提交 3529c6c3 编写于 作者: Y Yi Wang

Put all layers and costs in package paddle.layer

上级 8b70f0f3
...@@ -16,11 +16,11 @@ Some essential concepts that our API have to provide include: ...@@ -16,11 +16,11 @@ Some essential concepts that our API have to provide include:
1. In some topologies, layers share parameters. For 1. In some topologies, layers share parameters. For
example, example,
[the network for training a ranking model](https://github.com/PaddlePaddle/Paddle/issues/1311#issuecomment-279121850). [the network for training a ranking model](https://github.com/PaddlePaddle/Paddle/issues/1311#issuecomment-279121850).
1. At programming time, users specify topologies and possible sharing 1. At programming time, users specify topologies and possible sharing
of parameters. PaddlePaddle can figure out and create parameters of parameters. PaddlePaddle can figure out and create parameters
required (and possibly shared) by one or more topologies. required (and possibly shared) by one or more topologies.
## Starting from Examples ## Starting from Examples
...@@ -59,9 +59,9 @@ fA = f(paddle.layer.data(input_name="A")) ...@@ -59,9 +59,9 @@ fA = f(paddle.layer.data(input_name="A"))
fB = f(paddle.layer.data(input_name="B")) fB = f(paddle.layer.data(input_name="B"))
fQ = f(paddle.layer.data(input_name="Q")) fQ = f(paddle.layer.data(input_name="Q"))
topology = paddle.cost.less_than( topology = paddle.layer.less_than(
paddle.cost.cross_entropy(fA, fQ), paddle.layer.cross_entropy(fA, fQ),
paddle.cost.corss_entropy(fB, fQ)) paddle.layer.corss_entropy(fB, fQ))
# Derive parameters required in topology and create them in model. # Derive parameters required in topology and create them in model.
parameters = paddle.parameters.create(topology) parameters = paddle.parameters.create(topology)
...@@ -86,7 +86,7 @@ correspond to the two networks in the following figure: ...@@ -86,7 +86,7 @@ correspond to the two networks in the following figure:
```python ```python
def G(in): def G(in):
# over-simplified example as G has only one layers: # over-simplified example as G has only one layers:
return paddle.layer.fc(in, parameter_name="G") return paddle.layer.fc(in, parameter_name="G")
def D(in); def D(in);
# again, over-simplified: # again, over-simplified:
...@@ -94,12 +94,12 @@ def D(in); ...@@ -94,12 +94,12 @@ def D(in);
# Construct the first topology, which contains both D and G. # Construct the first topology, which contains both D and G.
# By learning this topology, we update parameters of G. # By learning this topology, we update parameters of G.
d0 = paddle.cost.should_be_false(D(G(paddle.layer.data()))) d0 = paddle.layer.should_be_false(D(G(paddle.layer.data())))
# Construct a second topology d1, which contains only D. By # Construct a second topology d1, which contains only D. By
# training this topology, we update parameters of D. Note # training this topology, we update parameters of D. Note
# that d1 share parameters with d0. # that d1 share parameters with d0.
d1 = paddle.cost.should_be_true(D(paddle.layer.data())) d1 = paddle.layer.should_be_true(D(paddle.layer.data()))
# Create parameters from a list of multiple topologies (models) for # Create parameters from a list of multiple topologies (models) for
# the chance to share parameters between these topologies. # the chance to share parameters between these topologies.
...@@ -132,16 +132,16 @@ Above two programs reveal some important design concerns: ...@@ -132,16 +132,16 @@ Above two programs reveal some important design concerns:
1. At training and inference time, `paddle.train` and `paddle.infer` 1. At training and inference time, `paddle.train` and `paddle.infer`
requires both a topology and the parameter set that holds the parameters of that topology. There are some reasons: requires both a topology and the parameter set that holds the parameters of that topology. There are some reasons:
1. This prevents users from forgetting to call 1. This prevents users from forgetting to call
`paddle.parameters.create`. `paddle.parameters.create`.
1. `paddle.train` needs to know which parameter set to update. 1. `paddle.train` needs to know which parameter set to update.
1. Users could load another (pre-trained) parameter set and use it 1. Users could load another (pre-trained) parameter set and use it
with a topology in `train.infer`. with a topology in `train.infer`.
1. By specifying the `immutable_parameters` parameter of 1. By specifying the `immutable_parameters` parameter of
`paddle.train`, we can forbid the update of these parameters. `paddle.train`, we can forbid the update of these parameters.
## Reader ## Reader
...@@ -190,7 +190,7 @@ access a Kubernetes cluster, s/he should be able to call ...@@ -190,7 +190,7 @@ access a Kubernetes cluster, s/he should be able to call
```python ```python
paddle.dist_train(model, paddle.dist_train(model,
trainer=paddle.trainer.SGD(..., trainer=paddle.trainer.SGD(...,
paddle.updater.Adam(...)), paddle.updater.Adam(...)),
reader=read, reader=read,
k8s_user="yi", k8s_user="yi",
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册