def run_mlp():
# test the new way to automatically fill in inputs for models
mlp = Prototype()
x = ((None, 784), matrix("x"))
mlp.add(Dense(inputs=x, outputs=1000, activation='rectifier'))
mlp.add(Dense, outputs=1500, activation='tanh')
mlp.add(Softmax, outputs=10, out_as_probs=False)
# define our loss to optimize for the model (and the target variable)
# targets from MNIST are int64 numbers 0-9
y = lvector('y')
loss = Neg_LL(inputs=mlp.models[-1].p_y_given_x, targets=y, one_hot=False)
mnist = MNIST()
optimizer = AdaDelta(model=mlp, loss=loss, dataset=mnist, epochs=10)
optimizer.train()
test_data, test_labels = mnist.test_inputs, mnist.test_targets
test_data = test_data[:25]
test_labels = test_labels[:25]
# use the run function!
yhat = mlp.run(test_data)
print('-------')
print('Prediction: %s' % str(yhat))
print('Actual: %s' % str(test_labels.astype('int32')))
def create_mlp():
# define the model layers
relu_layer1 = Dense(input_size=784,
output_size=1000,
activation='rectifier')
relu_layer2 = Dense(inputs_hook=(1000, relu_layer1.get_outputs()),
output_size=1000,
activation='rectifier')
class_layer3 = SoftmaxLayer(inputs_hook=(1000, relu_layer2.get_outputs()),
output_size=10,
out_as_probs=False)
# add the layers as a Prototype
mlp = Prototype(layers=[relu_layer1, relu_layer2, class_layer3])
mnist = MNIST()
optimizer = AdaDelta(model=mlp, dataset=mnist, epochs=20)
optimizer.train()
test_data, test_labels = mnist.test_inputs[:25], mnist.test_targets[:25]
# use the run function!
preds = mlp.run(test_data)
log.info('-------')
log.info("predicted: %s", str(preds))
log.info("actual: %s", str(test_labels.astype('int32')))
def main():
# First, let's create a simple feedforward MLP with one hidden layer as a Prototype.
mlp = Prototype()
mlp.add(
Dense(input_size=28 * 28,
output_size=1000,
activation='rectifier',
noise='dropout'))
mlp.add(SoftmaxLayer(output_size=10))
# Now, we get to choose what values we want to monitor, and what datasets we would like to monitor on!
# Each Model (in our case, the Prototype), has a get_monitors method that will return a useful
# dictionary of {string_name: monitor_theano_expression} for various computations of the model we might
# care about. By default, this method returns an empty dictionary - it was the model creator's job to
# include potential monitor values.
mlp_monitors = mlp.get_monitors()
mlp_channel = MonitorsChannel(name="error")
for name, expression in mlp_monitors.items():
mlp_channel.add(
Monitor(name=name,
expression=expression,
train=True,
valid=True,
test=True))
# create some monitors for statistics about the hidden and output weights!
# let's look at the mean, variance, and standard deviation of the weights matrices.
weights_channel = MonitorsChannel(name="weights")
hiddens_1 = mlp[0].get_params()[0]
hiddens1_mean = T.mean(hiddens_1)
weights_channel.add(
Monitor(name="hiddens_mean", expression=hiddens1_mean, train=True))
hiddens_2 = mlp[1].get_params()[0]
hiddens2_mean = T.mean(hiddens_2)
weights_channel.add(
Monitor(name="out_mean", expression=hiddens2_mean, train=True))
# create our plot object to do live plotting!
plot = Plot(bokeh_doc_name="Monitor Tutorial",
monitor_channels=[mlp_channel, weights_channel],
open_browser=True)
# use SGD optimizer
optimizer = SGD(model=mlp,
dataset=MNIST(concat_train_valid=False),
epochs=500,
save_freq=100,
batch_size=600,
learning_rate=.01,
lr_decay=False,
momentum=.9,
nesterov_momentum=True)
# train, with the plot!
optimizer.train(plot=plot)
def get_sequenced_mnist(sequence_number=0, seq_3d=False, seq_length=30, rng=None, flatten=True):
"""
sequence_number : int, optional
The sequence method to use if we want to put the input images into a specific order. 0 defaults to random.
seq_3d : bool, optional
When sequencing, whether the output should be
3D tensors (batches, subsequences, data) or 2D (sequence, data).
seq_length: int, optional
The length of subsequences to split the data.
rng : random, optional
The random number generator to use when sequencing.
"""
mnist = MNIST(flatten=flatten)
# sequence the dataset
if sequence_number is not None:
_sequence(mnist, sequence_number=sequence_number, rng=rng)
# optionally make 3D instead of 2D
if seq_3d:
print("Making 3D....")
# chop up into sequences of length seq_length
# first make sure to chop off the remainder of the data so seq_length can divide evenly.
if mnist.train_inputs.shape[0] % seq_length != 0:
length, dim = mnist.train_inputs.shape
if mnist.train_targets.ndim == 1:
ydim = 1
else:
ydim = mnist.train_targets.shape[-1]
mnist.train_inputs = mnist.train_inputs[:seq_length * int(math.floor(length / seq_length))]
mnist.train_targets = mnist.train_targets[:seq_length * int(math.floor(length / seq_length))]
# now create the 3D tensor of sequences - they will be (num_sequences, sequence_size, 784)
mnist.train_inputs = numpy.reshape(mnist.train_inputs, (length / seq_length, seq_length, dim))
mnist.train_targets = numpy.reshape(mnist.train_targets, (length / seq_length, seq_length, ydim))
if mnist.valid_inputs.shape[0] % seq_length != 0:
length, dim = mnist.valid_inputs.shape
if mnist.valid_targets.ndim == 1:
ydim = 1
else:
ydim = mnist.valid_targets.shape[-1]
mnist.valid_inputs = mnist.valid_inputs[:seq_length * int(math.floor(length / seq_length))]
mnist.valid_targets = mnist.valid_targets[:seq_length * int(math.floor(length / seq_length))]
# now create the 3D tensor of sequences - they will be (num_sequences, sequence_size, 784)
mnist.valid_inputs = numpy.reshape(mnist.valid_inputs, (length / seq_length, seq_length, dim))
mnist.valid_targets = numpy.reshape(mnist.valid_targets, (length / seq_length, seq_length, ydim))
if mnist.test_inputs.shape[0] % seq_length != 0:
length, dim = mnist.test_inputs.shape
if mnist.test_targets.ndim == 1:
ydim = 1
else:
ydim = mnist.test_targets.shape[-1]
mnist.test_inputs = mnist.test_inputs[:seq_length * int(math.floor(length / seq_length))]
mnist.test_targets = mnist.test_targets[:seq_length * int(math.floor(length / seq_length))]
# now create the 3D tensor of sequences - they will be (num_sequences, sequence_size, 784)
mnist.test_inputs = numpy.reshape(mnist.test_inputs, (length / seq_length, seq_length, dim))
mnist.test_targets = numpy.reshape(mnist.test_targets, (length / seq_length, seq_length, ydim))
print('Train shape is: {!s}, {!s}'.format(mnist.train_inputs.shape, mnist.train_targets.shape))
print('Valid shape is: {!s}, {!s}'.format(mnist.valid_inputs.shape, mnist.valid_targets.shape))
print('Test shape is: {!s}, {!s}'.format(mnist.test_inputs.shape, mnist.test_targets.shape))
return mnist
# first need a target variable
labels = T.lvector('ys')
# negative log-likelihood for classification cost
loss = Neg_LL(inputs=lenet.models[-1].p_y_given_x,
targets=labels,
one_hot=False)
# make a monitor to view average accuracy per batch
accuracy = Monitor(name='Accuracy',
expression=1 -
(T.mean(T.neq(lenet.models[-1].y_pred, labels))),
valid=True,
test=True)
# Now grab our MNIST dataset. The version given here has each image as a single 784-dimensional vector.
# because convolutions work over 2d, let's reshape our data into the (28,28) images they originally were
# (only one channel because they are black/white images not rgb)
mnist = MNIST()
process_image = lambda img: np.reshape(img, (1, 28, 28))
mnist.train_inputs = ModifyStream(mnist.train_inputs, process_image)
mnist.valid_inputs = ModifyStream(mnist.valid_inputs, process_image)
mnist.test_inputs = ModifyStream(mnist.test_inputs, process_image)
# finally define our optimizer and train the model!
optimizer = AdaDelta(model=lenet,
dataset=mnist,
loss=loss,
epochs=10,
batch_size=64)
# train!
optimizer.train(monitor_channels=accuracy)
output_size=512,
activation='rectifier',
noise='dropout')
hidden2 = Dense(inputs_hook=(512, hidden1.get_outputs()),
output_size=512,
activation='rectifier',
noise='dropout')
class_layer = SoftmaxLayer(inputs_hook=(512, hidden2.get_outputs()),
output_size=10)
mlp = Prototype([hidden1, hidden2, class_layer])
return mlp
if __name__ == '__main__':
mlp = sequential_add_layers()
data = MNIST(concat_train_valid=True)
print data.train_inputs.shape
print data.valid_inputs.shape
print data.test_inputs.shape
optimizer = SGD(model=mlp,
dataset=data,
epochs=500,
batch_size=600,
learning_rate=.01,
momentum=.9,
nesterov_momentum=True)
optimizer.train()
# Optimization #
################
# Now that our model is complete, let's define the loss function to optimize
# first need a target variable
labels = T.lvector('ys')
# negative log-likelihood for classification cost
loss = Neg_LL(inputs=lenet.models[-1].p_y_given_x, targets=labels, one_hot=False)
# make a monitor to view average accuracy per batch
accuracy = Monitor(name='Accuracy',
expression=1-(T.mean(T.neq(lenet.models[-1].y_pred, labels))),
valid=True, test=True)
# Now grab our MNIST dataset. The version given here has each image as a single 784-dimensional vector.
# because convolutions work over 2d, let's reshape our data into the (28,28) images they originally were
# (only one channel because they are black/white images not rgb)
mnist = MNIST()
process_image = lambda img: np.reshape(img, (1, 28, 28))
mnist.train_inputs = ModifyStream(mnist.train_inputs, process_image)
mnist.valid_inputs = ModifyStream(mnist.valid_inputs, process_image)
mnist.test_inputs = ModifyStream(mnist.test_inputs, process_image)
# finally define our optimizer and train the model!
optimizer = AdaDelta(
model=lenet,
dataset=mnist,
loss=loss,
epochs=10,
batch_size=64
)
# train!
optimizer.train(monitor_channels=accuracy)
def testLeNet(self):
try:
# quick and dirty way to create a model from arbitrary layers
lenet = Prototype(outdir=None)
# our input is going to be 4D tensor of images with shape (batch_size, 1, 28, 28)
x = ((None, 1, 28, 28), ftensor4('x'))
# our first convolutional layer
lenet.add(
Conv2D(inputs=x, n_filters=20, filter_size=(5, 5),
outdir=None))
# our first pooling layer, automatically hooking inputs to the previous convolutional outputs
lenet.add(Pool2D, size=(2, 2))
# our second convolutional layer
lenet.add(Conv2D, n_filters=50, filter_size=(5, 5), outdir=None)
# our second pooling layer
lenet.add(Pool2D, size=(2, 2))
# now we need to flatten the 4D convolution outputs into 2D matrix (just flatten the trailing dimensions)
lenet.add(Flatten, ndim=2)
# one dense hidden layer
lenet.add(Dense, outputs=500, activation='tanh', outdir=None)
# hook a softmax classification layer, outputting the probabilities.
lenet.add(Softmax, outputs=10, out_as_probs=True, outdir=None)
# Grab the MNIST dataset
data = MNIST(path="../../../datasets/{!s}".format(mnist_name),
concat_train_valid=False,
flatten=False)
# define our loss to optimize for the model (and the target variable)
# targets from MNIST are int64 numbers 0-9
y = lvector('y')
loss = Neg_LL(inputs=lenet.get_outputs(), targets=y, one_hot=False)
# monitor
error_monitor = Monitor(
name='error',
expression=mean(neq(lenet.models[-1].y_pred, y)),
valid=True,
test=True,
out_service=FileService('outputs/lenet_error.txt'))
# optimize our model to minimize loss given the dataset using SGD
optimizer = SGD(model=lenet,
dataset=data,
loss=loss,
epochs=10,
batch_size=128,
learning_rate=.1,
momentum=False)
print("Training LeNet...")
optimizer.train(monitor_channels=error_monitor)
def test_subset(filename, expected, conf=0.001):
with open(filename, 'r') as f:
errs = [float(err) for err in f]
for i, (err, exp) in enumerate(zip(errs, expected)):
if i == 0:
c = conf * 10
else:
c = conf
self.assertTrue(
exp - c < round(err, 4) < exp + c,
"Errors: {!s} and Expected: {!s} -- Error at {!s} and {!s}"
.format(errs, expected, err, exp))
test_subset('outputs/lenet_error_train.txt', [
.0753, .0239, .0159, .0113, .0088, .0064, .0050, .0037, .0026,
.0019
])
test_subset('outputs/lenet_error_valid.txt', [
.0283, .0209, .0170, .0151, .0139, .0129, .0121, .0118, .0112,
.0113
])
test_subset('outputs/lenet_error_test.txt', [
.0319, .0213, .0167, .0134, .0122, .0119, .0116, .0107, .0104,
.0105
])
shutil.rmtree('outputs/')
finally:
if 'lenet' in locals():
del lenet
if 'data' in locals():
del data
if 'y' in locals():
del y
if 'x' in locals():
del x
if 'loss' in locals():
del loss
if 'optimizer' in locals():
del optimizer
)
# one dense hidden layer
lenet.add(
Dense, outputs=500, activation='tanh'
)
# hook a softmax classification layer, outputting the probabilities.
lenet.add(
Softmax, outputs=10, out_as_probs=True
)
return lenet
if __name__ == '__main__':
# Grab the MNIST dataset
data = MNIST(concat_train_valid=False)
# we need to convert the (784,) flat example from MNIST to (1, 28, 28) for a 2D greyscale image
process_mnist = lambda img: np.reshape(img, (1, 28, 28))
# we can do this by using ModifyStreams over the inputs!
data.train_inputs = ModifyStream(data.train_inputs, process_mnist)
data.valid_inputs = ModifyStream(data.valid_inputs, process_mnist)
data.test_inputs = ModifyStream(data.test_inputs, process_mnist)
# now build the actual model
lenet = build_lenet()
# define our loss to optimize for the model (and the target variable)
# targets from MNIST are int64 numbers 0-9
y = lvector('y')
loss = Neg_LL(inputs=lenet.get_outputs(), targets=y, one_hot=False)
mlp.add(Softmax, outputs=10, out_as_probs=False)
################
# Optimization #
################
# Now that our model is complete, let's define the loss function to optimize
# first need a target variable
labels = T.lvector('ys')
# negative log-likelihood for classification cost
loss = Neg_LL(inputs=mlp.models[-1].p_y_given_x,
targets=labels,
one_hot=False)
# make a monitor to view average accuracy per batch
accuracy = Monitor(name='Accuracy',
expression=1 -
(T.mean(T.neq(mlp.models[-1].y_pred, labels))),
valid=True,
test=True)
# Now grab our MNIST dataset.
mnist = MNIST()
# finally define our optimizer and train the model!
optimizer = AdaDelta(model=mlp,
dataset=mnist,
loss=loss,
epochs=10,
batch_size=64)
# train!
optimizer.train(monitor_channels=accuracy)
lenet.add(Conv2D, n_filters=50, filter_size=(5, 5))
# our second pooling layer
lenet.add(Pool2D, size=(2, 2))
# now we need to flatten the 4D convolution outputs into 2D matrix (just flatten the trailing dimensions)
lenet.add(Flatten, ndim=2)
# one dense hidden layer
lenet.add(Dense, outputs=500, activation='tanh')
# hook a softmax classification layer, outputting the probabilities.
lenet.add(Softmax, outputs=10, out_as_probs=True)
return lenet
if __name__ == '__main__':
# Grab the MNIST dataset
data = MNIST(concat_train_valid=False)
# we need to convert the (784,) flat example from MNIST to (1, 28, 28) for a 2D greyscale image
process_mnist = lambda img: np.reshape(img, (1, 28, 28))
# we can do this by using ModifyStreams over the inputs!
data.train_inputs = ModifyStream(data.train_inputs, process_mnist)
data.valid_inputs = ModifyStream(data.valid_inputs, process_mnist)
data.test_inputs = ModifyStream(data.test_inputs, process_mnist)
# now build the actual model
lenet = build_lenet()
# define our loss to optimize for the model (and the target variable)
# targets from MNIST are int64 numbers 0-9
y = lvector('y')
loss = Neg_LL(inputs=lenet.get_outputs(), targets=y, one_hot=False)
try:
import PIL.Image as Image
except ImportError:
import Image
if __name__ == '__main__':
# set up the logging environment to display outputs (optional)
# although this is recommended over print statements everywhere
import logging
from opendeep import config_root_logger
config_root_logger()
log = logging.getLogger(__name__)
log.info("Creating RBM!")
# grab the MNIST dataset
mnist = MNIST(concat_train_valid=False)
# create the RBM
rng = numpy.random.RandomState(1234)
mrg = theano.tensor.shared_randomstreams.RandomStreams(rng.randint(2**30))
config_args = {
'input_size': 28 * 28,
'hidden_size': 500,
'k': 15,
'weights_init': 'uniform',
'weights_interval': 4 * numpy.sqrt(6. / 28 * 28 + 500),
'mrg': mrg
}
rbm = RBM(**config_args)
# rbm.load_params('rbm_trained.pkl')
optimizer = Optimizer(learning_rate=0.1,
