def tag_recurrent_dropout(self,
variables,
recurrent_dropout,
rng=None,
**hyperparameters):
from blocks.roles import OUTPUT, has_roles
ancestors = graph.deep_ancestors(variables)
for lstm in self.rnn.transitions:
variables = [
var for var in ancestors
if (has_roles(var, [OUTPUT]) and lstm in var.tag.annotations
and var.name.endswith("states"))
]
# get one dropout mask for all time steps. use the very
# first state to get the hidden state shape, else we get
# graph cycles.
initial_state = util.the(
[var for var in variables if "initial_state" in var.name])
mask = util.get_dropout_mask(initial_state.shape,
recurrent_dropout,
rng=rng)
subsequent_states = [
var for var in variables if "initial_state" not in var.name
]
graph.add_transform(subsequent_states,
graph.DropoutTransform("recurrent_dropout",
mask=mask),
reason="regularization")
def tag_convnet_dropout(outputs, rng=None, **kwargs):
from blocks.roles import has_roles, OUTPUT
cnn_outputs = OrderedDict()
for var in theano.gof.graph.ancestors(outputs):
if (has_roles(var, [OUTPUT])
and util.annotated_by_a(util.get_convolution_classes(), var)):
cnn_outputs.setdefault(util.get_path(var), []).append(var)
unique_outputs = []
for path, vars in cnn_outputs.items():
vars = util.dedup(vars, equal=util.equal_computations)
unique_outputs.append(util.the(vars))
graph.add_transform(unique_outputs,
graph.DropoutTransform("convnet_dropout", rng=rng),
reason="regularization")
def tag_convnet_dropout(outputs, rng=None, **kwargs):
from blocks.roles import has_roles, OUTPUT
cnn_outputs = OrderedDict()
for var in theano.gof.graph.ancestors(outputs):
if (has_roles(var, [OUTPUT]) and util.annotated_by_a(
util.get_convolution_classes(), var)):
cnn_outputs.setdefault(util.get_path(var), []).append(var)
unique_outputs = []
for path, vars in cnn_outputs.items():
vars = util.dedup(vars, equal=util.equal_computations)
unique_outputs.append(util.the(vars))
graph.add_transform(
unique_outputs,
graph.DropoutTransform("convnet_dropout", rng=rng),
reason="regularization")
