Skip to content

Update binomial_cross_entropy_layer #97

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jzilly wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Update binomial_cross_entropy_layer #97

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 16 additions & 17 deletions brainstorm/layers/binomial_cross_entropy_layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,12 +39,12 @@ class BinomialCrossEntropyLayerImpl(Layer):
computes_no_input_deltas_for = ['targets']

def setup(self, kwargs, in_shapes):
if in_shapes['default'] != in_shapes['targets']:
if in_shapes['default'].feature_size != in_shapes['targets'].feature_size:
raise LayerValidationError("{}: default and targets must have the "
"same shapes but got {} and {}"
.format(self.name,
in_shapes['default'],
in_shapes['targets']))
in_shapes['default'].feature_shape,
in_shapes['targets'].feature_shape))
outputs = OrderedDict()
outputs['default'] = BufferStructure('T', 'B', 1)

Expand All @@ -59,10 +59,10 @@ def setup(self, kwargs, in_shapes):
def forward_pass(self, buffers, training_pass=True):
# prepare
_h = self.handler
y = buffers.inputs.default
t = buffers.inputs.targets
cee = buffers.internals.cee
cee_sum = buffers.outputs.default
y = flatten_time_and_features(buffers.inputs.default)
t = flatten_time_and_features(buffers.inputs.targets)
cee = flatten_time_and_features(buffers.internals.cee)
cee_sum = flatten_time(buffers.outputs.default)

# the binomial cross entropy error is given by
# - t * ln(y) - (1-t) * ln(1-y)
Expand All @@ -79,9 +79,7 @@ def forward_pass(self, buffers, training_pass=True):

_h.add_tt(tmp, cee, cee) # cee = (1-t) * ln(1-y) + t * ln(y)

# reshape for summation
cee = flatten_time_and_features(cee)
cee_sum = flatten_time(cee_sum)
# summation
_h.sum_t(cee, axis=1, out=cee_sum)
_h.mult_st(-1, cee_sum, cee_sum) # * -1

Expand All @@ -90,12 +88,13 @@ def backward_pass(self, buffers):
_h = self.handler
ceed_sum = buffers.output_deltas.default
ceed = buffers.internals.ceed
tmp2 = _h.allocate(ceed.shape)
ceed = flatten_time_and_features(ceed)
tmp = _h.allocate(ceed.shape)
y = flatten_time_and_features(buffers.inputs.default)
t = flatten_time_and_features(buffers.inputs.targets)

y = buffers.inputs.default
t = buffers.inputs.targets

yd = buffers.input_deltas.default
yd = flatten_time_and_features(buffers.input_deltas.default)

# the derivative of the binomial cross entropy error is given by
# (y - t) / (y - y²)
Expand All @@ -110,7 +109,7 @@ def backward_pass(self, buffers):

# ceed_sum has only one feature dimension due to summation,
# so we broadcast to all feature dimensions
_h.broadcast_t(ceed_sum, 2, tmp)
_h.mult_tt(ceed, tmp, ceed)

_h.add_tt(ceed, yd, yd)
_h.broadcast_t(ceed_sum, 2, tmp2)
tmp2 = flatten_time(tmp2)
_h.mult_add_tt(ceed, tmp2, yd)