From 0bd75a57e866bc1cff0b25cc1ea433f3104bf77e Mon Sep 17 00:00:00 2001 From: qiaolongfei Date: Tue, 14 Feb 2017 15:15:56 +0800 Subject: [PATCH] change layers to layer --- demo/mnist/api_train.py | 21 +++---- python/paddle/v2/__init__.py | 4 +- python/paddle/v2/{layers.py => layer.py} | 70 +++++++++++++++--------- 3 files changed, 58 insertions(+), 37 deletions(-) rename python/paddle/v2/{layers.py => layer.py} (70%) diff --git a/demo/mnist/api_train.py b/demo/mnist/api_train.py index 6d071cdffe..f0c56b77cf 100644 --- a/demo/mnist/api_train.py +++ b/demo/mnist/api_train.py @@ -10,7 +10,7 @@ import random import numpy as np import paddle.trainer.PyDataProvider2 as dp -import paddle.v2 +import paddle.v2 as paddle_v2 import py_paddle.swig_paddle as api from paddle.trainer_config_helpers import * from py_paddle import DataProviderConverter @@ -58,7 +58,7 @@ def input_order_converter(generator): def main(): api.initPaddle("-use_gpu=false", "-trainer_count=4") # use 4 cpu cores - optimizer = paddle.v2.optimizer.Adam( + optimizer = paddle_v2.optimizer.Adam( learning_rate=1e-4, batch_size=1000, model_average=ModelAverage(average_window=0.5), @@ -71,16 +71,17 @@ def main(): assert isinstance(updater, api.ParameterUpdater) # define network - images = paddle.v2.layers.data_layer(name='pixel', size=784) - label = paddle.v2.layers.data_layer(name='label', size=10) - hidden1 = paddle.v2.layers.fc_layer(input=images, size=200) - hidden2 = paddle.v2.layers.fc_layer(input=hidden1, size=200) - inference = paddle.v2.layers.fc_layer( - input=hidden2, size=10, act=SoftmaxActivation()) - cost = paddle.v2.layers.classification_cost(input=inference, label=label) + images = paddle_v2.layer.data(name='pixel', size=784) + label = paddle_v2.layer.data(name='label', size=10) + hidden1 = paddle_v2.layer.fc(input=images, size=200) + hidden2 = paddle_v2.layer.fc(input=hidden1, size=200) + inference = paddle_v2.layer.fc(input=hidden2, + size=10, + act=SoftmaxActivation()) + cost = paddle_v2.layer.classification_cost(input=inference, label=label) # Create Simple Gradient Machine. - model_config = paddle.v2.layers.parse_network(cost) + model_config = paddle_v2.layer.parse_network(cost) m = api.GradientMachine.createFromConfigProto(model_config, api.CREATE_MODE_NORMAL, optimizer.enable_types()) diff --git a/python/paddle/v2/__init__.py b/python/paddle/v2/__init__.py index c04c0cc73a..590fb5cd52 100644 --- a/python/paddle/v2/__init__.py +++ b/python/paddle/v2/__init__.py @@ -13,6 +13,6 @@ # limitations under the License. import optimizer -import layers +import layer -__all__ = ['optimizer', 'layers'] +__all__ = ['optimizer', 'layer'] diff --git a/python/paddle/v2/layers.py b/python/paddle/v2/layer.py similarity index 70% rename from python/paddle/v2/layers.py rename to python/paddle/v2/layer.py index 14efe9412c..4cdb2a3638 100644 --- a/python/paddle/v2/layers.py +++ b/python/paddle/v2/layer.py @@ -19,6 +19,21 @@ from paddle.trainer_config_helpers.default_decorators import wrap_name_default import collections +def parse_network(*outputs): + """ + parse all output layers and then generate a model config proto. + :param outputs: + :return: + """ + + def __real_func__(): + context = dict() + real_output = [each.to_proto(context=context) for each in outputs] + conf_helps.outputs(real_output) + + return __parse__(__real_func__) + + class Layer(object): def __init__(self, name, parent_layer): assert isinstance(parent_layer, dict) @@ -49,22 +64,13 @@ class Layer(object): raise NotImplementedError() -def parse_network(*outputs): - def __real_func__(): - context = dict() - real_output = [each.to_proto(context=context) for each in outputs] - conf_helps.outputs(real_output) - - return __parse__(__real_func__) - - -def __convert__(method_name, name_prefix, parent_names): +def __convert_to_v2__(method_name, name_prefix, parent_names): if name_prefix is not None: wrapper = wrap_name_default(name_prefix=name_prefix) else: wrapper = None - class __Impl__(Layer): + class V2LayerImpl(Layer): def __init__(self, name=None, **kwargs): parent_layers = dict() other_kwargs = dict() @@ -75,7 +81,7 @@ def __convert__(method_name, name_prefix, parent_names): if key not in parent_names: other_kwargs[key] = kwargs[key] - super(__Impl__, self).__init__(name, parent_layers) + super(V2LayerImpl, self).__init__(name, parent_layers) self.__other_kwargs__ = other_kwargs if wrapper is not None: @@ -89,24 +95,38 @@ def __convert__(method_name, name_prefix, parent_names): args[each] = self.__other_kwargs__[each] return getattr(conf_helps, method_name)(name=self.name, **args) - return __Impl__ + return V2LayerImpl -data_layer = __convert__('data_layer', None, []) -fc_layer = __convert__('fc_layer', name_prefix='fc', parent_names=['input']) -classification_cost = __convert__( +data = __convert_to_v2__('data_layer', None, []) +fc = __convert_to_v2__('fc_layer', name_prefix='fc', parent_names=['input']) +max_id = __convert_to_v2__( + 'maxid_layer', name_prefix='maxid_layer', parent_names=['input']) +classification_cost = __convert_to_v2__( 'classification_cost', name_prefix='classification_cost', parent_names=['input', 'label']) +cross_entropy_cost = __convert_to_v2__( + 'cross_entropy', + name_prefix='cross_entropy', + parent_names=['input', 'label']) -__all__ = ['data_layer', 'fc_layer', 'classification_cost', 'parse_network'] +__all__ = [ + 'parse_network', 'data', 'fc', 'max_id', 'classification_cost', + 'cross_entropy_cost' +] if __name__ == '__main__': - data = data_layer(name='pixel', size=784) - hidden = fc_layer(input=data, size=100, act=conf_helps.SigmoidActivation()) - predict = fc_layer( - input=[hidden, data], size=10, act=conf_helps.SoftmaxActivation()) - cost = classification_cost( - input=predict, label=data_layer( - name='label', size=10)) - print parse_network(cost) + pixel = data(name='pixel', size=784) + label = data(name='label', size=10) + hidden = fc(input=pixel, size=100, act=conf_helps.SigmoidActivation()) + inference = fc(input=hidden, size=10, act=conf_helps.SoftmaxActivation()) + maxid = max_id(input=inference) + cost1 = classification_cost(input=inference, label=label) + cost2 = cross_entropy_cost(input=inference, label=label) + + print parse_network(cost1) + print parse_network(cost2) + print parse_network(cost1, cost2) + print parse_network(cost2) + print parse_network(inference, maxid) -- GitLab