fluid 如何获取特定层的参数

• 问题描述：在训练好一个模型之后，想打印模型的参数类似TF里面，w = session.run([fc1.W])
  然后直接获取到了第一层FClayer的W矩阵，fluid有类似的接口吗？

• 报错输出：

Traceback (most recent call last):
  File "test.py", line 12, in <module>
    ret = scope.find_var("target_fc").get_tensor()
AttributeError: 'NoneType' object has no attribute 'get_tensor'

• 问题复现：

import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
x = fluid.layers.data(name='x', shape=[5], dtype='float32'),                          
param_attr = ParamAttr(name='target_fc')
bias_attr = ParamAttr(name='target_fc_b')
fc1 = fluid.layers.fc(input=x, size=128, act='relu', param_attr=param_attr, bias_attr=bias_attr)
scope = fluid.executor.global_scope()

# get tensor
ret = scope.find_var("target_fc").get_tensor()
print(ret)            
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())

• 问题分析：

在PaddlePaddle中，program程序没有运行是，数据是不会进行初始化的，所以要将获得数据的操作diam放在运行逻辑之后，具体而言就是将ret = scope.find_var("target_fc").get_tensor()移动到exe.run(fluid.default_startup_program())之后

• 解决方法：

import paddle.fluid as fluid
from paddle.fluid.param_attr import ParamAttr
x = fluid.layers.data(name='x', shape=[5], dtype='float32'),                          
param_attr = ParamAttr(name='target_fc')
bias_attr = ParamAttr(name='target_fc_b')
fc1 = fluid.layers.fc(input=x, size=128, act='relu', param_attr=param_attr, bias_attr=bias_attr)

# get tensor
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
ret = fluid.global_scope().find_var("target_fc").get_tensor()
import numpy as np
ret = np.array(ret)
print (ret.shape)
print(ret)