold=0.995)
start_time = time.clock()
while (trainer.continue_training(state)):
print('epoch %d, validation error %f%%' %
(state.epoch, state.epoch_losses[-1][0] * 100.0))
end_time = time.clock()
print >> sys.stderr, ('The code for file ' + os.path.split(__file__)[1] +
' ran for %.2fm' % ((end_time - start_time) / 60.))
print(('Optimization complete. Best validation score of %f%% '
'obtained at iteration %i, with test performance %f%%') %
class TestTutorials(unittest.TestCase):
"""docstring for TestTutorials"""
@classmethod
def setUpClass(self):
self.dataset = DataSet()
self.dataset.load(100)
def test_convolutional_multilayer_perceptron(self):
lenet5 = ConvolutionalMultilayerPerceptronTrainer(self.dataset,
n_epochs=1,
batch_size=2)
lenet5.initialize(nkerns=[2, 5])
epoch_losses, best_validation_loss, best_iter, test_score = lenet5.train(
patience=10000, patience_increase=2, improvement_threshold=0.995)
self.assertEqual(epoch_losses, [[0.52000000000000002, 49]])
self.assertEqual(test_score, 0.45000000000000001)
def test_convolutional_multilayer_perceptron_incremental(self):
lenet5 = ConvolutionalMultilayerPerceptronTrainer(self.dataset,
n_epochs=1,
batch_size=2)
lenet5.initialize(nkerns=[2, 5])
state = lenet5.start_training(patience=10000,
patience_increase=2,
improvement_threshold=0.995)
while lenet5.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.52000000000000002, 49]])
self.assertEqual(state.test_score, 0.45000000000000001)
def test_deep_belief_network(self):
dbn = DeepBeliefNetworkTrainer(self.dataset,
batch_size=2,
pretraining_epochs=1,
training_epochs=1)
dbn.initialize()
layer_epoch_costs = dbn.pretrain()
self.assertTrue(layer_epoch_costs[0][0] > -229.574659742916
and layer_epoch_costs[0][0] < -229.574659742915)
self.assertTrue(layer_epoch_costs[1][0] > -724.564076667859
and layer_epoch_costs[1][0] < -724.564076667856)
self.assertTrue(layer_epoch_costs[2][0] > -237.068920458976
and layer_epoch_costs[2][0] < -237.068920458975)
epoch_losses, best_validation_loss, best_iter, test_score = dbn.train()
self.assertEqual(best_validation_loss, 0.79)
self.assertEqual(best_iter, 49)
self.assertEqual(test_score, 0.76)
def test_deep_belief_network_incremental(self):
dbn = DeepBeliefNetworkTrainer(self.dataset,
batch_size=2,
pretraining_epochs=1,
training_epochs=1)
dbn.initialize()
state = dbn.start_pretraining()
while dbn.continue_pretraining(state):
pass
self.assertTrue(state.layer_epoch_costs[0] > -229.574659742916
and state.layer_epoch_costs[0] < -229.574659742915)
self.assertTrue(state.layer_epoch_costs[1] > -724.564076667859
and state.layer_epoch_costs[1] < -724.564076667856)
self.assertTrue(state.layer_epoch_costs[2] > -237.068920458976
and state.layer_epoch_costs[2] < -237.068920458975)
state = dbn.start_training()
while dbn.continue_training(state):
pass
self.assertEqual(state.best_validation_loss, 0.79)
self.assertEqual(state.best_iter, 49)
self.assertEqual(state.test_score, 0.76)
def test_denoising_autoencoder(self):
da = DenoisingAutoencoderTrainer(self.dataset,
training_epochs=1,
batch_size=2)
da.initialize()
uncorrupt_costs = da.train()
self.assertEqual(uncorrupt_costs, [149.16503228187111])
da.initialize(corruption_level=0.3)
corrupt_costs = da.train()
self.assertTrue(corrupt_costs[0] > 173.6649940882978
and corrupt_costs[0] < 173.6649940882979)
def test_denoising_autoencoder_incremental(self):
da = DenoisingAutoencoderTrainer(self.dataset,
training_epochs=1,
batch_size=2)
da.initialize()
state = da.start_training()
while da.continue_training(state):
pass
self.assertEqual(state.costs, [149.16503228187111])
da.initialize(corruption_level=0.3)
state = da.start_training()
while da.continue_training(state):
pass
self.assertTrue(state.costs[0] > 173.6649940882978
and state.costs[0] < 173.6649940882979)
def test_logistic(self):
lc = LogisticTrainer(self.dataset, batch_size=2, n_epochs=1)
lc.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = lc.train(
patience=5000, patience_increase=2, improvement_threshold=0.995)
self.assertEqual(epoch_losses, [[0.40000000000000002, 49]])
self.assertEqual(test_score, 0.30)
def test_logistic_incremental(self):
lc = LogisticTrainer(self.dataset, batch_size=2, n_epochs=1)
lc.initialize()
state = lc.start_training(patience=5000,
patience_increase=2,
improvement_threshold=0.995)
while lc.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.40000000000000002, 49]])
self.assertEqual(state.test_score, 0.30)
def test_multilayer_perceptron(self):
mp = MultilayerPerceptronTrainer(self.dataset,
n_epochs=1,
batch_size=2)
mp.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = mp.train(
patience=10000, patience_increase=2, improvement_threshold=0.995)
self.assertEqual(epoch_losses, [[0.54, 49]])
self.assertEqual(test_score, 0.52)
def test_multilayer_perceptron_incremental(self):
mp = MultilayerPerceptronTrainer(self.dataset,
n_epochs=1,
batch_size=2)
mp.initialize()
state = mp.start_training(patience=10000,
patience_increase=2,
improvement_threshold=0.995)
while mp.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.54, 49]])
self.assertEqual(state.test_score, 0.52)
def test_restricted_boltzmann_machine(self):
rbm = RestrictedBoltzmannMachineTrainer(self.dataset,
training_epochs=1,
batch_size=2)
rbm.initialize(n_chains=2, n_samples=2, n_hidden=5)
epoch_costs = rbm.train()
self.assertEqual(epoch_costs, [-174.86070176730175])
def test_restricted_boltzmann_machine_incremental(self):
rbm = RestrictedBoltzmannMachineTrainer(self.dataset,
training_epochs=1,
batch_size=2)
rbm.initialize(n_chains=2, n_samples=2, n_hidden=5)
state = rbm.start_training()
while rbm.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [-174.86070176730175])
def test_stacked_denoising_autoencoder(self):
sda = StackedDenoisingAutoencoderTrainer(self.dataset,
pretraining_epochs=1,
n_epochs=1,
batch_size=2)
sda.preinitialize()
layer_epoch_costs = sda.pretrain()
self.assertEqual(
layer_epoch_costs,
[[328.15852933515004], [771.56755018914123], [661.65193991637716]])
sda.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = sda.train(
None)
self.assertEqual(epoch_losses, [[0.73, 49]])
self.assertEqual(best_validation_loss, 0.73)
self.assertEqual(best_iter, 49)
self.assertEqual(test_score, 0.67)
def test_stacked_denoising_autoencoder_incremental(self):
sda = StackedDenoisingAutoencoderTrainer(self.dataset,
pretraining_epochs=1,
n_epochs=1,
batch_size=2)
sda.preinitialize()
state = sda.start_pretraining()
while sda.continue_pretraining(state):
pass
self.assertEqual(
state.layer_epoch_costs,
[[328.15852933515004], [771.56755018914123], [661.65193991637716]])
sda.initialize()
state = sda.start_training()
while sda.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.73, 49]])
self.assertEqual(state.best_validation_loss, 0.73)
self.assertEqual(state.best_iter, 49)
self.assertEqual(state.test_score, 0.67)
inputs,
outputs
)
self.test_eval_function = self.compiled_test_function(
self.classifier,
minibatch_index,
inputs,
outputs
)
from data_set import DataSet
if __name__ == '__main__':
dataset = DataSet()
dataset.load()
dbn = DeepBeliefNetworkTrainer(dataset)
dbn.initialize()
start_time = time.clock()
layer_epoch_costs = dbn.pretrain()
end_time = time.clock()
print >> sys.stderr, ('The pretraining code for file ' +
os.path.split(__file__)[1] +
' ran for %.2fm' % ((end_time - start_time) / 60.))
start_time = time.clock()
epoch_losses, best_validation_loss, best_iter, test_score = dbn.train()
end_time = time.clock()
print >> sys.stderr, ('The fine tuning code for file ' +
os.path.split(__file__)[1] +
class TestTutorials(unittest.TestCase):
"""docstring for TestTutorials"""
@classmethod
def setUpClass(self):
self.dataset = DataSet()
self.dataset.load(100)
def test_convolutional_multilayer_perceptron(self):
lenet5 = ConvolutionalMultilayerPerceptronTrainer(self.dataset, n_epochs = 1, batch_size = 2)
lenet5.initialize(nkerns = [2, 5])
epoch_losses, best_validation_loss, best_iter, test_score = lenet5.train(patience = 10000, patience_increase = 2, improvement_threshold = 0.995)
self.assertEqual(epoch_losses, [[0.52000000000000002, 49]])
self.assertEqual(test_score, 0.45000000000000001)
def test_convolutional_multilayer_perceptron_incremental(self):
lenet5 = ConvolutionalMultilayerPerceptronTrainer(self.dataset, n_epochs = 1, batch_size = 2)
lenet5.initialize(nkerns = [2, 5])
state = lenet5.start_training(patience = 10000, patience_increase = 2, improvement_threshold = 0.995)
while lenet5.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.52000000000000002, 49]])
self.assertEqual(state.test_score, 0.45000000000000001)
def test_deep_belief_network(self):
dbn = DeepBeliefNetworkTrainer(self.dataset, batch_size = 2, pretraining_epochs = 1, training_epochs = 1)
dbn.initialize()
layer_epoch_costs = dbn.pretrain()
self.assertTrue(layer_epoch_costs[0][0] > -229.574659742916 and layer_epoch_costs[0][0] < -229.574659742915)
self.assertTrue(layer_epoch_costs[1][0] > -724.564076667859 and layer_epoch_costs[1][0] < -724.564076667856)
self.assertTrue(layer_epoch_costs[2][0] > -237.068920458976 and layer_epoch_costs[2][0] < -237.068920458975)
epoch_losses, best_validation_loss, best_iter, test_score = dbn.train()
self.assertEqual(best_validation_loss, 0.79)
self.assertEqual(best_iter, 49)
self.assertEqual(test_score, 0.76)
def test_deep_belief_network_incremental(self):
dbn = DeepBeliefNetworkTrainer(
self.dataset,
batch_size=2,
pretraining_epochs=1,
training_epochs=1
)
dbn.initialize()
state = dbn.start_pretraining()
while dbn.continue_pretraining(state):
pass
self.assertTrue(
state.layer_epoch_costs[0] > -229.574659742916
and state.layer_epoch_costs[0] < -229.574659742915
)
self.assertTrue(
state.layer_epoch_costs[1] > -724.564076667859
and state.layer_epoch_costs[1] < -724.564076667856
)
self.assertTrue(
state.layer_epoch_costs[2] > -237.068920458976
and state.layer_epoch_costs[2] < -237.068920458975
)
state = dbn.start_training()
while dbn.continue_training(state):
pass
self.assertEqual(state.best_validation_loss, 0.79)
self.assertEqual(state.best_iter, 49)
self.assertEqual(state.test_score, 0.76)
def test_denoising_autoencoder(self):
da = DenoisingAutoencoderTrainer(self.dataset, training_epochs=1, batch_size=2)
da.initialize()
uncorrupt_costs = da.train()
self.assertEqual(uncorrupt_costs, [149.16503228187111])
da.initialize(corruption_level = 0.3)
corrupt_costs = da.train()
self.assertTrue(
corrupt_costs[0] > 173.6649940882978
and corrupt_costs[0] < 173.6649940882979
)
def test_denoising_autoencoder_incremental(self):
da = DenoisingAutoencoderTrainer(self.dataset, training_epochs=1, batch_size=2)
da.initialize()
state = da.start_training()
while da.continue_training(state):
pass
self.assertEqual(state.costs, [149.16503228187111])
da.initialize(corruption_level = 0.3)
state = da.start_training()
while da.continue_training(state):
pass
self.assertTrue(
state.costs[0] > 173.6649940882978
and state.costs[0] < 173.6649940882979
)
def test_logistic(self):
lc = LogisticTrainer(self.dataset, batch_size=2, n_epochs=1)
lc.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = lc.train(patience = 5000, patience_increase = 2, improvement_threshold = 0.995)
self.assertEqual(epoch_losses, [[0.40000000000000002, 49]])
self.assertEqual(test_score, 0.30)
def test_logistic_incremental(self):
lc = LogisticTrainer(self.dataset, batch_size=2, n_epochs=1)
lc.initialize()
state = lc.start_training(patience=5000, patience_increase=2, improvement_threshold=0.995)
while lc.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.40000000000000002, 49]])
self.assertEqual(state.test_score, 0.30)
def test_multilayer_perceptron(self):
mp = MultilayerPerceptronTrainer(self.dataset, n_epochs = 1, batch_size = 2)
mp.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = mp.train(patience = 10000, patience_increase = 2, improvement_threshold = 0.995)
self.assertEqual(epoch_losses, [[0.54, 49]])
self.assertEqual(test_score, 0.52)
def test_multilayer_perceptron_incremental(self):
mp = MultilayerPerceptronTrainer(self.dataset, n_epochs = 1, batch_size = 2)
mp.initialize()
state = mp.start_training(patience = 10000, patience_increase = 2, improvement_threshold = 0.995)
while mp.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.54, 49]])
self.assertEqual(state.test_score, 0.52)
def test_restricted_boltzmann_machine(self):
rbm = RestrictedBoltzmannMachineTrainer(self.dataset, training_epochs = 1, batch_size = 2)
rbm.initialize(n_chains = 2, n_samples = 2, n_hidden = 5)
epoch_costs = rbm.train()
self.assertEqual(epoch_costs, [-174.86070176730175])
def test_restricted_boltzmann_machine_incremental(self):
rbm = RestrictedBoltzmannMachineTrainer(self.dataset, training_epochs = 1, batch_size = 2)
rbm.initialize(n_chains = 2, n_samples = 2, n_hidden = 5)
state = rbm.start_training()
while rbm.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [-174.86070176730175])
def test_stacked_denoising_autoencoder(self):
sda = StackedDenoisingAutoencoderTrainer(
self.dataset,
pretraining_epochs=1,
n_epochs=1,
batch_size=2
)
sda.preinitialize()
layer_epoch_costs = sda.pretrain()
self.assertEqual(layer_epoch_costs, [[328.15852933515004], [771.56755018914123], [661.65193991637716]])
sda.initialize()
epoch_losses, best_validation_loss, best_iter, test_score = sda.train(None)
self.assertEqual(epoch_losses, [[0.73, 49]])
self.assertEqual(best_validation_loss, 0.73)
self.assertEqual(best_iter, 49)
self.assertEqual(test_score, 0.67)
def test_stacked_denoising_autoencoder_incremental(self):
sda = StackedDenoisingAutoencoderTrainer(self.dataset, pretraining_epochs = 1, n_epochs = 1, batch_size = 2)
sda.preinitialize()
state = sda.start_pretraining()
while sda.continue_pretraining(state):
pass
self.assertEqual(state.layer_epoch_costs, [[328.15852933515004], [771.56755018914123], [661.65193991637716]])
sda.initialize()
state = sda.start_training()
while sda.continue_training(state):
pass
self.assertEqual(state.epoch_losses, [[0.73, 49]])
self.assertEqual(state.best_validation_loss, 0.73)
self.assertEqual(state.best_iter, 49)
self.assertEqual(state.test_score, 0.67)
