飞桨百度架构师手把手带你零基础实践深度学习21天学习笔记——使用飞桨重写波士顿房价预测模型

使用飞桨书写模型的时候，流程与前面介绍的是一摸一样的。
下面开始总结一下如何使用飞桨写模型

加载飞桨库

主库：paddle/fluid
动态图类库：digraph（便于调试）

静态图模式（声明式编程范式，类比C++）	动态图模式（命令式编程范式，类比Python）
先编译后执行的方式。用户需预先定义完整的网络结构，再对网络结构进行编译优化后，才能执行获得计算结果。	解析式的执行方式。用户无需预先定义完整的网络结构，每写一行网络代码，即可同时获得计算结果。

全连接层：Linear（可激活指定性函数）

数据处理

此部分包括数据导入、修改格式、分为训练测试集、数据归一化处理。
在飞桨中书写与前面完全一致。

模型设计

实质为定义线性回归的结构。
需要用到两种函数，init函数和forward函数。

init	forward
在类的初始化函数中声明每一层网络的实现函数。	构建神经网络结构，实现前向计算过程，并返回预测结果。

注：我们这里不使用激活函数

训练配置

定义工作环境
定义模型（有训练和预测状态）

模型实例有两种状态：训练状态.train()和预测状态.eval()。训练时要执行正向计算和反向传播梯度两个过程，而预测时只需要执行正向计算。为模型指定运行状态，有两点原因：
（1）部分高级的算子（例如Drop out和Batch Normalization，在计算机视觉的章节会详细介绍）在两个状态执行的逻辑不同。
（2）从性能和存储空间的考虑，预测状态时更节省内存，性能更好。
开启训练模式
读取数据
定义优化算法SDG，定义学习率0.01
开始训练
双层循环+四个步骤
（先将数据格式写为numpy.array格式）
保存模型
先保存，加载，再进行测试。因为模型可以应用到不同的场景。
加载模型，进行预测
（1）生产模型实例，配置资源。
（2）加载模型参数，设置模型状态为“eval”（因为预测时，不需要反向传播梯度）
（3）调用模型，打印预测值与真实值

总结：编写成本大降低

感悟：飞桨提供了一个深度学习框架，可以达到事半功倍的效果。学习这一章之后其实我能够发现代码编写过程大大的减少了，但是飞桨的一些函数的定义与调用还没有特别的清楚，在init函数和forward函数的定义上还是不够理解，需要在接下来的学习过程中，去更好的理解和了解。

下面书写一下使用飞桨重写波士顿房价预测模型
直接使用python的案例在这里：完整步骤

#加载飞桨、Numpy和相关类库
import paddle
import paddle.fluid as fluid
import paddle.fluid.dygraph as dygraph
from paddle.fluid.dygraph import Linear
import numpy as np
import os
import random
#数据处理，此部分与前面的完全一致，飞桨的优化主要在模型设计和训练配置上

def load_data():

# 从文件导入数据

datafile = './work/housing.data'

data = np.fromfile(datafile, sep=' ')
<span class="token comment"># 每条数据包括14项，其中前面13项是影响因素，第14项是相应的房屋价格中位数</span>
feature_names <span class="token operator">=</span> <span class="token punctuation">[</span> <span class="token string">'CRIM'</span><span class="token punctuation">,</span> <span class="token string">'ZN'</span><span class="token punctuation">,</span> <span class="token string">'INDUS'</span><span class="token punctuation">,</span> <span class="token string">'CHAS'</span><span class="token punctuation">,</span> <span class="token string">'NOX'</span><span class="token punctuation">,</span> <span class="token string">'RM'</span><span class="token punctuation">,</span> <span class="token string">'AGE'</span><span class="token punctuation">,</span> \
                  <span class="token string">'DIS'</span><span class="token punctuation">,</span> <span class="token string">'RAD'</span><span class="token punctuation">,</span> <span class="token string">'TAX'</span><span class="token punctuation">,</span> <span class="token string">'PTRATIO'</span><span class="token punctuation">,</span> <span class="token string">'B'</span><span class="token punctuation">,</span> <span class="token string">'LSTAT'</span><span class="token punctuation">,</span> <span class="token string">'MEDV'</span> <span class="token punctuation">]</span>
feature_num <span class="token operator">=</span> <span class="token builtin">len</span><span class="token punctuation">(</span>feature_names<span class="token punctuation">)</span>

<span class="token comment"># 将原始数据进行Reshape，变成[N, 14]这样的形状</span>
data <span class="token operator">=</span> data<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span><span class="token punctuation">[</span>data<span class="token punctuation">.</span>shape<span class="token punctuation">[</span><span class="token number">0</span><span class="token punctuation">]</span> <span class="token operator">//</span> feature_num<span class="token punctuation">,</span> feature_num<span class="token punctuation">]</span><span class="token punctuation">)</span>

<span class="token comment"># 将原数据集拆分成训练集和测试集</span>
<span class="token comment"># 这里使用80%的数据做训练，20%的数据做测试</span>
<span class="token comment"># 测试集和训练集必须是没有交集的</span>
ratio <span class="token operator">=</span> <span class="token number">0.8</span>
offset <span class="token operator">=</span> <span class="token builtin">int</span><span class="token punctuation">(</span>data<span class="token punctuation">.</span>shape<span class="token punctuation">[</span><span class="token number">0</span><span class="token punctuation">]</span> <span class="token operator">*</span> ratio<span class="token punctuation">)</span>
training_data <span class="token operator">=</span> data<span class="token punctuation">[</span><span class="token punctuation">:</span>offset<span class="token punctuation">]</span>

<span class="token comment"># 计算train数据集的最大值，最小值，平均值</span>
maximums<span class="token punctuation">,</span> minimums<span class="token punctuation">,</span> avgs <span class="token operator">=</span> training_data<span class="token punctuation">.</span><span class="token builtin">max</span><span class="token punctuation">(</span>axis<span class="token operator">=</span><span class="token number">0</span><span class="token punctuation">)</span><span class="token punctuation">,</span> training_data<span class="token punctuation">.</span><span class="token builtin">min</span><span class="token punctuation">(</span>axis<span class="token operator">=</span><span class="token number">0</span><span class="token punctuation">)</span><span class="token punctuation">,</span> \
                             training_data<span class="token punctuation">.</span><span class="token builtin">sum</span><span class="token punctuation">(</span>axis<span class="token operator">=</span><span class="token number">0</span><span class="token punctuation">)</span> <span class="token operator">/</span> training_data<span class="token punctuation">.</span>shape<span class="token punctuation">[</span><span class="token number">0</span><span class="token punctuation">]</span>

<span class="token comment"># 记录数据的归一化参数，在预测时对数据做归一化</span>
<span class="token keyword">global</span> max_values
<span class="token keyword">global</span> min_values
<span class="token keyword">global</span> avg_values
max_values <span class="token operator">=</span> maximums
min_values <span class="token operator">=</span> minimums
avg_values <span class="token operator">=</span> avgs

<span class="token comment"># 对数据进行归一化处理</span>
<span class="token keyword">for</span> i <span class="token keyword">in</span> <span class="token builtin">range</span><span class="token punctuation">(</span>feature_num<span class="token punctuation">)</span><span class="token punctuation">:</span>
    <span class="token comment">#print(maximums[i], minimums[i], avgs[i])</span>
    data<span class="token punctuation">[</span><span class="token punctuation">:</span><span class="token punctuation">,</span> i<span class="token punctuation">]</span> <span class="token operator">=</span> <span class="token punctuation">(</span>data<span class="token punctuation">[</span><span class="token punctuation">:</span><span class="token punctuation">,</span> i<span class="token punctuation">]</span> <span class="token operator">-</span> avgs<span class="token punctuation">[</span>i<span class="token punctuation">]</span><span class="token punctuation">)</span> <span class="token operator">/</span> <span class="token punctuation">(</span>maximums<span class="token punctuation">[</span>i<span class="token punctuation">]</span> <span class="token operator">-</span> minimums<span class="token punctuation">[</span>i<span class="token punctuation">]</span><span class="token punctuation">)</span>

<span class="token comment"># 训练集和测试集的划分比例</span>
<span class="token comment">#ratio = 0.8</span>
<span class="token comment">#offset = int(data.shape[0] * ratio)</span>
training_data <span class="token operator">=</span> data<span class="token punctuation">[</span><span class="token punctuation">:</span>offset<span class="token punctuation">]</span>
test_data <span class="token operator">=</span> data<span class="token punctuation">[</span>offset<span class="token punctuation">:</span><span class="token punctuation">]</span>
<span class="token keyword">return</span> training_data<span class="token punctuation">,</span> test_data

#定义线性回归模型

class Regressor(fluid.dygraph.Layer):

def init(self):

super(Regressor, self).init()

#调用父类的函数init

# 定义一层全连接层，输出维度是1，激活函数为None，即不使用激活函数

self.fc = Linear(input_dim=13, output_dim=1, act=None)
<span class="token comment"># 网络的前向计算函数</span>
<span class="token keyword">def</span> <span class="token function">forward</span><span class="token punctuation">(</span>self<span class="token punctuation">,</span> inputs<span class="token punctuation">)</span><span class="token punctuation">:</span>
    x <span class="token operator">=</span> self<span class="token punctuation">.</span>fc<span class="token punctuation">(</span>inputs<span class="token punctuation">)</span>
    <span class="token keyword">return</span> x

# 定义飞桨动态图的工作环境

with fluid.dygraph.guard():

# 声明定义好的线性回归模型

model = Regressor()

# 开启模型训练模式

model.train()

# 加载数据

training_data, test_data = load_data()

# 定义优化算法，这里使用随机梯度下降-SGD

# 学习率设置为0.01

opt = fluid.optimizer.SGD(learning_rate=0.01, parameter_list=model.parameters())
with dygraph.guard(fluid.CPUPlace()):

EPOCH_NUM = 10   # 设置外层循环次数

BATCH_SIZE = 10  # 设置batch大小
<span class="token comment"># 定义外层循环</span>
<span class="token keyword">for</span> epoch_id <span class="token keyword">in</span> <span class="token builtin">range</span><span class="token punctuation">(</span>EPOCH_NUM<span class="token punctuation">)</span><span class="token punctuation">:</span>
    <span class="token comment"># 在每轮迭代开始之前，将训练数据的顺序随机的打乱</span>
    np<span class="token punctuation">.</span>random<span class="token punctuation">.</span>shuffle<span class="token punctuation">(</span>training_data<span class="token punctuation">)</span>
    <span class="token comment"># 将训练数据进行拆分，每个batch包含10条数据</span>
    mini_batches <span class="token operator">=</span> <span class="token punctuation">[</span>training_data<span class="token punctuation">[</span>k<span class="token punctuation">:</span>k<span class="token operator">+</span>BATCH_SIZE<span class="token punctuation">]</span> <span class="token keyword">for</span> k <span class="token keyword">in</span> <span class="token builtin">range</span><span class="token punctuation">(</span><span class="token number">0</span><span class="token punctuation">,</span> <span class="token builtin">len</span><span class="token punctuation">(</span>training_data<span class="token punctuation">)</span><span class="token punctuation">,</span> BATCH_SIZE<span class="token punctuation">)</span><span class="token punctuation">]</span>
    <span class="token comment"># 定义内层循环</span>
    <span class="token keyword">for</span> iter_id<span class="token punctuation">,</span> mini_batch <span class="token keyword">in</span> <span class="token builtin">enumerate</span><span class="token punctuation">(</span>mini_batches<span class="token punctuation">)</span><span class="token punctuation">:</span>
        x <span class="token operator">=</span> np<span class="token punctuation">.</span>array<span class="token punctuation">(</span>mini_batch<span class="token punctuation">[</span><span class="token punctuation">:</span><span class="token punctuation">,</span> <span class="token punctuation">:</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">.</span>astype<span class="token punctuation">(</span><span class="token string">'float32'</span><span class="token punctuation">)</span> <span class="token comment"># 获得当前批次训练数据</span>
        y <span class="token operator">=</span> np<span class="token punctuation">.</span>array<span class="token punctuation">(</span>mini_batch<span class="token punctuation">[</span><span class="token punctuation">:</span><span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">:</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">.</span>astype<span class="token punctuation">(</span><span class="token string">'float32'</span><span class="token punctuation">)</span> <span class="token comment"># 获得当前批次训练标签（真实房价）</span>
        <span class="token comment"># 将numpy数据转为飞桨动态图variable形式</span>
        house_features <span class="token operator">=</span> dygraph<span class="token punctuation">.</span>to_variable<span class="token punctuation">(</span>x<span class="token punctuation">)</span>
        prices <span class="token operator">=</span> dygraph<span class="token punctuation">.</span>to_variable<span class="token punctuation">(</span>y<span class="token punctuation">)</span>
        
        <span class="token comment"># 前向计算</span>
        predicts <span class="token operator">=</span> model<span class="token punctuation">(</span>house_features<span class="token punctuation">)</span>
        
        <span class="token comment"># 计算损失</span>
        loss <span class="token operator">=</span> fluid<span class="token punctuation">.</span>layers<span class="token punctuation">.</span>square_error_cost<span class="token punctuation">(</span>predicts<span class="token punctuation">,</span> label<span class="token operator">=</span>prices<span class="token punctuation">)</span>
        avg_loss <span class="token operator">=</span> fluid<span class="token punctuation">.</span>layers<span class="token punctuation">.</span>mean<span class="token punctuation">(</span>loss<span class="token punctuation">)</span>
        <span class="token keyword">if</span> iter_id<span class="token operator">%</span><span class="token number">20</span><span class="token operator">==</span><span class="token number">0</span><span class="token punctuation">:</span>
            <span class="token keyword">print</span><span class="token punctuation">(</span><span class="token string">"epoch: {}, iter: {}, loss is: {}"</span><span class="token punctuation">.</span><span class="token builtin">format</span><span class="token punctuation">(</span>epoch_id<span class="token punctuation">,</span> iter_id<span class="token punctuation">,</span> avg_loss<span class="token punctuation">.</span>numpy<span class="token punctuation">(</span><span class="token punctuation">)</span><span class="token punctuation">)</span><span class="token punctuation">)</span>
        
        <span class="token comment"># 反向传播</span>
        avg_loss<span class="token punctuation">.</span>backward<span class="token punctuation">(</span><span class="token punctuation">)</span>
        <span class="token comment"># 最小化loss,更新参数</span>
        opt<span class="token punctuation">.</span>minimize<span class="token punctuation">(</span>avg_loss<span class="token punctuation">)</span>
        <span class="token comment"># 清除梯度</span>
        model<span class="token punctuation">.</span>clear_gradients<span class="token punctuation">(</span><span class="token punctuation">)</span>
<span class="token comment"># 保存模型</span>
fluid<span class="token punctuation">.</span>save_dygraph<span class="token punctuation">(</span>model<span class="token punctuation">.</span>state_dict<span class="token punctuation">(</span><span class="token punctuation">)</span><span class="token punctuation">,</span> <span class="token string">'LR_model'</span><span class="token punctuation">)</span>

# 定义飞桨动态图工作环境

with fluid.dygraph.guard():

# 保存模型参数，文件名为LR_model

fluid.save_dygraph(model.state_dict(), 'LR_model')

print("模型保存成功，模型参数保存在LR_model中")
#测试模型

def load_one_example(data_dir):

f = open(data_dir, 'r')

datas = f.readlines()

# 选择倒数第10条数据用于测试

tmp = datas[-10]

tmp = tmp.strip().split()

one_data = [float(v) for v in tmp]
<span class="token comment"># 对数据进行归一化处理</span>
<span class="token keyword">for</span> i <span class="token keyword">in</span> <span class="token builtin">range</span><span class="token punctuation">(</span><span class="token builtin">len</span><span class="token punctuation">(</span>one_data<span class="token punctuation">)</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">)</span><span class="token punctuation">:</span>
    one_data<span class="token punctuation">[</span>i<span class="token punctuation">]</span> <span class="token operator">=</span> <span class="token punctuation">(</span>one_data<span class="token punctuation">[</span>i<span class="token punctuation">]</span> <span class="token operator">-</span> avg_values<span class="token punctuation">[</span>i<span class="token punctuation">]</span><span class="token punctuation">)</span> <span class="token operator">/</span> <span class="token punctuation">(</span>max_values<span class="token punctuation">[</span>i<span class="token punctuation">]</span> <span class="token operator">-</span> min_values<span class="token punctuation">[</span>i<span class="token punctuation">]</span><span class="token punctuation">)</span>

data <span class="token operator">=</span> np<span class="token punctuation">.</span>reshape<span class="token punctuation">(</span>np<span class="token punctuation">.</span>array<span class="token punctuation">(</span>one_data<span class="token punctuation">[</span><span class="token punctuation">:</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">,</span> <span class="token punctuation">[</span><span class="token number">1</span><span class="token punctuation">,</span> <span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span><span class="token punctuation">.</span>astype<span class="token punctuation">(</span>np<span class="token punctuation">.</span>float32<span class="token punctuation">)</span>
label <span class="token operator">=</span> one_data<span class="token punctuation">[</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span>
<span class="token keyword">return</span> data<span class="token punctuation">,</span> label

with dygraph.guard():

# 参数为保存模型参数的文件地址

model_dict, _ = fluid.load_dygraph('LR_model')

model.load_dict(model_dict)

model.eval()
<span class="token comment"># 参数为数据集的文件地址</span>
test_data<span class="token punctuation">,</span> label <span class="token operator">=</span> load_one_example<span class="token punctuation">(</span><span class="token string">'./work/housing.data'</span><span class="token punctuation">)</span>
<span class="token comment"># 将数据转为动态图的variable格式</span>
test_data <span class="token operator">=</span> dygraph<span class="token punctuation">.</span>to_variable<span class="token punctuation">(</span>test_data<span class="token punctuation">)</span>
results <span class="token operator">=</span> model<span class="token punctuation">(</span>test_data<span class="token punctuation">)</span>

<span class="token comment"># 对结果做反归一化处理</span>
results <span class="token operator">=</span> results <span class="token operator">*</span> <span class="token punctuation">(</span>max_values<span class="token punctuation">[</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span> <span class="token operator">-</span> min_values<span class="token punctuation">[</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span><span class="token punctuation">)</span> <span class="token operator">+</span> avg_values<span class="token punctuation">[</span><span class="token operator">-</span><span class="token number">1</span><span class="token punctuation">]</span>
<span class="token keyword">print</span><span class="token punctuation">(</span><span class="token string">"Inference result is {}, the corresponding label is {}"</span><span class="token punctuation">.</span><span class="token builtin">format</span><span class="token punctuation">(</span>results<span class="token punctuation">.</span>numpy<span class="token punctuation">(</span><span class="token punctuation">)</span><span class="token punctuation">,</span> label<span class="token punctuation">)</span><span class="token punctuation">)</span>