pytorch简单线性回归

最新推荐文章于 2024-10-01 23:51:03 发布

红糖果仁沙琪玛

最新推荐文章于 2024-10-01 23:51:03 发布

阅读量90

点赞数

分类专栏： pytorch 文章标签： pytorch 线性回归 python

本文链接：https://blog.csdn.net/K_ZhJ18/article/details/130138495

版权

pytorch 专栏收录该内容

4 篇文章 0 订阅

订阅专栏

import torch as tc
import matplotlib.pyplot as plt
class EasyReg(object):
    def __init__(self, x, y, w, b):
        self.x = x
        self.y = y
        self.w = w
        self.b = b
    def predict(self):
        return tc.matmul(self.x, self.w) + b
    def lossFunc(self, ypre):
        return (self.y - ypre).pow(2).mean()
    def training(self, learningRate, iterNum):
        for i in range(iterNum):
            ypre = self.predict()
            loss = self.lossFunc(ypre)
            if self.w.grad != None:
                self.w.grad.data.zero_()
            if self.b.grad != None:
                self.b.grad.data.zero_()
            loss.backward()
            self.w.data -= learningRate * self.w.grad.data
            self.b.data -= learningRate * self.b.grad.data
            if i % 30 == 0:
                print("w: ", self.w.data, "b: ", self.b.data, "loss: ", loss.data)
        return self.predict()

x = tc.rand([100, 1], dtype = tc.float32) * 5
r = tc.rand([100, 1])
y = x * 3 + r
w = tc.rand([1, 1], dtype = tc.float32, requires_grad = True)
b = tc.rand(1, dtype = tc.float32, requires_grad = True)
test = EasyReg(x, y, w, b)
newY = test.training(0.1, 1000)

plt.scatter(x.numpy().reshape(-1), y.data.numpy().reshape(-1))
plt.plot(x.numpy().reshape(-1), newY.data.numpy().reshape(-1), c = "red")
plt.show()

对于带有扰动的y(x) = y + e ，寻找一条直线能尽可能的反应y，则令y = w*x+b，损失函数

loss = 实际值和预测值的均方根误差。在训练中利用梯度下降法使loss不断减小，便可以最终找到

一条最优的直线。