def train(cost, error_rate, batch_size=100, num_epochs=150):
# Setting Loggesetr
timestr = time.strftime("%Y_%m_%d_at_%H_%M")
save_path = 'results/memory_' + timestr
log_path = os.path.join(save_path, 'log.txt')
os.makedirs(save_path)
fh = logging.FileHandler(filename=log_path)
fh.setLevel(logging.DEBUG)
logger.addHandler(fh)
# Training
blocks_model = Model(cost)
all_params = blocks_model.parameters
print "Number of found parameters:" + str(len(all_params))
print all_params
training_algorithm = GradientDescent(
cost=cost, parameters=all_params,
step_rule=Adam(learning_rate=0.001))
# training_algorithm = GradientDescent(
# cost=cost, params=all_params,
# step_rule=Scale(learning_rate=model.default_lr))
monitored_variables = [cost, error_rate]
# the rest is for validation
# train_data_stream, valid_data_stream = get_mnist_streams(
# 50000, batch_size)
train_data_stream, valid_data_stream = get_mnist_video_streams(batch_size)
train_monitoring = TrainingDataMonitoring(
variables=monitored_variables,
prefix="train",
after_epoch=True)
valid_monitoring = DataStreamMonitoring(
variables=monitored_variables,
data_stream=valid_data_stream,
prefix="valid",
after_epoch=True)
main_loop = MainLoop(
algorithm=training_algorithm,
data_stream=train_data_stream,
model=blocks_model,
extensions=[
train_monitoring,
valid_monitoring,
FinishAfter(after_n_epochs=num_epochs),
SaveParams('valid_misclassificationrate_apply_error_rate', blocks_model, save_path),
SaveLog(save_path, after_epoch=True),
ProgressBar(),
Printing()])
main_loop.run()
def train(cost, error_rate, batch_size=100, num_epochs=150):
# Setting Loggesetr
timestr = time.strftime("%Y_%m_%d_at_%H_%M")
save_path = 'results/memory_' + timestr
log_path = os.path.join(save_path, 'log.txt')
os.makedirs(save_path)
fh = logging.FileHandler(filename=log_path)
fh.setLevel(logging.DEBUG)
logger.addHandler(fh)
# Training
blocks_model = Model(cost)
all_params = blocks_model.parameters
print "Number of found parameters:" + str(len(all_params))
print all_params
training_algorithm = GradientDescent(cost=cost,
parameters=all_params,
step_rule=Adam(learning_rate=0.001))
# training_algorithm = GradientDescent(
# cost=cost, params=all_params,
# step_rule=Scale(learning_rate=model.default_lr))
monitored_variables = [cost, error_rate]
# the rest is for validation
# train_data_stream, valid_data_stream = get_mnist_streams(
# 50000, batch_size)
train_data_stream, valid_data_stream = get_mnist_video_streams(batch_size)
train_monitoring = TrainingDataMonitoring(variables=monitored_variables,
prefix="train",
after_epoch=True)
valid_monitoring = DataStreamMonitoring(variables=monitored_variables,
data_stream=valid_data_stream,
prefix="valid",
after_epoch=True)
main_loop = MainLoop(
algorithm=training_algorithm,
data_stream=train_data_stream,
model=blocks_model,
extensions=[
train_monitoring, valid_monitoring,
FinishAfter(after_n_epochs=num_epochs),
SaveParams('valid_misclassificationrate_apply_error_rate',
blocks_model, save_path),
SaveLog(save_path, after_epoch=True),
ProgressBar(),
Printing()
])
main_loop.run()
hyperparameters["batched_window"] = True
cropper = LocallySoftRectangularCropper(
patch_shape=patch_shape,
hyperparameters=hyperparameters,
kernel=Gaussian())
patch = cropper.apply(
x.reshape((1, 1,) + image_shape),
np.array([list(image_shape)]),
location,
scale)
f = theano.function([x, location, scale], patch, allow_input_downcast=True)
tds, vds = get_mnist_video_streams(1)
image = tds.get_epoch_iterator().next()[0][10, 0, :]
patch1 = f(image, [[60.0, 65.0]], [[1.0, 1.0]])[0, 0]
patch2 = f(image, [[60.0, 65.0]], [[2.0, 2.0]])[0, 0]
patch3 = f(image, [[50.0, 50.0]], [[0.28, 0.28]])[0, 0]
plt.imshow(
image.reshape(image_shape),
cmap=plt.gray(),
interpolation='nearest')
plt.savefig('img0.png')
plt.imshow(
patch1.reshape(patch_shape),
cmap=plt.gray(),
def train(cost, monitorings, batch_size=100, num_epochs=500):
# Setting Loggesetr
timestr = time.strftime("%Y_%m_%d_at_%H_%M")
save_path = 'results/test_' + timestr
log_path = os.path.join(save_path, 'log.txt')
os.makedirs(save_path)
fh = logging.FileHandler(filename=log_path)
fh.setLevel(logging.DEBUG)
logger.addHandler(fh)
# Training
blocks_model = Model(cost)
all_params = blocks_model.parameters
print "Number of found parameters:" + str(len(all_params))
print all_params
# grads = T.grad(cost, all_params)
# from blocks.graph import ComputationGraph
# cg = ComputationGraph(cost)
# f = theano.function(cg.inputs, grads)
# tds, vds = get_mnist_video_streams(100)
# data = tds.get_epoch_iterator().next()
# res = f(data[1], data[0])
# res_norm = [np.mean(np.abs(r)) for r in res]
# params_dicts = blocks_model.get_parameter_dict()
# for e1, e2 in zip(params_dicts, res_norm):
# print str(e1) + ": " + str(e2)
# import ipdb; ipdb.set_trace()
clipping = StepClipping(threshold=np.cast[floatX](20))
adam = Adam(learning_rate=0.0001)
step_rule = CompositeRule([adam, clipping])
training_algorithm = GradientDescent(
cost=cost, parameters=all_params,
step_rule=step_rule)
monitored_variables = [
cost,
aggregation.mean(training_algorithm.total_gradient_norm)] + monitorings
blocks_model = Model(cost)
params_dicts = blocks_model.get_parameter_dict()
for name, param in params_dicts.iteritems():
to_monitor = training_algorithm.gradients[param].norm(2)
to_monitor.name = name + "_grad_norm"
monitored_variables.append(to_monitor)
to_monitor = param.norm(2)
to_monitor.name = name + "_norm"
monitored_variables.append(to_monitor)
train_data_stream, valid_data_stream = get_mnist_video_streams(batch_size)
train_monitoring = TrainingDataMonitoring(
variables=monitored_variables,
prefix="train",
after_epoch=True)
valid_monitoring = DataStreamMonitoring(
variables=monitored_variables,
data_stream=valid_data_stream,
prefix="valid",
after_epoch=True)
main_loop = MainLoop(
algorithm=training_algorithm,
data_stream=train_data_stream,
model=blocks_model,
extensions=[
train_monitoring,
valid_monitoring,
FinishAfter(after_n_epochs=num_epochs),
SaveParams('valid_misclassificationrate_apply_error_rate',
blocks_model, save_path),
SaveLog(save_path, after_epoch=True),
ProgressBar(),
Printing()])
main_loop.run()
hyperparameters["batched_window"] = True
cropper = LocallySoftRectangularCropper(
patch_shape=patch_shape,
hyperparameters=hyperparameters,
kernel=Gaussian())
patch = cropper.apply(
x.reshape((1, 1,) + image_shape),
np.array([list(image_shape)]),
location,
scale)
f = theano.function([x, location, scale], patch, allow_input_downcast=True)
tds, vds = get_mnist_video_streams(1)
image = tds.get_epoch_iterator().next()[0][2, 0, :]
import ipdb; ipdb.set_trace()
location = [[60.0, 65.0]]
scale = [[1.0, 1.0]]
patch1 = f(image, location, scale)[0, 0]
plt.imshow(
image.reshape(image_shape),
cmap=plt.gray(),
interpolation='nearest')
plt.savefig('img0.png')