Skip to content
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

Jump to bottom

Enable drop in average and max layer #1027

Merged
merged 7 commits into from
Mar 13, 2017
Merged

Enable drop in average and max layer #1027

Changes from 5 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
e022c06
enable dropout in average and max layer.
lcy-seso Dec 28, 2016
15a6252
Merge branch 'develop' into enable_drop_in_average_and_max_layer
lcy-seso Dec 28, 2016
ce939b3
enable dropout rate in several computation layers.
lcy-seso Dec 28, 2016
08a817e
delete unnecessary parameters and modifications for some mathmatical
lcy-seso Dec 28, 2016
cbbad42
delete the modification of SlopeInterceptLayer.
lcy-seso Dec 28, 2016
1d58949
Merge branch 'develop' into enable_drop_in_average_and_max_layer
lcy-seso Mar 7, 2017
4375a64
fix bug to pass travis
luotao1 Mar 13, 2017
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
110 changes: 40 additions & 70 deletions python/paddle/trainer/config_parser.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1803,9 +1803,8 @@ class ConvTransLayer(ConvTransLayerBase):

@config_layer('norm')
class NormLayer(LayerBase):
def __init__(self, name, inputs, device=None):
super(NormLayer, self).__init__(
name, 'norm', 0, inputs=inputs, device=device)
def __init__(self, name, inputs, **xargs):
super(NormLayer, self).__init__(name, 'norm', 0, inputs=inputs, **xargs)
for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index)
norm_conf = self.config.inputs[input_index].norm_conf
Expand All @@ -1817,9 +1816,8 @@ def __init__(self, name, inputs, device=None):

@config_layer('pool')
class PoolLayer(LayerBase):
def __init__(self, name, inputs, device=None):
super(PoolLayer, self).__init__(
name, 'pool', 0, inputs=inputs, device=device)
def __init__(self, name, inputs, **xargs):
super(PoolLayer, self).__init__(name, 'pool', 0, inputs=inputs, **xargs)
for input_index in xrange(len(self.inputs)):
input_layer = self.get_input_layer(input_index)
pool_conf = self.config.inputs[input_index].pool_conf
Expand Down Expand Up @@ -1851,7 +1849,6 @@ def __init__(self,
inputs,
active_type="linear",
bias=True,
device=None,
use_global_stats=True,
moving_average_fraction=0.9,
batch_norm_type=None,
Expand Down Expand Up @@ -1893,7 +1890,6 @@ def __init__(self,
0,
active_type=active_type,
inputs=inputs,
device=device,
**xargs)

if use_global_stats is not None:
Expand Down Expand Up @@ -1927,9 +1923,9 @@ def calc_parameter_size(self, image_conf):

@config_layer('trans')
class TransLayer(LayerBase):
def __init__(self, name, inputs, device=None):
def __init__(self, name, inputs, **xargs):
super(TransLayer, self).__init__(
name, 'trans', 0, inputs=inputs, device=device)
name, 'trans', 0, inputs=inputs, **xargs)
config_assert(
len(self.inputs) == 1,
'TransLayer must have one and only one input')
Expand All @@ -1938,9 +1934,9 @@ def __init__(self, name, inputs, device=None):

@config_layer('resize')
class ResizeLayer(LayerBase):
def __init__(self, name, size, inputs, device=None):
def __init__(self, name, size, inputs, **xargs):
super(ResizeLayer, self).__init__(
name, 'resize', size=size, inputs=inputs, device=device)
name, 'resize', size=size, inputs=inputs, **xargs)
config_assert(
len(self.inputs) == 1,
'ResizeLayer must have one and only one input')
Expand Down Expand Up @@ -2265,14 +2261,9 @@ def Generator(

@config_layer('expand')
class ExpandLayer(LayerBase):
def __init__(self,
name,
inputs,
trans_type='non-seq',
device=None,
bias=False):
def __init__(self, name, inputs, trans_type='non-seq', bias=False, **xargs):
super(ExpandLayer, self).__init__(
name, 'expand', 0, inputs=inputs, device=device)
name, 'expand', 0, inputs=inputs, **xargs)
config_assert(
len(self.inputs) == 2, 'ExpandLayer takes 2 and only 2 inputs')
self.config.trans_type = trans_type
Expand Down Expand Up @@ -2303,11 +2294,10 @@ def __init__(self,
inputs,
trans_type='non-seq',
active_type='linear',
device=None,
bias=False,
output_max_index=None):
super(MaxLayer, self).__init__(
name, 'max', 0, inputs=inputs, device=device)
output_max_index=None,
**xargs):
super(MaxLayer, self).__init__(name, 'max', 0, inputs=inputs, **xargs)
config_assert(len(self.inputs) == 1, 'MaxLayer must have 1 input')
self.config.trans_type = trans_type
self.config.active_type = active_type
Expand Down Expand Up @@ -2354,15 +2344,15 @@ def __init__(self,
inputs,
active_type='linear',
trans_type='non-seq',
device=None,
bias=False):
bias=False,
**xargs):
super(SequenceLastInstanceLayer, self).__init__(
name,
'seqlastins',
0,
inputs=inputs,
device=device,
active_type=active_type)
active_type=active_type,
**xargs)
config_assert(
len(inputs) == 1, 'SequenceLastInstanceLayer must have 1 input')
self.config.trans_type = trans_type
Expand All @@ -2374,39 +2364,34 @@ def __init__(self,

@config_layer('seqfirstins')
class SequenceFirstInstanceLayer(SequenceLastInstanceLayer):
def __init__(
self,
name,
inputs,
active_type='linear',
trans_type='non-seq',
device=None,
bias=False, ):
def __init__(self,
name,
inputs,
active_type='linear',
trans_type='non-seq',
bias=False,
**xargs):
super(SequenceFirstInstanceLayer, self).__init__(
name,
inputs=inputs,
active_type=active_type,
device=device,
bias=bias)
bias=bias,
**xargs)
self.config.trans_type = trans_type
self.config.select_first = True


@config_layer('seqconcat')
class SequenceConcatLayer(LayerBase):
def __init__(self,
name,
inputs,
active_type='linear',
device=None,
bias=False):
def __init__(self, name, inputs, active_type='linear', bias=False, **xargs):
super(SequenceConcatLayer, self).__init__(
name,
'seqconcat',
0,
inputs=inputs,
device=device,
active_type=active_type)
active_type=active_type,
**xargs)
config_assert(
len(inputs) == 2, 'SequenceConcatLayer must have 2 inputs')
for input_index in xrange(len(self.inputs)):
Expand All @@ -2422,15 +2407,15 @@ def __init__(self,
size,
inputs,
active_type='linear',
device=None,
bias=False):
bias=False,
**xargs):
super(SequenceReshapeLayer, self).__init__(
name,
'seqreshape',
size,
inputs=inputs,
device=device,
active_type=active_type)
active_type=active_type,
**xargs)
config_assert(
len(inputs) == 1, 'SequenceReshapeLayer must have 1 inputs')
self.set_layer_size(size)
Expand All @@ -2439,19 +2424,9 @@ def __init__(self,

@config_layer('subseq')
class SubSequenceLayer(LayerBase):
def __init__(self,
name,
inputs,
active_type='linear',
device=None,
bias=False):
def __init__(self, name, inputs, active_type='linear', bias=False, **xargs):
super(SubSequenceLayer, self).__init__(
name,
'subseq',
0,
inputs=inputs,
device=device,
active_type=active_type)
name, 'subseq', 0, inputs=inputs, active_type=active_type, **xargs)
config_assert(len(inputs) == 3, 'SubSequenceLayer must have 3 inputs')
input_layer0 = self.get_input_layer(0)
size = input_layer0.size
Expand Down Expand Up @@ -2608,15 +2583,10 @@ def __init__(self,
average_strategy='average',
trans_type='non-seq',
active_type='linear',
device=None,
bias=False):
bias=False,
**xargs):
super(AverageLayer, self).__init__(
name,
'average',
0,
inputs=inputs,
device=device,
active_type=active_type)
name, 'average', 0, inputs=inputs, active_type=active_type, **xargs)
self.config.average_strategy = average_strategy
self.config.trans_type = trans_type
config_assert(len(inputs) == 1, 'AverageLayer must have 1 input')
Expand All @@ -2640,9 +2610,9 @@ def __init__(self, name, inputs, cos_scale=5, device=None):

@config_layer('tensor')
class TensorLayer(LayerBase):
def __init__(self, name, size, inputs, device=None, bias=True, **xargs):
def __init__(self, name, size, inputs, bias=True, **xargs):
super(TensorLayer, self).__init__(
name, 'tensor', size, inputs=inputs, device=device, **xargs)
name, 'tensor', size, inputs=inputs, **xargs)
config_assert(len(self.inputs) == 2, 'TensorLayer must have 2 inputs')
config_assert(size > 0, 'size must be positive')
config_assert(inputs[1].parameter_name == None,
Expand Down