PaddlePaddle 极简入门实践二：简单的线性回归

Posted 2020-12-12 zhulimin

import paddle.fluid as fluid
import numpy


#定义数据
train_data = [[0], [1], [2], [3], [4], [5], [10]]
y_true = [[3], [13], [23], [33], [43], [53], [103]]

#定义网络
x = fluid.layers.data(name="x", shape=[1], dtype="float32")
y = fluid.layers.data(name="y", shape=[1], dtype="float32")
y_predict = fluid.layers.fc(input=x, size=1, act=None)  # 定义x与其有关系

#定义损失函数
cost = fluid.layers.square_error_cost(input=y_predict,label=y)#平方差损失函数
avg_cost = fluid.layers.mean(cost)#求平均损失

#定义优化方法
sgd_optimizer = fluid.optimizer.SGD(learning_rate=0.01)#定义SGD-随机梯度下降的学习率
sgd_optimizer.minimize(avg_cost)

cpu = fluid.CPUPlace()#此处使用CPU进行训练 GPU训练则移步之后更新的文章
exe = fluid.Executor(cpu)#Executor是执行器
prog=fluid.default_startup_program()#将刚刚定义的一堆堆赋值给prog这个变量名
exe.run(prog)#准备开始！

for i in range(500):
    for data_id in range(len(y_true)):
        # 需转换为numpy数组类型，这样可以传入训练
        data_x=numpy.array(train_data[data_id]).astype("float32").reshape((1,1))
        data_y = numpy.array(y_true[data_id]).astype("float32").reshape((1,1))
        outs = exe.run(
            feed={‘x‘:data_x , ‘y‘: data_y},
            fetch_list=[y_predict.name, avg_cost])  # feed为数据表 输入数据和标签数据
        print("正在训练第" + str(i + 1) + "次")
        # 观察结果
        print(outs)


# 保存预测模型
dirname = "./test01.inference.model/"
fluid.io.save_inference_model(dirname, [‘x‘], [y_predict], exe)


######################################################################################

import paddle.fluid as fluid
import numpy

import paddle.fluid as fluid
import numpy

dirname = "./test01.inference.model/"


cpu = fluid.CPUPlace()
exe = fluid.Executor(cpu)


# 加载模型
[inference_program, feed_target_names, fetch_targets] = fluid.io.load_inference_model(dirname, exe)

# 目标数据
datatype = "float32"
test_data = numpy.array([[input("请输入数值")]]).astype(datatype)


# 启动预测
results = exe.run(inference_program,
                  feed={feed_target_names[0]: test_data},
                  fetch_list=fetch_targets)


print(results[0][0])