paddlepaddle demo——基于y=3*x+2的线性回归预测

开发环境

---

Jupyter Notebook

Python版本   python 3.7

框架版本   PaddlePaddle 1.8.4

具体实现

---

import paddle
import numpy as np
import paddle.fluid as fluid
import matplotlib.pyplot as plt

BATCH_SIZE=10
iter=0;
iters=[] #训练轮数
train_costs=[] #损失值，判断收敛程度

#创建神经网络
x = fluid.layers.data(name='x',shape=[1],dtype='float32')
hidden = fluid.layers.fc(input=x,size=100,act='relu')
net = fluid.layers.fc(input=hidden,size=1,act=None)

#定义损失函数
y = fluid.layers.data(name='y',shape=[1],dtype='float32')
cost = fluid.layers.square_error_cost(input=net,label=y)
avg_cost = fluid.layers.mean(cost) #square_error_cost()求的是一个Batch的损失值，所以还要用mean()再求一个平均值。

#优化函数
optimizer = fluid.optimizer.SGDOptimizer(learning_rate=0.01)
opts = optimizer.minimize(avg_cost)

#创建cpu解释器
place = fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())

x_train_data = np.array([[1.0],[2.0],[3.0],[4.0],[5.0]]).astype('float32')
y_train_data = np.array([[5.0], [8.0], [11.0], [14.0], [17.0]]).astype('float32') #对应关系：y=3*x+2
test_data = np.array([[6.0]]).astype('float32') #预测数据

#定义可视化函数
def draw_train_process(iters,train_costs):
    title="training cost"
    plt.title(title, fontsize=24)
    plt.xlabel("iter", fontsize=14) #横坐标表示训练轮数
    plt.ylabel("cost", fontsize=14) #纵坐标表示收敛程度
    plt.plot(iters, train_costs,color='red',label='training cost') 
    plt.grid()
    plt.show()
    
#训练n轮
n = 400
for pass_id in range(n):
    train_cost = exe.run(
        program=fluid.default_main_program(),
        feed={'x':x_train_data,'y':y_train_data},
        fetch_list=[avg_cost]
    )
    if pass_id % 10 ==0: #每BATCH_SIZE条打印一次数据，BATCH_SIZE=10
        print("id:%d,cost:%0.5f"%(pass_id,train_cost[0])) 
        iter=iter+BATCH_SIZE
        iters.append(iter)
        train_costs.append(train_cost[0])
    
draw_train_process(iters,train_costs)

test_program = fluid.default_main_program().clone(for_test=True)

#预测
result = exe.run(
    program=test_program,
    feed={'x':test_data,'y':np.array([[0.0]]).astype('float32')},
    fetch_list=[net]
)
#输出预测结果
print("当x为6.0时，y为：%0.5f" % result[0])

预测结果

---