提交 c7b6d2c4 编写于 作者: W wanghaoshuang

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into clip_op

......@@ -66,7 +66,7 @@ endif()
if(ANDROID OR IOS)
if(ANDROID)
if(AND ${CMAKE_SYSTEM_VERSION} VERSION_LESS "16")
if(${CMAKE_SYSTEM_VERSION} VERSION_LESS "16")
message(FATAL_ERROR "Unsupport standalone toolchains with Android API level lower than 16")
elseif(${CMAKE_SYSTEM_VERSION} VERSION_LESS "21")
# TODO: support glog for Android api 16 ~ 19 in the future
......
......@@ -3,7 +3,7 @@
## Ingredients
As our design principle is starting from the essence: how could we
allow users to express and solve their problems at neural networks.
allow users to express and solve their problems as neural networks.
Some essential concepts that our API have to provide include:
1. A *topology* is an expression of *layers*.
......@@ -233,7 +233,7 @@ paddle.dist_train(model,
num_parameter_servers=15)
```
The pseudo code if `paddle.dist_train` is as follows:
The pseudo code of `paddle.dist_train` is as follows:
```python
def dist_train(topology, parameters, trainer, reader, ...):
......
## Auto Gradient Checker Design
## Backgraound:
- Operator forward computing is easy to check if the result is right because it has a clear definition. **But** backpropagation is a notoriously difficult algorithm to debug and get right:
- 1. you should get the right backpropagation formula according to the forward computation.
- 2. you should implement it right in CPP.
- 3. it's difficult to prepare test data.
- Generally, it is easy to check whether the forward computation of an Operator is correct or not. However, backpropagation is a notoriously difficult algorithm to debug and get right:
1. you should get the right backpropagation formula according to the forward computation.
2. you should implement it right in CPP.
3. it's difficult to prepare test data.
- Auto gradient check gets a numeric gradient by forward Operator and use it as a reference of the backward Operator's result. It has several advantages:
- 1. numeric gradient checker only need forward operator.
- 2. user only need to prepare the input data for forward Operator.
- Auto gradient checking gets a numerical gradient by forward Operator and use it as a reference of the backward Operator's result. It has several advantages:
1. numerical gradient checker only need forward operator.
2. user only need to prepare the input data for forward Operator.
## Mathematical Theory
The following two document from stanford has a detailed explanation of how to get numeric gradient and why it's useful.
The following two document from Stanford has a detailed explanation of how to get numerical gradient and why it's useful.
- [Gradient checking and advanced optimization(en)](http://deeplearning.stanford.edu/wiki/index.php/Gradient_checking_and_advanced_optimization)
- [Gradient checking and advanced optimization(cn)](http://ufldl.stanford.edu/wiki/index.php/%E6%A2%AF%E5%BA%A6%E6%A3%80%E9%AA%8C%E4%B8%8E%E9%AB%98%E7%BA%A7%E4%BC%98%E5%8C%96)
......@@ -20,7 +20,7 @@ The following two document from stanford has a detailed explanation of how to ge
## Numeric Gradient Implementation
### Python Interface
```python
def get_numeric_gradient(op,
def get_numerical_gradient(op,
input_values,
output_name,
input_to_check,
......@@ -30,13 +30,13 @@ def get_numeric_gradient(op,
Get Numeric Gradient for an operator's input.
:param op: C++ operator instance, could be an network
:param input_values: The input variables. Should be an dictionary, key is
variable name. Value is numpy array.
:param input_values: The input variables. Should be an dictionary, whose key is
variable name, and value is numpy array.
:param output_name: The final output variable name.
:param input_to_check: The input variable need to get gradient.
:param input_to_check: The input variable with respect to which to compute the gradient.
:param delta: The perturbation value for numeric gradient method. The
smaller delta is, the more accurate result will get. But if that delta is
too small, it could occur numerical stability problem.
too small, it will suffer from numerical stability problem.
:param local_scope: The local scope used for get_numeric_gradient.
:return: The gradient array in numpy format.
"""
......@@ -45,28 +45,28 @@ def get_numeric_gradient(op,
### Explaination:
- Why need `output_name`
- One Operator may have multiple Output, you can get independent gradient from each Output. So user should set one output to calculate.
- An Operator may have multiple Output, one can get independent gradient from each Output. So caller should specify the name of the output variable.
- Why need `input_to_check`
- One operator may have multiple inputs. Gradient Op can calculate the gradient of these Inputs at the same time. But Numeric Gradient needs to calculate them one by one. So `get_numeric_gradient` is designed to calculate the gradient for one input. If you need to compute multiple inputs, you can call `get_numeric_gradient` multiple times.
- One operator may have multiple inputs. Gradient Op can calculate the gradient of these inputs at the same time. But Numeric Gradient needs to calculate them one by one. So `get_numeric_gradient` is designed to calculate the gradient for one input. If you need to compute multiple inputs, you can call `get_numeric_gradient` multiple times.
### Core Algorithm Implementation
```python
# we only compute gradient of one element each time.
# we use a for loop to compute the gradient of every element.
# we only compute gradient of one element a time.
# we use a for loop to compute the gradient of each element.
for i in xrange(tensor_size):
# get one input element throw it's index i.
# get one input element by its index i.
origin = tensor_to_check.get_float_element(i)
# add delta to it, run op and then get the sum of the result tensor.
# add delta to it, run op and then get the new value of the result tensor.
x_pos = origin + delta
tensor_to_check.set_float_element(i, x_pos)
y_pos = get_output()
# plus delta to this element, run op and get the sum of the result tensor.
# plus delta to this element, run op and get the new value of the result tensor.
x_neg = origin - delta
tensor_to_check.set_float_element(i, x_neg)
y_neg = get_output()
......@@ -85,15 +85,15 @@ def get_numeric_gradient(op,
Each Operator Kernel has three kinds of Gradient:
- 1. Numeric Gradient
- 2. CPU Operator Gradient
- 3. GPU Operator Gradient(if supported)
1. Numerical gradient
2. CPU kernel gradient
3. GPU kernel gradient (if supported)
Numeric Gradient Only relies on forward Operator. So we use Numeric Gradient as the reference value.
The numerical gradient only relies on forward Operator. So we use the numerical gradient as the reference value. And the gradient checking is performed in the following three steps:
- 1. calculate the numeric gradient.
- 2. calculate CPU kernel Gradient with the backward Operator and compare it with the numeric gradient.
- 3. calculate GPU kernel Gradient with the backward Operator and compare it with the numeric gradient.(if support GPU)
1. calculate the numerical gradient
2. calculate CPU kernel gradient with the backward Operator and compare it with the numerical gradient
3. calculate GPU kernel gradient with the backward Operator and compare it with the numeric gradient (if supported)
#### Python Interface
......@@ -110,8 +110,8 @@ Numeric Gradient Only relies on forward Operator. So we use Numeric Gradient as
:param forward_op: used to create backward_op
:param input_vars: numpy value of input variable. The following
computation will use these variables.
:param inputs_to_check: inputs var names that should check gradient.
:param output_name: output name that used to
:param inputs_to_check: the input variable with respect to which to compute the gradient.
:param output_name: The final output variable name.
:param max_relative_error: The relative tolerance parameter.
:param no_grad_set: used when create backward ops
:param only_cpu: only compute and check gradient on cpu kernel.
......@@ -120,24 +120,24 @@ Numeric Gradient Only relies on forward Operator. So we use Numeric Gradient as
```
### How to check if two numpy array is close enough?
if `abs_numeric_grad` is nearly zero, then use abs error for numeric_grad, not relative
if `abs_numerical_grad` is nearly zero, then use abs error for numerical_grad
```python
numeric_grad = ...
numerical_grad = ...
operator_grad = numpy.array(scope.find_var(grad_var_name(name)).get_tensor())
abs_numeric_grad = numpy.abs(numeric_grad)
# if abs_numeric_grad is nearly zero, then use abs error for numeric_grad, not relative
abs_numerical_grad = numpy.abs(numerical_grad)
# if abs_numerical_grad is nearly zero, then use abs error for numeric_grad, not relative
# error.
abs_numeric_grad[abs_numeric_grad < 1e-3] = 1
abs_numerical_grad[abs_numerical_grad < 1e-3] = 1
diff_mat = numpy.abs(abs_numeric_grad - operator_grad) / abs_numeric_grad
diff_mat = numpy.abs(abs_numerical_grad - operator_grad) / abs_numerical_grad
max_diff = numpy.max(diff_mat)
```
#### Notes:
1,The Input data for auto gradient checker should be reasonable to avoid numeric problem.
The Input data for auto gradient checker should be reasonable to avoid numerical stability problem.
#### Refs:
......
......@@ -53,12 +53,12 @@ Let's explain using an example. Suppose that we are going to compose the FC usi
```python
def operator.mul(X1, X2):
O = Var()
paddle.cpp.create_operator("mul", input={X1, Y1], output=O)
paddle.cpp.create_operator("mul", input={X1, Y1}, output=O)
return O
def operator.add(X1, X2):
O = Var()
paddle.cpp.create_operator("add", input={X1, X2], output=O)
paddle.cpp.create_operator("add", input={X1, X2}, output=O)
return O
```
......
......@@ -56,7 +56,7 @@ For each parameter, like W and b created by `layer.fc`, marked as double circles
## Block and Graph
The word block and graph are interchangable in the desgin of PaddlePaddle. A [Block[(https://github.com/PaddlePaddle/Paddle/pull/3708) is a metaphore of the code and local variables in a pair of curly braces in programming languages, where operators are like statements or instructions. A graph of operators and variables is a representation of the block.
The word block and graph are interchangable in the desgin of PaddlePaddle. A [Block](https://github.com/PaddlePaddle/Paddle/pull/3708) is a metaphore of the code and local variables in a pair of curly braces in programming languages, where operators are like statements or instructions. A graph of operators and variables is a representation of the block.
A Block keeps operators in an array `BlockDesc::ops`
......@@ -67,4 +67,4 @@ message BlockDesc {
}
```
in the order that there appear in user programs, like the Python program at the beginning of this article. We can imagine that in `ops`, we have some forward operators, followed by some gradient operators, and then some optimization operators.
in the order that they appear in user programs, like the Python program at the beginning of this article. We can imagine that in `ops`, we have some forward operators, followed by some gradient operators, and then some optimization operators.
# Design Doc: The C++ Class `Parameters`
`Parameters` is a concept we designed in Paddle V2 API. `Parameters` is a container of parameters, and make Paddle can shared parameter between topologies. We described usages of `Parameter` in [api.md](./api.md).
`Parameters` is a concept we designed in PaddlePaddle V2 API. `Parameters` is a container of parameters, which makes PaddlePaddle capable of sharing parameter between topologies. We described usages of `Parameter` in [api.md](./api.md).
We used Python to implement Parameters when designing V2 API before. There are several defects for current implementation:
We used Python to implement Parameters when designing V2 API before. There are several defects for the current implementation:
* We just use `memcpy` to share Parameters between topologies, but this is very inefficient.
* We did not implement share Parameters while training. We just trigger `memcpy` when start training.
* We did not support sharing Parameters while training. We just trigger `memcpy` when start training.
It is necessary that we implement Parameters in CPP side. However, it could be a code refactoring for Paddle, because Paddle was designed for training only one topology before, i.e., each GradientMachine contains its Parameter as a data member. In current Paddle implementation, there are three concepts associated with `Parameters`:
It is necessary that we implement Parameters in CPP side. However, it could result a code refactoring for PaddlePaddle, because PaddlePaddle was designed for training only one topology before, i.e., each GradientMachine contains its Parameter as a data member. In current PaddlePaddle implementation, there are three concepts associated with `Parameters`:
1. `paddle::Parameter`. A `Parameters` is a container for `paddle::Parameter`.
It is evident that we should use `paddle::Parameter` when developing `Parameters`.
However, the `Parameter` class contains many functions and does not have a clear interface.
It contains `create/store Parameter`, `serialize/deserialize`, `optimize(i.e SGD)`, `randomize/zero`.
When we developing `Parameters`, we only use `create/store Parameter` functionality.
We should extract functionalities of Parameter into many classes to clean Paddle CPP implementation.
We should extract functionalities of Parameter into many classes to clean PaddlePaddle CPP implementation.
2. `paddle::GradientMachine` and its sub-classes, e.g., `paddle::MultiGradientMachine`, `paddle::NeuralNetwork`.
We should pass `Parameters` to `paddle::GradientMachine` when `forward/backward` to avoid `memcpy` between topologies.
......@@ -24,7 +24,7 @@ Also, we should handle multi-GPU/CPU training, because `forward` and `backward`
So `Parameters` should be used by `paddle::ParameterUpdater`, and `paddle::ParameterUpdater` should optimize `Parameters` (by SGD).
The step by step approach for implementation Parameters in Paddle C++ core is listed below. Each step should be a PR and could be merged into Paddle one by one.
The step by step approach for implementation Parameters in PaddlePaddle C++ core is listed below. Each step should be a PR and could be merged into PaddlePaddle one by one.
1. Clean `paddle::Parameter` interface. Extract the functionalities of `paddle::Parameter` to prepare for the implementation of Parameters.
......
# Design Doc: ProgramDesc
The basic structure of a PaddlePaddle program is some nested blocks, as a C++ or Java program.
As described in [graph.md](./graph.md), the first five lines of the following PaddlePaddle program
```python
x = layer.data("images")
l = layer.data("label")
y = layer.fc(x)
cost = layer.mse(y, l)
optimize(cost)
train(cost, reader=mnist.train())
```
generates, or compiles, a PaddelPaddle program, which is represented by the following protobuf message:
```protobuf
message ProgramDesc {
repeated BlockDesc blocks = 1;
}
message BlockDesc {
required int32 parent = 1;
repeated VarDesc vars = 2;
repeated OpDesc ops = 3;
}
message OpDesc {
AttrDesc attrs = 1;
...
}
message AttrDesc {
required AttrType type = 1;
// index into ProgramDesc::blocks when type==BLOCK
optional int32 block = 2;
...
}
```
When each of the first five lines runs, related Python function, e.g., `layer.fc`, calls C++ InferShape functions. This InferShape function needs to access the properties of VarDesc's accessed by the current OpDesc. These VarDesc's might not be defined in the current block, but in some ancestor blocks. This requires that we can trace the parent of a block.
A nested block is often an attribute of an operator, most likely, an IfElseOp or a WhileOp. In above solution, all blocks are in `ProgramDesc::blocks`, this implicitly assigns a zero-based ID to each block -- the index of the block in `ProgramDesc::blocks`. So that `AttrDesc::block` could be an integer block ID.
With this design, the InferShape function should take the following parameters:
```c++
void InferShape(int current_block,
int current_operator,
ProgramDesc* program // might change VarDesc values.
) {
...
}
```
where
- `current_block` indices into `ProgramDesc::blocks`,
- `current_operator` indices into `BlockDesc::ops`.
......@@ -52,7 +52,7 @@ Here are valid outputs:
# a mini batch of three data items, each data item is a list (single column).
[([1,1,1],),
([2,2,2],),
([3,3,3],),
([3,3,3],)]
```
Please note that each item inside the list must be a tuple, below is an invalid output:
......
......@@ -15,7 +15,7 @@ The goal of refactorizaiton include:
1. Users write Python programs to describe the graphs and run it (locally or remotely).
1. A graph is composed of *variabels* and *operators*.
1. A graph is composed of *variables* and *operators*.
1. The description of graphs must be able to be serialized/deserialized, so it
......@@ -140,7 +140,7 @@ Compile Time -> IR -> Runtime
* `thrust` has the same API as C++ standard library. Using `transform` can quickly implement a customized elementwise kernel.
* `thrust` has more complex API, like `scan`, `reduce`, `reduce_by_key`.
* Hand-writing `GPUKernel` and `CPU` code
* Do not write `.h`. CPU Kernel should be in `.cc`. CPU kernel should be in `.cu`. (`GCC` cannot compile GPU code.)
* Do not write `.h`. CPU Kernel should be in `.cc`. GPU kernel should be in `.cu`. (`GCC` cannot compile GPU code.)
---
# Operator Register
......
# Paddle发行规范
# PaddlePaddle发行规范
Paddle使用git-flow branching model做分支管理,使用[Semantic Versioning](http://semver.org/)标准表示Paddle版本号。
PaddlePaddle使用git-flow branching model做分支管理,使用[Semantic Versioning](http://semver.org/)标准表示PaddlePaddle版本号。
Paddle每次发新的版本,遵循以下流程:
PaddlePaddle每次发新的版本,遵循以下流程:
1.`develop`分支派生出新的分支,分支名为`release/版本号`。例如,`release/0.10.0`
2. 将新分支的版本打上tag,tag为`版本号rc.Patch号`。第一个tag为`0.10.0rc1`,第二个为`0.10.0rc2`,依次类推。
......@@ -27,14 +27,14 @@ Paddle每次发新的版本,遵循以下流程:
需要注意的是:
* `release/版本号`分支一旦建立,一般不允许再从`develop`分支合入`release/版本号`。这样保证`release/版本号`分支功能的封闭,方便测试人员测试Paddle的行为。
* `release/版本号`分支一旦建立,一般不允许再从`develop`分支合入`release/版本号`。这样保证`release/版本号`分支功能的封闭,方便测试人员测试PaddlePaddle的行为。
*`release/版本号`分支存在的时候,如果有bugfix的行为,需要将bugfix的分支同时merge到`master`, `develop``release/版本号`这三个分支。
# Paddle 分支规范
# PaddlePaddle 分支规范
Paddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范,并适应github的特性做了一些区别。
PaddlePaddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范,并适应github的特性做了一些区别。
* Paddle的主版本库遵循[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范。其中:
* PaddlePaddle的主版本库遵循[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范。其中:
* `master`分支为稳定(stable branch)版本分支。每一个`master`分支的版本都是经过单元测试和回归测试的版本。
* `develop`分支为开发(develop branch)版本分支。每一个`develop`分支的版本都经过单元测试,但并没有经过回归测试。
* `release/版本号`分支为每一次Release时建立的临时分支。在这个阶段的代码正在经历回归测试。
......@@ -42,18 +42,18 @@ Paddle开发过程使用[git-flow](http://nvie.com/posts/a-successful-git-branch
* 其他用户的fork版本库并不需要严格遵守[git-flow](http://nvie.com/posts/a-successful-git-branching-model/)分支规范,但所有fork的版本库的所有分支都相当于特性分支。
* 建议,开发者fork的版本库使用`develop`分支同步主版本库的`develop`分支
* 建议,开发者fork的版本库中,再基于`develop`版本fork出自己的功能分支。
* 当功能分支开发完毕后,向Paddle的主版本库提交`Pull Reuqest`,进而进行代码评审。
* 当功能分支开发完毕后,向PaddlePaddle的主版本库提交`Pull Reuqest`,进而进行代码评审。
* 在评审过程中,开发者修改自己的代码,可以继续在自己的功能分支提交代码。
* BugFix分支也是在开发者自己的fork版本库维护,与功能分支不同的是,BugFix分支需要分别给主版本库的`master``develop`与可能有的`release/版本号`分支,同时提起`Pull Request`
# Paddle回归测试列表
# PaddlePaddle回归测试列表
本列表说明Paddle发版之前需要测试的功能点。
本列表说明PaddlePaddle发版之前需要测试的功能点。
## Paddle Book中所有章节
## PaddlePaddle Book中所有章节
Paddle每次发版本首先要保证Paddle Book中所有章节功能的正确性。功能的正确性包括验证Paddle目前的`paddle_trainer`训练和纯使用`Python`训练模型正确性。
PaddlePaddle每次发版本首先要保证PaddlePaddle Book中所有章节功能的正确性。功能的正确性包括验证PaddlePaddle目前的`paddle_trainer`训练和纯使用`Python`训练模型正确性。
| | 新手入门章节 | 识别数字 | 图像分类 | 词向量 | 情感分析 | 语意角色标注 | 机器翻译 | 个性化推荐 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
......
......@@ -17,7 +17,7 @@ Scope is an association of a name to variable. All variables belong to `Scope`.
1. Scope only contains a map of a name to variable.
All parameters, data, states in a Net should be variables and stored inside a scope. Each op should get inputs and outputs to do computation from a scope, such as data buffer, state(momentum) etc.
All parameters, data, states in a Net should be variables and stored inside a scope. Each op should get inputs and outputs to do computation from a scope, such as data buffer, state (momentum) etc.
1. Variable can only be created by Scope and a variable can only be got from Scope. User cannot create or get a variable outside a scope. This is a constraints of our framework, and will keep our framework simple and clear.
......@@ -32,7 +32,7 @@ Scope is an association of a name to variable. All variables belong to `Scope`.
1. Scope should destruct all Variables inside it when itself is destructed. User can never store `Variable` pointer somewhere else.
Because Variable can only be got from Scope. When destroying Scope, we also need to destroy all the Variables in it. If user store `Variable` pointer to private data member or some global variable, the pointer will be a invalid pointer when associated `Scope` is destroyed.
Because Variable can only be got from Scope. When destroying Scope, we also need to destroy all the Variables in it. If user store `Variable` pointer to private data member or some global variable, the pointer will be an invalid pointer when associated `Scope` is destroyed.
```cpp
class Scope {
......@@ -50,7 +50,7 @@ class Scope {
Just like [scope](https://en.wikipedia.org/wiki/Scope_(computer_science)) in programming languages, `Scope` in the neural network can also be a local scope. There are two attributes about local scope.
1. We can create local variables in a local scope. When that local scope are destroyed, all local variables should also be destroyed.
1. We can create local variables in a local scope. When that local scope is destroyed, all local variables should also be destroyed.
2. Variables in a parent scope can be retrieved from local scopes of that parent scope, i.e., when user get a variable from a scope, it will try to search this variable in current scope. If there is no such variable in the local scope, `scope` will keep searching from its parent, until the variable is found or there is no parent.
```cpp
......@@ -121,4 +121,4 @@ Also, as the parent scope is a `shared_ptr`, we can only `Create()` a scope shar
## Orthogonal interface
`FindVar` will return `nullptr` when `name` is not found. It can be used as `Contains` method. `NewVar` will return a `Error` when there is a name conflict locally. Combine `FindVar` and `NewVar`, we can implement `NewVar` easily.
`FindVar` will return `nullptr` when `name` is not found. It can be used as `Contains` method. `NewVar` will return an `Error` when there is a name conflict locally. Combine `FindVar` and `NewVar`, we can implement `NewVar` easily.
......@@ -6,9 +6,9 @@ The Interaction between Python and C++ can be simplified as two steps:
1. C++ tells Python how many Ops there are, and what parameter do users need to offer to initialize a new Op. Python then builds API for each Op at compile time.
2. Users invoke APIs built by Python and provide necessary parameters. These parameters will be sent to C++ fo finish Op construction task.
2. Users invoke APIs built by Python and provide necessary parameters. These parameters will be sent to C++ for finishing the Op construction task.
### Message form C++ to Python
### Message from C++ to Python
We define a Protobuf message class `OpProto` to hold message needed in the first step. What should an `OpProto` contain? This question is equivalent to “What message do we need to offer, to build a Python API which is legal and user oriented and can use to describe a whole Op.”
......@@ -193,7 +193,7 @@ def fc_layer(input, size, with_bias, activation):
elif:
# ...
return act_output;
```
```
### Low Leval API
......
## Background
PaddlePaddle divides the description of neural network computation graph into two stages: compile time and runtime.
PaddlePaddle use proto message to describe compile time graph for
PaddlePaddle use proto message to describe compile time graph because
1. Computation graph should be able to be saved to a file.
1. In distributed training, the graph will be serialized and send to multiple workers.
......
......@@ -321,3 +321,55 @@ pip uninstall py_paddle paddle
然后安装paddle的python环境, 在build目录下执行
pip install python/dist/paddle*.whl && pip install ../paddle/dist/py_paddle*.whl
16. PaddlePaddle存储的参数格式是什么,如何和明文进行相互转化
---------------------------------------------------------
PaddlePaddle保存的模型参数文件内容由16字节头信息和网络参数两部分组成。头信息中,1~4字节表示PaddlePaddle版本信息,请直接填充0;5~8字节表示每个参数占用的字节数,当保存的网络参数为float类型时为4,double类型时为8;9~16字节表示保存的参数总个数。
将PaddlePaddle保存的模型参数还原回明文时,可以使用相应数据类型的 :code:`numpy.array` 加载具体网络参数,此时可以跳过PaddlePaddle模型参数文件的头信息。若在PaddlePaddle编译时,未指定按照double精度编译,默认情况下按照float精度计算,保存的参数也是float类型。这时在使用 :code:`numpy.array` 时,一般设置 :code:`dtype=float32` 。示例如下:
.. code-block:: python
def read_parameter(fname, width):
s = open(fname).read()
# skip header
vec = np.fromstring(s[16:], dtype=np.float32)
# width is the size of the corresponding layer
np.savetxt(fname + ".csv", vec.reshape(width, -1),
fmt="%.6f", delimiter=",")
将明文参数转化为PaddlePaddle可加载的模型参数时,首先构造头信息,再写入网络参数。下面的代码将随机生成的矩阵转化为可以被PaddlePaddle加载的模型参数。
.. code-block:: python
def gen_rand_param(param_file, width, height, need_trans):
np.random.seed()
header = struct.pack("iil", 0, 4, height * width)
param = np.float32(np.random.rand(height, width))
with open(param_file, "w") as fparam:
fparam.write(header + param.tostring())
17. 如何加载预训练参数
------------------------------
* 对加载预训练参数的层,设置其参数属性 :code:`is_static=True`,使该层的参数在训练过程中保持不变。以embedding层为例,代码如下:
.. code-block:: python
emb_para = paddle.attr.Param(name='emb', is_static=True)
paddle.layer.embedding(size=word_dim, input=x, param_attr=emb_para)
* 从模型文件将预训练参数载入 :code:`numpy.array`,在创建parameters后,使用 :code:`parameters.set()` 加载预训练参数。PaddlePaddle保存的模型参数文件前16字节为头信息,用户将参数载入 :code:`numpy.array` 时须从第17字节开始。以embedding层为例,代码如下:
.. code-block:: python
def load_parameter(file_name, h, w):
with open(file_name, 'rb') as f:
f.read(16) # skip header.
return np.fromfile(f, dtype=np.float32).reshape(h, w)
parameters = paddle.parameters.create(my_cost)
parameters.set('emb', load_parameter(emb_param_file, 30000, 256))
......@@ -19,12 +19,14 @@ cc_test(scope_test SRCS scope_test.cc DEPS scope)
proto_library(framework_proto SRCS framework.proto)
cc_library(attribute SRCS attribute.cc DEPS framework_proto)
cc_library(op_proto_maker SRCS op_proto_maker.cc DEPS framework_proto attribute)
cc_test(op_proto_maker_test SRCS op_proto_maker_test.cc DEPS op_proto_maker)
cc_library(op_info SRCS op_info.cc DEPS attribute framework_proto)
cc_library(operator SRCS operator.cc DEPS op_info device_context tensor scope)
cc_test(operator_test SRCS operator_test.cc DEPS operator op_registry)
cc_library(grad_op_builder SRCS grad_op_builder.cc DEPS operator)
cc_library(op_registry SRCS op_registry.cc DEPS grad_op_builder)
cc_library(op_registry SRCS op_registry.cc DEPS grad_op_builder op_proto_maker)
cc_test(op_registry_test SRCS op_registry_test.cc DEPS op_registry)
cc_test(grad_op_builder_test SRCS grad_op_builder_test.cc DEPS grad_op_builder op_registry add_op)
......
......@@ -4,13 +4,13 @@ PaddlePaddle's RNN doesn't require that all instances have the same length. To
## Challenge of Variable-length Inputs
People usually represent a mini-batch by a Tensor. For example, a mini-batch of 32 images, each of size 32x32, is a 10x32x32 Tensor. So a transformation, T, of all images can be a matrix multiplication of the 32x32xO-dimensional tensor T and the 10x32x32 Tensor.
People usually represent a mini-batch by a Tensor. For example, a mini-batch of 10 images, each of size 32x32, is a 10x32x32 Tensor. So a transformation, T, of all images can be a matrix multiplication of the 10xOx32-dimensional tensor T and the 10x32x32 Tensor.
Another example is that each mini-batch contains 32 sentences, where each word is a D-dimensional one-hot vector. If all sentences have the same length L, we can represent this mini-batch by a 32xLxD tensor. However, in most cases, sentences have variable lengths, and we will need an index data structure to record these variable lengths.
## LoD as a Solution
### Mini-Batch of variable-length sentenses
### Mini-Batch of variable-length sentences
Let's imagine a mini-batch of 3 variable lengths sentences, containing 3, 1, and 2 words respectively. We can represent it by a (3+1+2)xD tensor plus some index information:
......@@ -51,17 +51,17 @@ The many 1's on the second level seem duplicated. For this particular case of 2
In summary, as long as that the essential elements (words or images) have the same size, we can represent mini-batches by a LoD Tensor:
- The underlying tensor has size LxD1xD2x..., where D1xD2... is the size of the essential elements, and
- the first dimension size L has an additon property -- a LoD index as a nested vector:
- The first dimension size L has an additonal property -- a LoD index as a nested vector:
```c++
typedef std::vector<std::vector> > LoD;
typedef std::vector<std::<vector>> LoD;
```
- The LoD index can is not necessary when there are only two levels and all elements of the second level have length 1.
- The LoD index is not necessary when there are only two levels and all elements of the second level have length 1.
## Slicing of LoD Tensor
Consider that we have a network with three levels of RNN: the top level one handles articles, the second level one handles sentences, and the basic level one handles words. This network requires that mini-batches represented by 4 level LoD Tensor, for example,
Consider that we have a network with three levels of RNN: the top level one handles articles, the second level one handles sentences, and the basic level one handles words. This network requires that mini-batches represented by 3 level LoD Tensor, for example,
```
3
......@@ -90,8 +90,9 @@ and the <1,2>-slice of above example is
Let's go on slicing this slice. Its <1,1>-slice is
```
3
|||
1
1
|
```
### The Slicing Algorithm
......@@ -99,7 +100,7 @@ Let's go on slicing this slice. Its <1,1>-slice is
The algorithm, with over-simplified data structure, is defined as
```c++
typedef vector<vector<int> > LoD;
typedef std::vector<std::vector<int>> LoD;
struct LoDTensor {
LoD lod_;
......@@ -128,7 +129,7 @@ Suppose that we want to retrieve the <1,2>-slice
we will need to find out the starting position of this slice by summing over all leaf nodes in `LoD` to the left of the slice, i.e., 3 + 2 + 4 + 1 = 10.
To avoid the traversal of the LoD tree at slcing time, we can do it at the construction time -- instead of saving the lengths of the next level in the LoD tree, we can save the starting offset of the next level. For example, above LoD Tensor can be transformed into
To avoid the traversal of the LoD tree at slicing time, we can do it at the construction time -- instead of saving the lengths of the next level in the LoD tree, we can save the starting offset of the next level. For example, above LoD Tensor can be transformed into
```
0
......
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/framework/op_proto_maker.h"
namespace paddle {
namespace framework {
void OpProtoAndCheckerMaker::Validate() {
validated_ = true;
CheckNoDuplicatedInOutAttrs();
}
OpProtoAndCheckerMaker::VariableBuilder OpProtoAndCheckerMaker::AddInput(
const std::string& name, const std::string& comment) {
auto* input = proto_->add_inputs();
input->set_name(name);
input->set_comment(comment);
return OpProtoAndCheckerMaker::VariableBuilder{input};
}
OpProtoAndCheckerMaker::VariableBuilder OpProtoAndCheckerMaker::AddOutput(
const std::string& name, const std::string& comment) {
auto* output = proto_->add_outputs();
output->set_name(name);
output->set_comment(comment);
return OpProtoAndCheckerMaker::VariableBuilder{output};
}
void OpProtoAndCheckerMaker::CheckNoDuplicatedInOutAttrs() {
std::unordered_set<std::string> names;
auto checker = [&](const std::string& name) {
PADDLE_ENFORCE(!names.count(name), "[%s] is duplicated", name);
names.insert(name);
};
for (auto& attr : proto_->attrs()) {
checker(attr.name());
}
for (auto& input : proto_->inputs()) {
checker(input.name());
}
for (auto& output : proto_->outputs()) {
checker(output.name());
}
}
} // namespace framework
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/framework/attribute.h"
#include "paddle/framework/framework.pb.h"
namespace paddle {
namespace framework {
// this class not only make proto but also init attribute checkers.
class OpProtoAndCheckerMaker {
public:
OpProtoAndCheckerMaker(OpProto* proto, OpAttrChecker* op_checker)
: proto_(proto), op_checker_(op_checker) {}
virtual ~OpProtoAndCheckerMaker() {
PADDLE_ENFORCE(validated_, "should call Validate after build");
}
void Validate();
protected:
struct VariableBuilder {
OpProto::Var* var_;
VariableBuilder& AsDuplicable() {
var_->set_duplicable(true);
return *this;
}
VariableBuilder& AsIntermediate() {
var_->set_intermediate(true);
return *this;
}
VariableBuilder& NotInGradient() {
var_->set_not_in_gradient(true);
return *this;
}
};
VariableBuilder AddInput(const std::string& name, const std::string& comment);
VariableBuilder AddOutput(const std::string& name,
const std::string& comment);
template <typename T>
TypedAttrChecker<T>& AddAttr(const std::string& name,
const std::string& comment,
bool generated = false) {
auto* attr = proto_->add_attrs();
attr->set_name(name);
attr->set_comment(comment);
attr->set_generated(generated);
attr->set_type(AttrTypeID<T>());
return op_checker_->AddAttrChecker<T>(name);
}
void AddComment(const std::string& comment) { proto_->set_comment(comment); }
private:
void CheckNoDuplicatedInOutAttrs();
OpProto* proto_;
OpAttrChecker* op_checker_;
bool validated_{false};
};
class NOPMaker : public OpProtoAndCheckerMaker {
public:
NOPMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {}
};
} // namespace framework
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/framework/op_proto_maker.h"
#include "gtest/gtest.h"
class TestAttrProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
public:
TestAttrProtoMaker(paddle::framework::OpProto* proto,
paddle::framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddAttr<float>("scale", "scale of test op");
AddAttr<float>("scale", "scale of test op");
}
};
TEST(ProtoMaker, DuplicatedAttr) {
paddle::framework::OpProto op_proto;
paddle::framework::OpAttrChecker op_checker;
auto proto_maker = TestAttrProtoMaker(&op_proto, &op_checker);
ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);
}
class TestInOutProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
public:
TestInOutProtoMaker(paddle::framework::OpProto* proto,
paddle::framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("input", "input of test op");
AddInput("input", "input of test op");
}
};
TEST(ProtoMaker, DuplicatedInOut) {
paddle::framework::OpProto op_proto;
paddle::framework::OpAttrChecker op_checker;
auto proto_maker = TestInOutProtoMaker(&op_proto, &op_checker);
ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);
}
\ No newline at end of file
......@@ -24,6 +24,7 @@ limitations under the License. */
#include "paddle/framework/framework.pb.h"
#include "paddle/framework/grad_op_builder.h"
#include "paddle/framework/op_info.h"
#include "paddle/framework/op_proto_maker.h"
#include "paddle/framework/operator.h"
#include "paddle/framework/scope.h"
......
......@@ -228,43 +228,5 @@ std::vector<Tensor*> ExecutionContext::MultiOutput<Tensor>(
return res;
}
void OpProtoAndCheckerMaker::Validate() {
validated_ = true;
CheckNoDuplicatedInOutAttrs();
}
OpProtoAndCheckerMaker::VariableBuilder OpProtoAndCheckerMaker::AddInput(
const std::string& name, const std::string& comment) {
auto* input = proto_->add_inputs();
input->set_name(name);
input->set_comment(comment);
return OpProtoAndCheckerMaker::VariableBuilder{input};
}
OpProtoAndCheckerMaker::VariableBuilder OpProtoAndCheckerMaker::AddOutput(
const std::string& name, const std::string& comment) {
auto* output = proto_->add_outputs();
output->set_name(name);
output->set_comment(comment);
return OpProtoAndCheckerMaker::VariableBuilder{output};
}
void OpProtoAndCheckerMaker::CheckNoDuplicatedInOutAttrs() {
std::unordered_set<std::string> names;
auto checker = [&](const std::string& name) {
PADDLE_ENFORCE(!names.count(name), "[%s] is duplicated", name);
names.insert(name);
};
for (auto& attr : proto_->attrs()) {
checker(attr.name());
}
for (auto& input : proto_->inputs()) {
checker(input.name());
}
for (auto& output : proto_->outputs()) {
checker(output.name());
}
}
} // namespace framework
} // namespace paddle
......@@ -167,71 +167,6 @@ class NOP : public OperatorBase {
}
};
// this class not only make proto but also init attribute checkers.
class OpProtoAndCheckerMaker {
public:
OpProtoAndCheckerMaker(OpProto* proto, OpAttrChecker* op_checker)
: proto_(proto), op_checker_(op_checker) {}
~OpProtoAndCheckerMaker() {
PADDLE_ENFORCE(validated_, "should call Validate after build");
}
void Validate();
protected:
struct VariableBuilder {
OpProto::Var* var_;
VariableBuilder& AsDuplicable() {
var_->set_duplicable(true);
return *this;
}
VariableBuilder& AsIntermediate() {
var_->set_intermediate(true);
return *this;
}
VariableBuilder& NotInGradient() {
var_->set_not_in_gradient(true);
return *this;
}
};
VariableBuilder AddInput(const std::string& name, const std::string& comment);
VariableBuilder AddOutput(const std::string& name,
const std::string& comment);
template <typename T>
TypedAttrChecker<T>& AddAttr(const std::string& name,
const std::string& comment,
bool generated = false) {
auto* attr = proto_->add_attrs();
attr->set_name(name);
attr->set_comment(comment);
attr->set_generated(generated);
attr->set_type(AttrTypeID<T>());
return op_checker_->AddAttrChecker<T>(name);
}
void AddComment(const std::string& comment) { proto_->set_comment(comment); }
private:
void CheckNoDuplicatedInOutAttrs();
OpProto* proto_;
OpAttrChecker* op_checker_;
bool validated_{false};
};
class NOPMaker : public OpProtoAndCheckerMaker {
public:
NOPMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {}
};
class InferShapeContext {
public:
InferShapeContext(const OperatorBase& op, const Scope& scope)
......
......@@ -264,37 +264,3 @@ TEST(Operator, Clone) {
auto b = a.Clone();
ASSERT_EQ(a.Type(), b->Type());
}
class TestAttrProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
public:
TestAttrProtoMaker(paddle::framework::OpProto* proto,
paddle::framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddAttr<float>("scale", "scale of test op");
AddAttr<float>("scale", "scale of test op");
}
};
TEST(ProtoMaker, DuplicatedAttr) {
paddle::framework::OpProto op_proto;
paddle::framework::OpAttrChecker op_checker;
auto proto_maker = TestAttrProtoMaker(&op_proto, &op_checker);
ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);
}
class TestInOutProtoMaker : public paddle::framework::OpProtoAndCheckerMaker {
public:
TestInOutProtoMaker(paddle::framework::OpProto* proto,
paddle::framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("input", "input of test op");
AddInput("input", "input of test op");
}
};
TEST(ProtoMaker, DuplicatedInOut) {
paddle::framework::OpProto op_proto;
paddle::framework::OpAttrChecker op_checker;
auto proto_maker = TestInOutProtoMaker(&op_proto, &op_checker);
ASSERT_THROW(proto_maker.Validate(), paddle::platform::EnforceNotMet);
}
\ No newline at end of file
......@@ -131,8 +131,9 @@ public:
fwdPD_.reset(new eltwise_fwd::primitive_desc(fwdDesc, eng));
// use inplace for forward but save input value before submit
inVal_ = val_;
if (act.grad) {
// only copy when need do backward
copyInVal_ = nullptr;
if (act.grad && algo == mkldnn::algorithm::eltwise_tanh) {
// tanh need save src input for backward
inVal_ = MKLDNNMatrix::create(nullptr, val_->getPrimitiveDesc());
copyInVal_ = std::make_shared<mkldnn::reorder>(*val_, *inVal_);
CHECK(copyInVal_) << "should not be emptry";
......
......@@ -14,26 +14,12 @@ limitations under the License. */
#include "paddle/utils/Util.h"
#include "CostLayer.h"
#include "ValidationLayer.h"
#include "paddle/math/SparseMatrix.h"
#include "paddle/utils/Error.h"
#include "paddle/utils/Logging.h"
#include "AddtoLayer.h"
#include "CRFLayer.h"
#include "CosSimLayer.h"
#include "CostLayer.h"
#include "DataLayer.h"
#include "ExpandConvLayer.h"
#include "FullyConnectedLayer.h"
#include "HierarchicalSigmoidLayer.h"
#include "MaxLayer.h"
#include "MixedLayer.h"
#include "NormLayer.h"
#include "PoolLayer.h"
#include "TensorLayer.h"
#include "TransLayer.h"
#include "ValidationLayer.h"
DEFINE_bool(log_error_clipping, false, "enable log error clipping or not");
namespace paddle {
......@@ -109,6 +95,10 @@ ClassRegistrar<Layer, LayerConfig> Layer::registrar_;
LayerPtr Layer::create(const LayerConfig& config) {
std::string type = config.type();
// NOTE: As following types have illegal character '-',
// they can not use REGISTER_LAYER to registrar.
// Besides, to fit with old training models,
// they can not use '_' instead.
if (type == "multi-class-cross-entropy")
return LayerPtr(new MultiClassCrossEntropy(config));
else if (type == "rank-cost")
......@@ -117,8 +107,6 @@ LayerPtr Layer::create(const LayerConfig& config) {
return LayerPtr(new AucValidation(config));
else if (type == "pnpair-validation")
return LayerPtr(new PnpairValidation(config));
// NOTE: stop adding "if" statements here.
// Instead, use REGISTER_LAYER to add more layer types
return LayerPtr(registrar_.createByType(config.type(), config));
}
......
......@@ -449,13 +449,14 @@ void MKLDNNConvLayer::resetOutGrad(
cvtOutGrad_ = nullptr;
if (!outputIsOnlyMKLDNN()) {
const MatrixPtr& cpuOut = getOutput(CPU_DEVICE).grad;
outMat->setData(cpuOut->getData());
// same PrimitiveDesc with cpuInVal_
CHECK(cpuOutVal_);
cpuOutGrad_ = MKLDNNMatrix::create(cpuOut, cpuOutVal_->getPrimitiveDesc());
if (cpuOutGrad_->getPrimitiveDesc() == out->getPrimitiveDesc()) {
outMat->setData(cpuOut->getData());
out = cpuOutGrad_;
} else {
out = MKLDNNMatrix::create(nullptr, wgtPD->diff_dst_primitive_desc());
cvtOutGrad_ = MKLDNNMatrix::createReorder(cpuOutGrad_, out);
CHECK(cvtOutGrad_);
}
......
......@@ -232,6 +232,7 @@ void MKLDNNFcLayer::resetBwdBuffers(MKLDNNMatrixPtr& in,
void MKLDNNFcLayer::resetOutGrad(MKLDNNMatrixPtr& out) {
// TODO(TJ): merge outgrad
int device = outputIsOnlyMKLDNN() ? MKLDNN_DEVICE : CPU_DEVICE;
output_.grad->setData(getOutput(device).grad->getData());
// for MKLDNN device:
// can not directly cast outputgrad to mkldnnmatrix,
// since each layer can not write the inputgrad to mkldnn inputgrad.
......
......@@ -141,18 +141,16 @@ public:
}
void backward(const UpdateCallback& callback) override {
/* Do derivation */ {
if (needResetBwd_) {
resetBwd(pipelineBwd_, inGrad_, wgtGrad_, biasGrad_, outGrad_);
needResetBwd_ = false;
}
{
REGISTER_TIMER_INFO("BpActTimer", getName().c_str());
backwardActivation();
}
{
REGISTER_TIMER_INFO("mkldnn_bwdTimer", getName().c_str());
if (needResetBwd_) {
resetBwd(pipelineBwd_, inGrad_, wgtGrad_, biasGrad_, outGrad_);
needResetBwd_ = false;
}
stream_->submit(pipelineBwd_);
}
......
......@@ -26,17 +26,26 @@ DECLARE_bool(thread_local_rand_use_global_seed);
DECLARE_bool(use_gpu);
DECLARE_bool(use_mkldnn);
struct testFCDesc {
#define RUN_MKLDNN_TEST(DNN_CONFIG, REF_CONFIG, DESC) \
MKLDNNTester tester; \
for (auto bs : {DESC.bs, 1}) { \
tester.run(DNN_CONFIG, REF_CONFIG, bs, DESC.ih, DESC.iw); \
}
#define RUN_MKLDNN_TEST_LAYER(DNN_CONFIG, REF_TYPE, DESC) \
TestConfig ref = DNN_CONFIG; \
ref.layerConfig.set_type(REF_TYPE); \
RUN_MKLDNN_TEST(DNN_CONFIG, ref, DESC)
struct testFcDesc {
int bs;
int ic;
int oc;
int ih, iw; // oh == ow == 1
};
void testFcLayer(const testFCDesc& pm) {
const std::string compareTypes[] = {"mkldnn_fc", "fc"};
TestConfig cfg;
cfg.layerConfig.set_type(compareTypes[0]);
static void getMKLDNNFcConfig(TestConfig& cfg, const testFcDesc& pm) {
cfg.layerConfig.set_type("mkldnn_fc");
cfg.layerConfig.set_size(pm.oc);
cfg.inputDefs.push_back(
{INPUT_DATA,
......@@ -44,25 +53,25 @@ void testFcLayer(const testFCDesc& pm) {
/* size of input layer= */ size_t(pm.ic * pm.ih * pm.iw),
/* size of weight= */ size_t(pm.oc * pm.ic * pm.ih * pm.iw)});
cfg.layerConfig.add_inputs();
}
MKLDNNTester tester;
void testFcLayer(const testFcDesc& pm) {
TestConfig dnnConfig;
getMKLDNNFcConfig(dnnConfig, pm);
for (auto biasSize : {pm.oc, 0}) {
cfg.biasSize = biasSize;
TestConfig ref = cfg;
ref.layerConfig.set_type(compareTypes[1]);
for (auto bs : {pm.bs, 1}) {
tester.run(cfg, ref, bs, pm.ih, pm.iw);
}
dnnConfig.biasSize = biasSize;
RUN_MKLDNN_TEST_LAYER(dnnConfig, "fc", pm)
}
}
TEST(MKLDNNLayer, FcLayer) {
testFcLayer({/*bs*/ 2, /*ic*/ 2, /*oc*/ 3, /*ih*/ 1, /*iw*/ 1});
testFcLayer({/*bs*/ 3, /*ic*/ 7, /*oc*/ 19, /*ih*/ 1, /*iw*/ 1});
testFcLayer({/*bs*/ 8, /*ic*/ 16, /*oc*/ 32, /*ih*/ 13, /*iw*/ 13});
testFcLayer({/*bs*/ 4, /*ic*/ 12, /*oc*/ 18, /*ih*/ 13, /*iw*/ 11});
testFcLayer({/*bs*/ 2, /*ic*/ 64, /*oc*/ 32, /*ih*/ 16, /*iw*/ 16});
testFcLayer({/*bs*/ 15, /*ic*/ 3, /*oc*/ 6, /*ih*/ 16, /*iw*/ 16});
/* bs, ic, ih, iw, oc */
testFcLayer({2, 2, 1, 1, 3});
testFcLayer({3, 7, 1, 1, 19});
testFcLayer({8, 16, 13, 13, 32});
testFcLayer({4, 12, 13, 13, 18});
testFcLayer({2, 64, 16, 16, 32});
testFcLayer({15, 3, 16, 16, 6});
}
struct testConvDesc {
......@@ -75,13 +84,10 @@ struct testConvDesc {
int dh, dw;
};
void testConvLayer(const testConvDesc& pm) {
const std::string compareTypes[] = {"mkldnn_conv", "exconv"};
TestConfig cfg;
cfg.layerConfig.set_type(compareTypes[0]);
static void getMKLDNNConvConfig(TestConfig& cfg, const testConvDesc& pm) {
cfg.layerConfig.set_type("mkldnn_conv");
cfg.layerConfig.set_num_filters(pm.oc);
cfg.layerConfig.set_size(pm.oc * pm.oh * pm.ow);
// cfg.layerConfig.set_partial_sum(1); // TODO: check it
cfg.layerConfig.set_shared_biases(true);
cfg.inputDefs.push_back(
{INPUT_DATA,
......@@ -115,15 +121,14 @@ void testConvLayer(const testConvDesc& pm) {
int oh = outputSize(pm.ih, fh, pm.ph, pm.sh, true);
CHECK_EQ(ow, pm.ow) << "output size check failed";
CHECK_EQ(oh, pm.oh) << "output size check failed";
}
MKLDNNTester tester;
void testConvLayer(const testConvDesc& pm) {
TestConfig dnnConfig;
getMKLDNNConvConfig(dnnConfig, pm);
for (auto biasSize : {pm.oc, 0}) {
cfg.biasSize = biasSize;
TestConfig ref = cfg;
ref.layerConfig.set_type(compareTypes[1]);
for (auto bs : {pm.bs, 1}) {
tester.run(cfg, ref, bs, pm.ih, pm.iw);
}
dnnConfig.biasSize = biasSize;
RUN_MKLDNN_TEST_LAYER(dnnConfig, "exconv", pm)
}
}
......@@ -143,7 +148,7 @@ TEST(MKLDNNLayer, ConvLayer) {
}
struct testPoolDesc {
int bs, ch; // input channel and output channel are the same
int bs, ic; // input channel and output channel are the same
int ih, iw;
int oh, ow;
int fh, fw;
......@@ -151,19 +156,18 @@ struct testPoolDesc {
int sh, sw;
};
void testPoolLayer(const testPoolDesc& pm) {
const std::string compareTypes[] = {"mkldnn_pool", "pool"};
TestConfig cfg;
cfg.layerConfig.set_type(compareTypes[0]);
cfg.layerConfig.set_size(pm.ch * pm.oh * pm.ow);
static void getMKLDNNPoolConfig(TestConfig& cfg, const testPoolDesc& pm) {
cfg.layerConfig.set_type("mkldnn_pool");
cfg.layerConfig.set_size(pm.ic * pm.oh * pm.ow);
cfg.inputDefs.push_back(
{INPUT_DATA,
"layer_0",
/* size of input layer= */ size_t(pm.ch * pm.ih * pm.iw),
/* size of input layer= */ size_t(pm.ic * pm.ih * pm.iw),
0});
LayerInputConfig* input = cfg.layerConfig.add_inputs();
PoolConfig* pool = input->mutable_pool_conf();
pool->set_channels(pm.ch);
pool->set_pool_type("avg-projection");
pool->set_channels(pm.ic);
pool->set_img_size(pm.iw);
pool->set_img_size_y(pm.ih);
pool->set_output_x(pm.ow);
......@@ -179,20 +183,21 @@ void testPoolLayer(const testPoolDesc& pm) {
int ow = outputSize(pm.iw, pm.fw, pm.pw, pm.sw, false);
CHECK_EQ(ow, pm.ow) << "output size check failed";
CHECK_EQ(oh, pm.oh) << "output size check failed";
}
MKLDNNTester tester;
void testPoolLayer(const testPoolDesc& pm) {
TestConfig dnnConfig;
getMKLDNNPoolConfig(dnnConfig, pm);
LayerInputConfig* input = dnnConfig.layerConfig.mutable_inputs(0);
PoolConfig* pool = input->mutable_pool_conf();
for (auto type : {"max-projection", "avg-projection"}) {
pool->set_pool_type(type);
TestConfig ref = cfg;
ref.layerConfig.set_type(compareTypes[1]);
for (auto bs : {pm.bs, 1}) {
tester.run(cfg, ref, bs, pm.ih, pm.iw);
}
RUN_MKLDNN_TEST_LAYER(dnnConfig, "pool", pm)
}
}
TEST(MKLDNNLayer, PoolLayer) {
/* bs, ch, ih, iw, oh, ow, fh, fw, ph, pw, sh, sw*/
/* bs, ch, ih, iw, oh, ow, fh, fw, ph, pw, sh, sw */
testPoolLayer({2, 1, 4, 4, 2, 2, 3, 3, 0, 0, 2, 2});
testPoolLayer({10, 8, 16, 16, 8, 8, 2, 2, 0, 0, 2, 2});
testPoolLayer({4, 2, 5, 5, 3, 3, 3, 3, 1, 1, 2, 2});
......@@ -204,44 +209,36 @@ TEST(MKLDNNLayer, PoolLayer) {
}
struct testActDesc {
int bs, ch;
int ih, iw;
int bs, ic, ih, iw;
};
static void getAddtoConfig(TestConfig& cfg, const testActDesc& pm) {
cfg.biasSize = 0;
cfg.layerConfig.set_type("addto");
cfg.layerConfig.set_size(pm.ch * pm.ih * pm.iw);
cfg.inputDefs.push_back(
{INPUT_DATA,
"layer_0",
/* size of input layer= */ size_t(pm.ch * pm.ih * pm.iw),
0});
size_t layerSize = pm.ih * pm.ih * pm.iw;
cfg.layerConfig.set_size(layerSize);
cfg.inputDefs.push_back({INPUT_DATA, "layer_0", layerSize, 0});
cfg.layerConfig.add_inputs();
}
void testActivation(std::string& type, const testActDesc& pm) {
const std::string compareTypes[] = {type, type.erase(0, 7)};
void testActivation(std::string& actType, const testActDesc& pm) {
// TODO(TJ): mkldnn_softmax not implemented, paddle do not have elu activation
if (actType == "mkldnn_softmax" || actType == "mkldnn_elu") {
return;
}
const std::string compareTypes[] = {actType, actType.erase(0, 7)};
TestConfig cfg;
getAddtoConfig(cfg, pm);
TestConfig ref = cfg;
cfg.layerConfig.set_active_type(compareTypes[0]);
ref.layerConfig.set_active_type(compareTypes[1]);
MKLDNNTester tester;
for (auto bs : {pm.bs, 1}) {
tester.run(cfg, ref, bs, pm.ih, pm.iw);
}
RUN_MKLDNN_TEST(cfg, ref, pm)
}
TEST(MKLDNNActivation, Activations) {
auto types = MKLDNNActivation::getAllRegisteredTypes();
// TODO(TJ): mkldnn_softmax not implemented, paddle do not have elu activation
std::set<string> excluded{"mkldnn_softmax", "mkldnn_elu"};
for (auto type : types) {
if (excluded.count(type)) {
continue;
}
/* bs, c, h, w*/
testActivation(type, {16, 64, 32, 32});
}
}
......
......@@ -96,3 +96,4 @@ set(GLOB_OP_LIB ${OP_LIBRARY} CACHE INTERNAL "Global OP library")
cc_test(gather_test SRCS gather_test.cc DEPS tensor)
cc_test(net_op_test SRCS net_op_test.cc DEPS net_op)
cc_test(scatter_test SRCS scatter_test.cc DEPS tensor)
cc_test(strided_memcpy_test SRCS strided_memcpy_test.cc DEPS tensor paddle_memory)
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/framework/ddim.h"
#include "paddle/memory/memcpy.h"
#include "paddle/platform/device_context.h"
namespace paddle {
namespace operators {
namespace detail {
template <typename T, int Rank>
struct StridedMemcpyFunctor;
template <typename T>
struct StridedMemcpyFunctor<T, 1> {
void operator()(const platform::DeviceContext& dev_ctx, const T* src,
framework::Dim<1> src_stride, framework::Dim<1> dst_dim,
framework::Dim<1> dst_stride, T* dst) const {
auto place = dev_ctx.GetPlace();
if (platform::is_cpu_place(place)) {
auto& cpu_place = boost::get<platform::CPUPlace>(place);
memory::Copy(cpu_place, dst, cpu_place, src, sizeof(T) * dst_dim.head);
} else {
#ifndef PADDLE_ONLY_CPU
auto& gpu_place = boost::get<platform::GPUPlace>(place);
auto& cuda_ctx =
reinterpret_cast<const platform::CUDADeviceContext&>(dev_ctx);
memory::Copy(gpu_place, dst, gpu_place, src, sizeof(T) * dst_dim.head,
cuda_ctx.stream());
#else
PADDLE_THROW("Paddle is not compiled with GPU");
#endif
}
}
};
template <typename T, int Rank>
struct StridedMemcpyFunctor {
void operator()(const platform::DeviceContext& dev_ctx, const T* src,
framework::Dim<Rank> src_stride, framework::Dim<Rank> dst_dim,
framework::Dim<Rank> dst_stride, T* dst) const {
for (int64_t i = 0; i < dst_dim.head; ++i) {
StridedMemcpyFunctor<T, Rank - 1> func;
func(dev_ctx, src, src_stride.tail, dst_dim.tail, dst_stride.tail, dst);
src += src_stride.head;
dst += dst_stride.head;
}
}
};
template <typename T>
struct StridedCopyDimVisitor : public boost::static_visitor<void> {
StridedCopyDimVisitor(const platform::DeviceContext& dev_ctx, const T* src,
const framework::DDim& src_stride,
const framework::DDim& dst_stride, T* dst)
: dev_ctx_(dev_ctx),
src_(src),
src_stride_(src_stride),
dst_stride_(dst_stride),
dst_(dst) {}
template <typename Dim>
void operator()(Dim dst_dim) const {
Dim src_stride = boost::get<Dim>(src_stride_);
Dim dst_stride = boost::get<Dim>(dst_stride_);
constexpr int dim = Dim::dimensions;
StridedMemcpyFunctor<T, dim> functor;
functor(dev_ctx_, src_, src_stride, dst_dim, dst_stride, dst_);
}
const platform::DeviceContext& dev_ctx_;
const T* src_;
const framework::DDim& src_stride_;
const framework::DDim& dst_stride_;
T* dst_;
};
} // namespace detail
} // namespace operators
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/operators/modified_huber_loss_op.h"
namespace paddle {
namespace operators {
class ModifiedHuberLossOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
protected:
void InferShape(const framework::InferShapeContext& context) const override {
PADDLE_ENFORCE_NOT_NULL(context.InputVar("X"), "X must be initialized.");
PADDLE_ENFORCE_NOT_NULL(context.InputVar("Y"), "Y must be initialized.");
auto* x = context.Input<Tensor>("X");
auto* y = context.Input<Tensor>("Y");
PADDLE_ENFORCE_EQ(x->dims(), y->dims(),
"The shape of X and Y must be the same.");
PADDLE_ENFORCE_EQ(x->dims().size(), 2, "The tensor rank of X must be 2.");
PADDLE_ENFORCE_EQ(x->dims()[1], 1, "The 2nd dimension of X must be 1.");
context.Output<framework::LoDTensor>("IntermediateVal")->Resize(x->dims());
context.Output<framework::LoDTensor>("Out")->Resize({x->dims()[0], 1});
}
};
class ModifiedHuberLossOpMaker : public framework::OpProtoAndCheckerMaker {
public:
ModifiedHuberLossOpMaker(framework::OpProto* proto,
framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("X",
"The input tensor of modified huber loss op."
"X is 2-D tensor with shape [batch_size, 1].");
AddInput("Y",
"The target labels of modified huber loss op."
"The shape of Y is same as X. Values of Y must be 0 or 1.");
AddOutput("IntermediateVal",
"Variable to save intermediate result which will be reused in "
"backward processing.")
.AsIntermediate();
AddOutput("Out", "Classification loss for X.");
AddComment(R"DOC(
Modified huber loss is used in binary classification problem. The shape of
input X and target Y are both [N, 1] and so is the shape of output loss.
Since target Y is not differentiable, cacluating gradient for Y is illegal.
The formulation of modified huber loss is:
L(y, f(x)) = max(0, 1 - yf(x))^2 for yf(x) >= -1,
-4yf(x) otherwise.
Make sure the values of target label Y are in {0, 1} here. The operator will
scale values of Y to {-1, +1} when computing losses and gradients.
)DOC");
}
};
class ModifiedHuberLossGradOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
protected:
void InferShape(const framework::InferShapeContext& context) const override {
auto* x = context.Input<Tensor>("X");
auto* y = context.Input<Tensor>("Y");
auto* intermediate_val = context.Input<Tensor>("IntermediateVal");
auto* out_grad = context.Input<Tensor>(framework::GradVarName("Out"));
auto* x_grad =
context.Output<framework::LoDTensor>(framework::GradVarName("X"));
PADDLE_ENFORCE_NOT_NULL(x, "X must be initialized.");
PADDLE_ENFORCE_NOT_NULL(y, "Y must be initialized.");
PADDLE_ENFORCE_NOT_NULL(intermediate_val,
"Intermediate value must not be null.");
PADDLE_ENFORCE_NOT_NULL(out_grad, "Input(Out@Grad) must not be null.");
PADDLE_ENFORCE_EQ(
intermediate_val->dims(), x->dims(),
"The shape of X and intermediate value must be the same.");
PADDLE_ENFORCE_EQ(out_grad->dims(), x->dims(),
"The shape of Input(Out@Grad) and X must be the same.");
if (x_grad) x_grad->Resize(x->dims());
}
};
} // namespace operators
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(modified_huber_loss, ops::ModifiedHuberLossOp,
ops::ModifiedHuberLossOpMaker, modified_huber_loss_grad,
ops::ModifiedHuberLossGradOp);
REGISTER_OP_CPU_KERNEL(
modified_huber_loss,
ops::ModifiedHuberLossKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(modified_huber_loss_grad,
ops::ModifiedHuberLossGradCPUKernel<float>);
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include <thrust/device_ptr.h>
#include <thrust/device_vector.h>
#include <thrust/for_each.h>
#include <thrust/tuple.h>
#include "paddle/framework/op_registry.h"
#include "paddle/operators/modified_huber_loss_op.h"
#include "paddle/platform/hostdevice.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
struct ModifiedHuberLossBackward {
template <typename Tuple>
HOSTDEVICE void operator()(Tuple t) const {
auto inter_val = thrust::get<1>(t);
auto y_val = thrust::get<2>(t);
auto out_grad = thrust::get<3>(t);
if (inter_val < -1) {
thrust::get<0>(t) = -4 * (2 * y_val - 1) * out_grad;
} else if (inter_val < 1) {
thrust::get<0>(t) = -2 * (1 - inter_val) * (2 * y_val - 1) * out_grad;
} else {
thrust::get<0>(t) = 0;
}
}
};
template <typename T>
class ModifiedHuberLossGradGPUKernel : public framework::OpKernel {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* in0 = context.Input<Tensor>("Y");
auto* in1 = context.Input<Tensor>("IntermediateVal");
auto* in2 = context.Input<Tensor>(framework::GradVarName("Out"));
auto* out0 = context.Output<Tensor>(framework::GradVarName("X"));
if (out0) {
auto counts = framework::product(in1->dims());
auto y_ptr = thrust::device_pointer_cast(in0->data<T>());
auto inter_val_ptr = thrust::device_pointer_cast(in1->data<T>());
auto out_grad_ptr = thrust::device_pointer_cast(in2->data<T>());
thrust::device_ptr<T> x_grad_ptr(
out0->mutable_data<T>(context.GetPlace()));
auto iter_begin = thrust::make_zip_iterator(
thrust::make_tuple(x_grad_ptr, inter_val_ptr, y_ptr, out_grad_ptr));
auto iter_end = thrust::make_zip_iterator(
thrust::make_tuple(x_grad_ptr + counts, inter_val_ptr + counts,
y_ptr + counts, out_grad_ptr + counts));
thrust::for_each(iter_begin, iter_end, ModifiedHuberLossBackward());
}
}
};
} // namespace operators
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP_GPU_KERNEL(
modified_huber_loss,
ops::ModifiedHuberLossKernel<paddle::platform::GPUPlace, float>);
REGISTER_OP_GPU_KERNEL(modified_huber_loss_grad,
ops::ModifiedHuberLossGradGPUKernel<float>);
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/platform/hostdevice.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
template <typename T>
struct CheckLabelValue {
HOSTDEVICE T operator()(const T& val) const {
PADDLE_ASSERT(val == static_cast<T>(0) || val == static_cast<T>(1));
}
};
template <typename T>
struct ModifiedHuberLossForward {
HOSTDEVICE T operator()(const T& val) const {
if (val < -1) {
return -4 * val;
} else if (val < 1) {
return (1 - val) * (1 - val);
} else {
return static_cast<T>(0);
}
}
};
template <typename Place, typename T>
class ModifiedHuberLossKernel : public framework::OpKernel {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* in0 = context.Input<Tensor>("X");
auto* in1 = context.Input<Tensor>("Y");
auto* out0 = context.Output<framework::LoDTensor>("IntermediateVal");
auto* out1 = context.Output<framework::LoDTensor>("Out");
out0->mutable_data<T>(context.GetPlace());
out1->mutable_data<T>(context.GetPlace());
auto place = context.GetEigenDevice<Place>();
auto x = EigenVector<T>::Flatten(*in0);
auto y = EigenVector<T>::Flatten(*in1);
// make sure value's of Y in {0, 1}
y.unaryExpr(CheckLabelValue<T>());
auto inter_val = EigenVector<T>::Flatten(*out0);
// scale y to {-1, +1} and compute x * y
inter_val.device(place) = x * (2 * y - static_cast<T>(1));
auto loss = EigenVector<T>::Flatten(*out1);
loss.device(place) = inter_val.unaryExpr(ModifiedHuberLossForward<T>());
}
};
// CPU backward kernel
template <typename T>
class ModifiedHuberLossGradCPUKernel : public framework::OpKernel {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* in0 = context.Input<Tensor>("Y");
auto* in1 = context.Input<framework::LoDTensor>("IntermediateVal");
auto* in2 =
context.Input<framework::LoDTensor>(framework::GradVarName("Out"));
auto* out0 =
context.Output<framework::LoDTensor>(framework::GradVarName("X"));
if (out0) {
const T* y_ptr = in0->data<T>();
const T* inter_val_ptr = in1->data<T>();
const T* out_grad_ptr = in2->data<T>();
size_t counts = static_cast<size_t>(framework::product(in1->dims()));
T* x_grad_ptr = out0->mutable_data<T>(context.GetPlace());
for (size_t i = 0; i < counts; ++i) {
if (inter_val_ptr[i] < -1) {
x_grad_ptr[i] = -4 * (2 * y_ptr[i] - 1) * out_grad_ptr[i];
} else if (inter_val_ptr[i] < 1) {
x_grad_ptr[i] = -2 * (1 - inter_val_ptr[i]) * (2 * y_ptr[i] - 1) *
out_grad_ptr[i];
} else {
x_grad_ptr[i] = 0;
}
}
}
}
};
} // namespace operators
} // namespace paddle
......@@ -96,7 +96,7 @@ class PReluGradKernel : public framework::OpKernel {
trans(context.device_context(), out_ptr, out_ptr + numel, dout_ptr, dx_ptr,
PReluGradFunctor<T>(alpha_ptr));
// TODO (Zhuoyuan): add dalpha upgrade when GPU kernels ready
// TODO(Zhuoyuan): add dalpha upgrade when GPU kernels ready
}
};
......
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/operators/smooth_l1_loss_op.h"
namespace paddle {
namespace operators {
class SmoothL1LossOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
protected:
void InferShape(const framework::InferShapeContext& ctx) const override {
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "X must be initialized.");
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("Y"), "Y must be initialized.");
auto* x = ctx.Input<framework::Tensor>("X");
auto* y = ctx.Input<framework::Tensor>("Y");
PADDLE_ENFORCE_EQ(x->dims(), y->dims(),
"The shape of X and Y must be the same.");
PADDLE_ENFORCE_GE(x->dims().size(), 2,
"The tensor rank of X must be at least 2.");
auto* inside_weight = ctx.Input<framework::Tensor>("InsideWeight");
if (inside_weight) {
auto* outside_weight = ctx.Input<framework::Tensor>("OutsideWeight");
PADDLE_ENFORCE_NOT_NULL(outside_weight,
"If weights are provided, must specify both "
"inside and outside weights.");
PADDLE_ENFORCE_EQ(inside_weight->dims(), x->dims(),
"The shape of InsideWeight must be same as X.");
PADDLE_ENFORCE_EQ(outside_weight->dims(), x->dims(),
"The shape of OutsideWeight must be same as X.");
}
auto* diff = ctx.Output<framework::LoDTensor>("Diff");
auto* out = ctx.Output<framework::LoDTensor>("Out");
diff->Resize(x->dims());
// loss is a two-rank tensor
out->Resize({x->dims()[0], 1});
}
};
template <typename AttrType>
class SmoothL1LossOpMaker : public framework::OpProtoAndCheckerMaker {
public:
SmoothL1LossOpMaker(framework::OpProto* proto,
framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {
AddInput("X",
"The input tensor of smooth l1 loss op."
"The rank should be greater or equal to 2 with shape "
"[batch_size, value_dim1, value_dim2, ..., value_dimN]");
AddInput("Y",
"The target tensor of smooth l1 loss op "
"with the same shape as X.");
AddInput("InsideWeight",
"Optional input tensor of smooth l1 loss op with the same shape "
"as X. If provided, the result of (X - Y) will be multiplied "
"by this tensor element by element.");
AddInput("OutsideWeight",
"Optinal input of smooth l1 loss op with the same shape as X."
"If provided, the output smooth l1 loss will be multiplied by "
"this tensor element by element.");
AddOutput("Diff", "Intermediate variable to cache InsideWeight*(X-Y).")
.AsIntermediate();
AddOutput("Out", "Smooth l1 loss.");
AddAttr<AttrType>("sigma",
"Hyper parameter of smooth l1 loss op."
"A float scalar with default value 3.0.")
.SetDefault(3.0);
AddComment(R"DOC(
Compute smooth l1 loss for input and target. The operator take the 1st
dimension of input as batch size. For each instance, it will compute
smooth l1 loss element by element first and sum all losses to one value.
So the output shape is [batch_size, 1].
The equation is:
loss = 0.5 * (sigma * (x-y))^2 if abs(x - y) < 1 / sigma^2
abs(x - y) - 0.5 / sigma^2 otherwise
)DOC");
}
};
class SmoothL1LossGradOp : public framework::OperatorWithKernel {
public:
using framework::OperatorWithKernel::OperatorWithKernel;
protected:
void InferShape(const framework::InferShapeContext& ctx) const override {
auto in_dims = ctx.Input<framework::Tensor>("X")->dims();
auto out_dims =
ctx.Input<framework::Tensor>(framework::GradVarName("Out"))->dims();
auto* x_grad =
ctx.Output<framework::LoDTensor>(framework::GradVarName("X"));
auto* y_grad =
ctx.Output<framework::LoDTensor>(framework::GradVarName("Y"));
PADDLE_ENFORCE_GE(out_dims.size(), 2,
"The tensor rank of Input(Out@Grad) should be 2.");
PADDLE_ENFORCE_EQ(out_dims[0], in_dims[0],
"The 1st dimension of Input(Out@Grad) must be "
"same as input.");
PADDLE_ENFORCE_EQ(out_dims[1], 1,
"The 2nd dimension of Input(Out@Grad) must be 1.");
if (x_grad) x_grad->Resize(in_dims);
if (y_grad) y_grad->Resize(in_dims);
}
};
} // namespace operators
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(smooth_l1_loss, ops::SmoothL1LossOp,
ops::SmoothL1LossOpMaker<float>, smooth_l1_loss_grad,
ops::SmoothL1LossGradOp);
REGISTER_OP_CPU_KERNEL(
smooth_l1_loss, ops::SmoothL1LossKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(
smooth_l1_loss_grad,
ops::SmoothL1LossGradKernel<paddle::platform::CPUPlace, float>);
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#define EIGEN_USE_GPU
#include "paddle/operators/smooth_l1_loss_op.h"
namespace ops = paddle::operators;
REGISTER_OP_GPU_KERNEL(
smooth_l1_loss, ops::SmoothL1LossKernel<paddle::platform::GPUPlace, float>);
REGISTER_OP_GPU_KERNEL(
smooth_l1_loss_grad,
ops::SmoothL1LossGradKernel<paddle::platform::GPUPlace, float>);
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/framework/eigen.h"
#include "paddle/framework/op_registry.h"
#include "paddle/platform/hostdevice.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
template <typename T>
struct SmoothL1LossForward {
HOSTDEVICE SmoothL1LossForward(const T& sigma2) : sigma2(sigma2) {}
HOSTDEVICE T operator()(const T& val) const {
T abs_val = std::abs(val);
if (abs_val < 1.0 / sigma2) {
return 0.5 * val * val * sigma2;
} else {
return abs_val - 0.5 / sigma2;
}
}
T sigma2;
};
template <typename Place, typename T, typename AttrType = T>
class SmoothL1LossKernel : public framework::OpKernel {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* in0 = context.Input<Tensor>("X");
auto* in1 = context.Input<Tensor>("Y");
auto* in2 = context.Input<Tensor>("InsideWeight");
auto* in3 = context.Input<Tensor>("OutsideWeight");
auto* out0 = context.Output<Tensor>("Diff");
auto* out1 = context.Output<Tensor>("Out");
out0->mutable_data<T>(context.GetPlace());
out1->mutable_data<T>(context.GetPlace());
auto place = context.GetEigenDevice<Place>();
auto sigma = static_cast<T>(context.Attr<AttrType>("sigma"));
T sigma2 = sigma * sigma;
bool has_weight = (in2 != nullptr) && (in3 != nullptr);
auto x = EigenVector<T>::Flatten(*in0);
auto y = EigenVector<T>::Flatten(*in1);
auto diff = EigenVector<T>::Flatten(*out0);
diff.device(place) = x - y;
// multiply inside weight
if (has_weight) {
auto inside_weight = EigenVector<T>::Flatten(*in2);
// cache diff, reused in bp
diff.device(place) = diff * inside_weight;
}
auto in_counts = in0->numel();
Tensor ptensor_errors;
ptensor_errors.mutable_data<T>({static_cast<int>(in_counts)},
context.GetPlace());
auto errors = EigenVector<T>::Flatten(ptensor_errors);
// apply smooth l1 forward
errors.device(place) = diff.unaryExpr(SmoothL1LossForward<T>(sigma2));
// multiply outside weight
if (has_weight) {
auto outside_weight = EigenVector<T>::Flatten(*in3);
errors.device(place) = errors * outside_weight;
}
auto loss = EigenVector<T>::Flatten(*out1);
// first dimension of 'X' is the number of samples
auto mat_dims =
framework::make_ddim({static_cast<int>(in0->dims()[0]),
static_cast<int>(in_counts / in0->dims()[0])});
auto errors_mat_view = EigenMatrix<T>::From(ptensor_errors, mat_dims);
loss.device(place) = errors_mat_view.sum(Eigen::array<int, 1>({{1}}));
}
};
template <typename T>
struct SmoothL1LossBackward {
HOSTDEVICE SmoothL1LossBackward(const T& sigma2) : sigma2(sigma2) {}
HOSTDEVICE T operator()(const T& val) const {
T abs_val = std::abs(val);
if (abs_val < 1.0 / sigma2) {
return sigma2 * val;
} else {
return (0 < val) - (val < 0);
}
}
T sigma2;
};
template <typename Place, typename T, typename AttrType = T>
class SmoothL1LossGradKernel : public framework::OpKernel {
public:
void Compute(const framework::ExecutionContext& context) const override {
auto* in0 = context.Input<Tensor>("InsideWeight");
auto* in1 = context.Input<Tensor>("OutsideWeight");
auto* in2 = context.Input<Tensor>("Diff");
auto* og = context.Input<Tensor>(framework::GradVarName("Out"));
auto sigma = static_cast<T>(context.Attr<AttrType>("sigma"));
T sigma2 = sigma * sigma;
bool has_weight = (in0 != nullptr) && (in1 != nullptr);
auto place = context.GetEigenDevice<Place>();
auto in_dims = in2->dims();
auto counts = in2->numel();
auto cols = counts / in_dims[0];
auto mat_dims = framework::make_ddim(
{static_cast<int>(in_dims[0]), static_cast<int>(cols)});
Tensor ptensor_diff;
ptensor_diff.mutable_data<T>({static_cast<int>(counts)},
context.GetPlace());
auto diff = EigenVector<T>::Flatten(ptensor_diff);
// apply smooth l1 backwoard
diff.device(place) = EigenVector<T>::Flatten(*in2).unaryExpr(
SmoothL1LossBackward<T>(sigma2));
// compute weights
Tensor ptensor_weights;
ptensor_weights.mutable_data<T>(mat_dims, context.GetPlace());
auto weights = EigenMatrix<T>::From(ptensor_weights);
// initialize to 1.0
weights.device(place) = weights.constant(static_cast<T>(1.0));
if (has_weight) {
auto inside_weight = EigenMatrix<T>::From(*in0, mat_dims);
auto outside_weight = EigenMatrix<T>::From(*in1, mat_dims);
weights.device(place) = inside_weight * outside_weight;
}
// compute gradients
auto out_grad = EigenMatrix<T>::From(*og);
auto diff_mat_view = EigenMatrix<T>::From(ptensor_diff, mat_dims);
auto gradients = out_grad.broadcast(
Eigen::array<int, 2>({{1, static_cast<int>(cols)}})) *
weights * diff_mat_view;
auto* out0 = context.Output<Tensor>(framework::GradVarName("X"));
auto* out1 = context.Output<Tensor>(framework::GradVarName("Y"));
if (out0) {
out0->mutable_data<T>(context.GetPlace());
auto x_grad = EigenMatrix<T>::From(*out0, mat_dims);
x_grad.device(place) = gradients;
}
if (out1) {
out1->mutable_data<T>(context.GetPlace());
auto y_grad = EigenMatrix<T>::From(*out1, mat_dims);
y_grad.device(place) = -1 * gradients;
}
}
};
} // namespace operators
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#pragma once
#include "paddle/operators/detail/strided_memcpy.h"
namespace paddle {
namespace operators {
// Strided memory copy from src to dst.
//
// The src and dst should be both on dev_ctx.GetPlace(), otherwise, there will
// be a segment fault.
//
// The stride of an array (also referred to as increment, pitch or step size) is
// the number of locations in memory between beginnings of successive array
// elements
//
// For example, for tensor like [1, 3, 300, 300]. If there is no padding, the
// stride is [270000, 90000, 300, 1].
//
// NOTE: When use GPU, the memcpy is async. To sync memcpy, please invoke
// `dev_ctx.Wait()`.
template <typename T>
inline void StridedMemcpy(const platform::DeviceContext& dev_ctx, const T* src,
const framework::DDim& src_stride,
const framework::DDim& dst_dim,
const framework::DDim& dst_stride, T* dst) {
using namespace detail;
StridedCopyDimVisitor<T> func(dev_ctx, src, src_stride, dst_stride, dst);
boost::apply_visitor(func, dst_dim);
}
} // namespace operators
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License. */
#include "paddle/operators/strided_memcpy.h"
#include "gtest/gtest.h"
#include "paddle/memory/memory.h"
namespace paddle {
namespace operators {
TEST(StridedMemcpy, CPUCrop) {
// clang-format off
int src[] = {
0, 1, 2, 0, 0,
0, 3, 4, 0, 0,
0, 0, 0, 0, 0,
};
// clang-format on
framework::DDim src_stride({5, 1});
int dst[4];
framework::DDim dst_dim({2, 2});
framework::DDim dst_stride({2, 1});
platform::CPUDeviceContext ctx;
StridedMemcpy<int>(ctx, src + 1, src_stride, dst_dim, dst_stride, dst);
ASSERT_EQ(1, dst[0]);
ASSERT_EQ(2, dst[1]);
ASSERT_EQ(3, dst[2]);
ASSERT_EQ(4, dst[3]);
}
TEST(StridedMemcpy, CPUConcat) {
// clang-format off
int src[] = {
1, 2,
3, 4
};
// clang-format on
int dst[8];
framework::DDim src_stride({2, 1});
framework::DDim dst_dim({2, 2});
framework::DDim dst_stride({4, 1});
platform::CPUDeviceContext ctx;
StridedMemcpy<int>(ctx, src, src_stride, dst_dim, dst_stride, dst);
StridedMemcpy<int>(ctx, src, src_stride, dst_dim, dst_stride, dst + 2);
// clang-format off
int expect_dst[] = {
1, 2, 1, 2,
3, 4, 3, 4
};
// clang-format on
for (size_t i = 0; i < sizeof(expect_dst) / sizeof(int); ++i) {
ASSERT_EQ(expect_dst[i], dst[i]);
}
}
#ifndef PADDLE_ONLY_CPU
TEST(StridedMemcpy, GPUCrop) {
// clang-format off
int src[] = {
0, 1, 2, 0, 0,
0, 3, 4, 0, 0,
0, 0, 0, 0, 0,
};
// clang-format on
platform::GPUPlace gpu0(0);
platform::CPUPlace cpu;
int* gpu_src = reinterpret_cast<int*>(memory::Alloc(gpu0, sizeof(src)));
memory::Copy(gpu0, gpu_src, cpu, src, sizeof(src));
framework::DDim src_stride({5, 1});
int dst[4];
int* gpu_dst = reinterpret_cast<int*>(memory::Alloc(gpu0, sizeof(dst)));
framework::DDim dst_dim({2, 2});
framework::DDim dst_stride({2, 1});
platform::CUDADeviceContext ctx(gpu0);
StridedMemcpy<int>(ctx, gpu_src + 1, src_stride, dst_dim, dst_stride,
gpu_dst);
memory::Copy(cpu, dst, gpu0, gpu_dst, sizeof(dst), ctx.stream());
ctx.Wait();
ASSERT_EQ(1, dst[0]);
ASSERT_EQ(2, dst[1]);
ASSERT_EQ(3, dst[2]);
ASSERT_EQ(4, dst[3]);
memory::Free(gpu0, gpu_dst);
memory::Free(gpu0, gpu_src);
}
TEST(StridedMemcpy, GPUConcat) {
// clang-format off
int src[] = {
1, 2,
3, 4
};
// clang-format on
platform::GPUPlace gpu0(0);
platform::CPUPlace cpu;
int* gpu_src = reinterpret_cast<int*>(memory::Alloc(gpu0, sizeof(src)));
memory::Copy(gpu0, gpu_src, cpu, src, sizeof(src));
int dst[8];
int* gpu_dst = reinterpret_cast<int*>(memory::Alloc(gpu0, sizeof(dst)));
framework::DDim src_stride({2, 1});
framework::DDim dst_dim({2, 2});
framework::DDim dst_stride({4, 1});
platform::CUDADeviceContext ctx(gpu0);
StridedMemcpy<int>(ctx, gpu_src, src_stride, dst_dim, dst_stride, gpu_dst);
StridedMemcpy<int>(ctx, gpu_src, src_stride, dst_dim, dst_stride,
gpu_dst + 2);
memory::Copy(cpu, dst, gpu0, gpu_dst, sizeof(dst), ctx.stream());
ctx.Wait();
// clang-format off
int expect_dst[] = {
1, 2, 1, 2,
3, 4, 3, 4
};
// clang-format on
for (size_t i = 0; i < sizeof(expect_dst) / sizeof(int); ++i) {
ASSERT_EQ(expect_dst[i], dst[i]);
}
memory::Free(gpu0, gpu_dst);
memory::Free(gpu0, gpu_src);
}
#endif
} // namespace operators
} // namespace paddle
\ No newline at end of file
......@@ -1565,6 +1565,10 @@ class LayerBase(object):
self.config = g_config.model_config.layers.add()
assert isinstance(self.config, LayerConfig)
use_mkldnn = bool(int(g_command_config_args.get("use_mkldnn", 0)))
mkldnn_acts = ['relu', 'tanh']
if use_mkldnn and active_type in mkldnn_acts:
active_type = "mkldnn_" + active_type
self.config.name = name
self.config.type = type
self.config.active_type = active_type
......
import unittest
import numpy as np
from op_test import OpTest
def modified_huber_loss_forward(val):
if val < -1:
return -4 * val
elif val < 1:
return (1 - val) * (1 - val)
else:
return 0
class TestModifiedHuberLossOp(OpTest):
def setUp(self):
self.op_type = 'modified_huber_loss'
samples_num = 32
self.inputs = {
'X': np.random.uniform(-1, 1., (samples_num, 1)).astype('float32'),
'Y': np.random.choice([0, 1], samples_num).reshape((samples_num, 1))
}
product_res = self.inputs['X'] * (2 * self.inputs['Y'] - 1)
loss = np.vectorize(modified_huber_loss_forward)(product_res)
self.outputs = {
'IntermediateVal': product_res,
'Out': loss.reshape((samples_num, 1))
}
def test_check_output(self):
self.check_output()
def test_check_grad(self):
self.check_grad(['X'], 'Out', max_relative_error=0.005)
if __name__ == '__main__':
unittest.main()
import unittest
import numpy as np
from op_test import OpTest
def smooth_l1_loss_forward(val, sigma2):
abs_val = abs(val)
if abs_val < 1.0 / sigma2:
return 0.5 * val * val * sigma2
else:
return abs_val - 0.5 / sigma2
class TestSmoothL1LossOp1(OpTest):
def setUp(self):
self.op_type = "smooth_l1_loss"
dims = (5, 10)
self.inputs = {
'X': np.random.random(dims).astype("float32"),
'Y': np.random.random(dims).astype("float32")
}
sigma = 3.0
self.attrs = {'sigma': sigma}
sigma2 = sigma * sigma
diff = self.inputs['X'] - self.inputs['Y']
loss = np.vectorize(smooth_l1_loss_forward)(diff, sigma2).sum(1)
loss = loss.reshape((dims[0], 1))
self.outputs = {'Diff': diff, 'Out': loss}
def test_check_output(self):
self.check_output()
def test_check_grad_normal(self):
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.02)
def test_check_grad_ingore_x(self):
self.check_grad(
['Y'], 'Out', max_relative_error=0.03, no_grad_set=set("X"))
def test_check_grad_ingore_y(self):
self.check_grad(
['X'], 'Out', max_relative_error=0.03, no_grad_set=set('Y'))
class TestSmoothL1LossOp2(OpTest):
def setUp(self):
self.op_type = "smooth_l1_loss"
dims = (5, 10)
self.inputs = {
'X': np.random.random(dims).astype("float32"),
'Y': np.random.random(dims).astype("float32"),
'InsideWeight': np.random.random(dims).astype("float32"),
'OutsideWeight': np.random.random(dims).astype("float32")
}
sigma = 3.0
self.attrs = {'sigma': sigma}
sigma2 = sigma * sigma
diff = self.inputs['X'] - self.inputs['Y']
diff = diff * self.inputs['InsideWeight']
loss = np.vectorize(smooth_l1_loss_forward)(diff, sigma2)
loss = loss * self.inputs['OutsideWeight']
loss = loss.sum(1).reshape((dims[0], 1))
self.outputs = {'Diff': diff, 'Out': loss}
def test_check_output(self):
self.check_output()
def test_check_grad_normal(self):
self.check_grad(['X', 'Y'], 'Out', max_relative_error=0.03)
def test_check_grad_ingore_x(self):
self.check_grad(
['Y'],
'Out',
max_relative_error=0.03,
no_grad_set=set(['X', 'InsideWeight', 'OutsideWeight']))
def test_check_grad_ingore_y(self):
self.check_grad(
['X'],
'Out',
max_relative_error=0.03,
no_grad_set=set(['Y', 'InsideWeight', 'OutsideWeight']))
if __name__ == '__main__':
unittest.main()
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