Skip to content

F
FluidDoc

PaddlePaddle / FluidDoc

通知 10
Star 2
Fork 0
F
FluidDoc
未验证 提交 521d094c 编写于 作者: A Aurelius84 提交者： GitHub
浏览文件
操作

Update related to paddle.nn.data (#2427) 

* modify nn.data into static.data

* Update index_cn.rst

* nn.data to static.data

* remove alias
上级 47018bc6
隐藏空白更改
内联 并排
Showing with 116 addition and 121 deletion
doc/fluid/api_cn/fluid_cn/data_cn.rst
浏览文件 @ 521d094c
......@@ -6,10 +6,6 @@ data
.. py:function:: paddle.fluid.data(name, shape, dtype='float32', lod_level=0)
:api_attr: 声明式编程模式（静态图)
:alias_main: paddle.nn.data
:alias: paddle.nn.data,paddle.nn.input.data
:old_api: paddle.fluid.data
......
doc/fluid/beginners_guide/dygraph/DyGraph.md
浏览文件 @ 521d094c
......@@ -55,8 +55,7 @@ import paddle
from paddle.imperative import to_variable
data = np.ones([2, 2], np.float32)
#x = paddle.data(name='x', shape=[2,2], dtype='float32')
x = paddle.nn.data(name='x', shape=[2,2], dtype='float32')
x = paddle.static.data(name='x', shape=[2,2], dtype='float32')
x += 10
exe = paddle.Executor()
exe.run(paddle.default_startup_program())
......@@ -67,7 +66,7 @@ print("result", out) #[[11, 11], [11, 11]]
paddle.enable_imperative()
x = paddle.imperative.to_variable(data)
x += 10
print('result', x.numpy()) #[[11, 11], [11, 11]]
print('result', x.numpy()) #[[11, 11], [11, 11]]
```
* 命令式编程下，所有操作在运行时就已经完成，更接近我们平时的编程方式，可以随时获取每一个操作的执行结果。
......@@ -152,7 +151,7 @@ class SimpleImgConvPool(paddle.nn.Layer):
param_attr=None,
bias_attr=None):
super(SimpleImgConvPool, self).__init__()
self._conv2d = Conv2D(
num_channels=num_channels,
num_filters=num_filters,
......@@ -165,7 +164,7 @@ class SimpleImgConvPool(paddle.nn.Layer):
bias_attr=None,
act=act,
use_cudnn=use_cudnn)
self._pool2d = Pool2D(
pool_size=pool_size,
pool_type=pool_type,
......@@ -203,12 +202,12 @@ class MNIST(paddle.nn.Layer):
1, 20, 5, 2, 2, act="relu")
self._simple_img_conv_pool_2 = SimpleImgConvPool(
20, 50, 5, 2, 2, act="relu")
self.pool_2_shape = 50 * 4 * 4
SIZE = 10
self.output_weight = self.create_parameter(
[self.pool_2_shape, 10])
def forward(self, inputs, label=None):
x = self._simple_img_conv_pool_1(inputs)
x = self._simple_img_conv_pool_2(x)
......@@ -275,25 +274,25 @@ adam = AdamOptimizer(learning_rate=0.001, parameter_list=mnist.parameters())
epoch_num = 5
for epoch in range(epoch_num):
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0].reshape(1, 28, 28) for x in data]).astype('float32')
y_data = np.array([x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
cost, acc = mnist(img, label)
loss = paddle.nn.functional.cross_entropy(cost, label)
avg_loss = paddle.mean(loss)
avg_loss.backward()
adam.minimize(avg_loss)
mnist.clear_gradients()
if batch_id % 100 == 0:
print("Loss at epoch {} step {}: {:}".format(
epoch, batch_id, avg_loss.numpy()))
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0].reshape(1, 28, 28) for x in data]).astype('float32')
y_data = np.array([x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
cost, acc = mnist(img, label)
loss = paddle.nn.functional.cross_entropy(cost, label)
avg_loss = paddle.mean(loss)
avg_loss.backward()
adam.minimize(avg_loss)
mnist.clear_gradients()
if batch_id % 100 == 0:
print("Loss at epoch {} step {}: {:}".format(
epoch, batch_id, avg_loss.numpy()))
model_dict = mnist.state_dict()
paddle.imperative.save(model_dict, "save_temp")
```
......@@ -307,7 +306,7 @@ paddle.imperative.save(model_dict, "save_temp")
model.eval() #切换到评估模式
model.train() #切换到训练模式
```
模型评估测试的实现如下：
* 首先定义 MNIST 类的对象 mnist_eval，然后通过 [load_dygraph](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/load_dygraph_cn.html#load-dygraph) 接口加载保存好的模型参数，通过 [Layer](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Layer_cn.html#layer) 的 [set_dict](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Layer_cn.html#set_dict) 接口将参数导入到模型中，通过 [Layer](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Layer_cn.html#layer) 的 eval 接口切换到预测评估模式。
......@@ -316,7 +315,7 @@ model.train() #切换到训练模式
```python
paddle.enable_imperative()
mnist_eval = MNIST()
mnist_eval = MNIST()
model_dict, _ = paddle.imperative.load("save_temp")
mnist_eval.set_dict(model_dict)
print("checkpoint loaded")
......@@ -326,21 +325,21 @@ mnist_eval.eval()
acc_set = []
avg_loss_set = []
for batch_id, data in enumerate(test_reader()):
dy_x_data = np.array([x[0].reshape(1, 28, 28)
for x in data]).astype('float32')
y_data = np.array(
[x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
prediction, acc = mnist_eval(img, label)
loss = paddle.nn.functional.cross_entropy(input=prediction, label=label)
avg_loss = paddle.mean(loss)
acc_set.append(float(acc.numpy()))
avg_loss_set.append(float(avg_loss.numpy()))
dy_x_data = np.array([x[0].reshape(1, 28, 28)
for x in data]).astype('float32')
y_data = np.array(
[x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
prediction, acc = mnist_eval(img, label)
loss = paddle.nn.functional.cross_entropy(input=prediction, label=label)
avg_loss = paddle.mean(loss)
acc_set.append(float(acc.numpy()))
avg_loss_set.append(float(avg_loss.numpy()))
acc_val_mean = np.array(acc_set).mean()
avg_loss_val_mean = np.array(avg_loss_set).mean()
print("Eval avg_loss is: {}, acc is: {}".format(avg_loss_val_mean, acc_val_mean))
......@@ -351,9 +350,9 @@ print("Eval avg_loss is: {}, acc is: {}".format(avg_loss_val_mean, acc_val_mean)
在命令式编程下，模型和优化器在不同的模块中，所以模型和优化器分别在不同的对象中存储，使得模型参数和优化器信息需分别存储。
因此模型的保存需要单独调用模型和优化器中的 state_dict() 接口，同样模型的加载也需要单独进行处理。
保存模型 ：
保存模型 ：
1. 保存模型参数：首先通过 minist.state_dict 函数获取 mnist 网络的所有参数，然后通过 paddle.imperative.save 函数将获得的参数保存至以 save_path 为前缀的文件中。
1. 保存优化器信息：首先通过 adam.state_dict 函数获取 adam 优化器的信息，然后通过 paddle.imperative.save 函数将获得的参数保存至以 save_path 为前缀的文件中。
1. 保存优化器信息：首先通过 adam.state_dict 函数获取 adam 优化器的信息，然后通过 paddle.imperative.save 函数将获得的参数保存至以 save_path 为前缀的文件中。
* [Layer](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Layer_cn.html#layer) 的 [state_dict](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/Layer_cn.html#state_dict) 接口：该接口可以获取当前层及其子层的所有参数，并将参数存放在 dict 结构中。
* [Optimizer](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/optimizer_cn/AdamOptimizer_cn.html#adamoptimizer) 的 state_dict 接口：该接口可以获取优化器的信息，并将信息存放在 dict 结构中。其中包含优化器使用的所有变量，例如对于 Adam 优化器，包括 beta1、beta2、momentum 等信息。注意如果该优化器的 minimize 函数没有被调用过，则优化器的信息为空。
* [paddle.imperative.save](https://www.paddlepaddle.org.cn/documentation/docs/zh/api_cn/dygraph_cn/save_dygraph_cn.html#save-dygraph) 接口：该接口将传入的参数或优化器的 dict 保存到磁盘上。
......@@ -363,7 +362,7 @@ print("Eval avg_loss is: {}, acc is: {}".format(avg_loss_val_mean, acc_val_mean)
# 保存优化器信息
2. paddle.imperative.save(adam.state_dict(), "save_path")
```
加载模型：
加载模型：
1. 通过 paddle.imperative.load 函数获取模型参数信息 model_state 和优化器信息 opt_state；
1. 通过 mnist.set_dict 函数用获取的模型参数信息设置 mnist 网络的参数
1. 通过 adam.set_dict 函数用获取的优化器信息设置 adam 优化器信息。
......@@ -406,35 +405,35 @@ adam = AdamOptimizer(learning_rate=0.001, parameter_list=mnist.parameters())
mnist = paddle.imperative.DataParallel(mnist, strategy)
train_reader = paddle.batch(
paddle.dataset.mnist.train(), batch_size=BATCH_SIZE, drop_last=True)
paddle.dataset.mnist.train(), batch_size=BATCH_SIZE, drop_last=True)
train_reader = paddle.incubate.reader.distributed_batch_reader(
train_reader)
train_reader)
for epoch in range(epoch_num):
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0].reshape(1, 28, 28)
for x in data]).astype('float32')
y_data = np.array(
[x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
label.stop_gradient = True
cost, acc = mnist(img, label)
loss = paddle.nn.functional.cross_entropy(cost, label)
avg_loss = paddle.mean(loss)
avg_loss = mnist.scale_loss(avg_loss)
avg_loss.backward()
mnist.apply_collective_grads()
adam.minimize(avg_loss)
mnist.clear_gradients()
if batch_id % 100 == 0 and batch_id is not 0:
print("epoch: {}, batch_id: {}, loss is: {}".format(epoch, batch_id, avg_loss.numpy()))
for batch_id, data in enumerate(train_reader()):
dy_x_data = np.array([x[0].reshape(1, 28, 28)
for x in data]).astype('float32')
y_data = np.array(
[x[1] for x in data]).astype('int64').reshape(-1, 1)
img = to_variable(dy_x_data)
label = to_variable(y_data)
label.stop_gradient = True
cost, acc = mnist(img, label)
loss = paddle.nn.functional.cross_entropy(cost, label)
avg_loss = paddle.mean(loss)
avg_loss = mnist.scale_loss(avg_loss)
avg_loss.backward()
mnist.apply_collective_grads()
adam.minimize(avg_loss)
mnist.clear_gradients()
if batch_id % 100 == 0 and batch_id is not 0:
print("epoch: {}, batch_id: {}, loss is: {}".format(epoch, batch_id, avg_loss.numpy()))
if paddle.imperative.ParallelEnv().local_rank == 0:
paddle.imperative.save(mnist.state_dict(), "work_0")
......@@ -477,7 +476,7 @@ trainers_endpoints: 127.0.0.1:6170,127.0.0.1:6171 , node_id: 0 , current_node_ip
总结一下，多卡训练相比单卡训练，有如下步骤不同：
1. 通过 ParallelEnv() 的 dev_id 设置程序运行的设备。
```
place = paddle.CUDAPlace(paddle.imperative.ParallelEnv().dev_id)
place = paddle.CUDAPlace(paddle.imperative.ParallelEnv().dev_id)
paddle.enable_imperative(place):
```
2. 准备多卡环境。
......@@ -511,7 +510,7 @@ mnist.apply_collective_grads()
和单卡不同，多卡训练时需逐个进程执行保存操作，多个进程同时保存会使模型文件格式出错。
```
if paddle.imperative.ParallelEnv().local_rank == 0：
paddle.imperative.save(mnist.state_dict(), "worker_0")
paddle.imperative.save(mnist.state_dict(), "worker_0")
```
7. 评估测试。
......@@ -532,18 +531,18 @@ if paddle.imperative.ParallelEnv().local_rank == 0：
```python
from paddle.imperative import TracedLayer
paddle.enable_imperative()
# 定义MNIST类的对象
mnist = MNIST()
in_np = np.random.random([10, 1, 28, 28]).astype('float32')
# 将numpy的ndarray类型的数据转换为Variable类型
input_var = paddle.imperative.to_variable(in_np)
# 通过 TracerLayer.trace 接口将命令式模型转换为声明式模型
out_dygraph, static_layer = TracedLayer.trace(mnist, inputs=[input_var])
save_dirname = './saved_infer_model'
# 将转换后的模型保存
from paddle.imperative import TracedLayer
paddle.enable_imperative()
# 定义MNIST类的对象
mnist = MNIST()
in_np = np.random.random([10, 1, 28, 28]).astype('float32')
# 将numpy的ndarray类型的数据转换为Variable类型
input_var = paddle.imperative.to_variable(in_np)
# 通过 TracerLayer.trace 接口将命令式模型转换为声明式模型
out_dygraph, static_layer = TracedLayer.trace(mnist, inputs=[input_var])
save_dirname = './saved_infer_model'
# 将转换后的模型保存
static_layer.save_inference_model(save_dirname, feed=[0], fetch=[0])
```
......@@ -573,9 +572,9 @@ in_np = np.array([-2]).astype('int')
input_var = paddle.imperative.to_variable(in_np)
# if判断与输入input_var的shape有关
if input_var.shape[0] > 1:
print("input_var's shape[0] > 1")
print("input_var's shape[0] > 1")
else:
print("input_var's shape[1] < 1")
print("input_var's shape[1] < 1")
```
* 针对依赖数据的控制流，解决流程如下 1. 添加declarative装饰器； 2. 利用ProgramTranslator进行转换
......@@ -584,10 +583,10 @@ else:
首先需要对给MNist类的forward函数添加一个declarative 装饰器，来标记需要转换的代码块，（注：需要在最外层的class的forward函数中添加）
```python
from paddle.imperative import declarative
# 定义MNIST网络，必须继承自paddle.nn.Layer
# 该网络由两个SimpleImgConvPool子网络、reshape层、matmul层、softmax层、accuracy层组成
from paddle.imperative import declarative
# 定义MNIST网络，必须继承自paddle.nn.Layer
# 该网络由两个SimpleImgConvPool子网络、reshape层、matmul层、softmax层、accuracy层组成
class MNIST(paddle.nn.Layer):
def __init__(self):
super(MNIST, self).__init__()
......@@ -595,13 +594,13 @@ class MNIST(paddle.nn.Layer):
1, 20, 5, 2, 2, act="relu")
self._simple_img_conv_pool_2 = SimpleImgConvPool(
20, 50, 5, 2, 2, act="relu")
self.pool_2_shape = 50 * 4 * 4
SIZE = 10
self.output_weight = self.create_parameter(
[self.pool_2_shape, 10])
@declarative
@declarative
def forward(self, inputs, label=None):
x = self._simple_img_conv_pool_1(inputs)
x = self._simple_img_conv_pool_2(x)
......@@ -612,8 +611,8 @@ class MNIST(paddle.nn.Layer):
acc = paddle.metric.accuracy(input=x, label=label)
return x, acc
else:
return x
return x
```
......@@ -622,19 +621,19 @@ class MNIST(paddle.nn.Layer):
```python
import paddle
paddle.enable_imperative()
prog_trans = paddle.imperative.ProgramTranslator()
mnist = MNIST()
in_np = np.random.random([10, 1, 28, 28]).astype('float32')
label_np = np.random.randint(0, 10, size=(10,1)).astype( "int64")
input_var = paddle.imperative.to_variable(in_np)
label_var = paddle.imperative.to_variable(label_np)
out = mnist( input_var, label_var)
import paddle
paddle.enable_imperative()
prog_trans = paddle.imperative.ProgramTranslator()
mnist = MNIST()
in_np = np.random.random([10, 1, 28, 28]).astype('float32')
label_np = np.random.randint(0, 10, size=(10,1)).astype( "int64")
input_var = paddle.imperative.to_variable(in_np)
label_var = paddle.imperative.to_variable(label_np)
out = mnist( input_var, label_var)
prog_trans.save_inference_model("./mnist_dy2stat", fetch=[0,1])
```
......@@ -654,13 +653,13 @@ class MNIST(paddle.nn.Layer):
1, 20, 5, 2, 2, act="relu")
self._simple_img_conv_pool_2 = SimpleImgConvPool(
20, 50, 5, 2, 2, act="relu")
self.pool_2_shape = 50 * 4 * 4
SIZE = 10
self.output_weight = self.create_parameter(
[self.pool_2_shape, 10])
@declarative
@declarative
def forward(self, inputs, label=None):
x = self._simple_img_conv_pool_1(inputs)
x = self._simple_img_conv_pool_2(x)
......@@ -672,7 +671,7 @@ class MNIST(paddle.nn.Layer):
return x, acc
else:
return x
```
......@@ -685,7 +684,7 @@ class MNIST(paddle.nn.Layer):
```
x = y * 10
print(x.numpy())
print(x.numpy())
```
来直接打印变量的值
......
doc/fluid/beginners_guide/index_cn.rst
浏览文件 @ 521d094c
......@@ -34,8 +34,8 @@
import numpy
import paddle
# 定义输入数据占位符
a = paddle.nn.data(name="a", shape=[1], dtype='int64')
b = paddle.nn.data(name="b", shape=[1], dtype='int64')
a = paddle.static.data(name="a", shape=[1], dtype='int64')
b = paddle.static.data(name="b", shape=[1], dtype='int64')
# 组建网络（此处网络仅由一个操作构成，即elementwise_add）
result = paddle.elementwise_add(a, b)
# 准备运行网络
......
Markdown is supported
0% 或 .
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 或