pytorch-TensorFlow-线性回归

import tensorflow as tf
import numpy as np
x =    #假设x.shape（n，2）
y =    #假设y.shape(n,1)
w = tf.Variable(tf.random.randn([2,1],name='weight')   #w,b这里的参数与x，y的            shape对应
b = tf.Variable(tf.random.randn([1],name='bias')
def linear_regression():
  return x*w + b
def square_loss(y_pred,y_true):
  return tf.reduce_mean(tf.pow(y_pred-y_true,2).sum())
optimizer = tf.optimizers.SGD(0.01)   #应用随机梯度下降算法，是梯度下降下降算法的一种改进。
def run_optimization():
  with tf.GradientTape() as g:      #获取 梯度信息
    pred = linear_regression()
    loss = square_loss(pred,y)
  gradients = g.gradients(loss,[w,b])  #得到loss对变量w，b的梯度
  optimizer.apply_gradients(zip(gradients,[w.b])  #通过梯度信息，更新变量w，b将loss最小化
for i in range(100):
  run_optimization()

import torch
import numpy as np
from torch.autograd import Variable
x =   #numpy.float32数据类型  
y =   #numpy.float32数据类型
x , y = torch.from_numpy(x), torch.from_numpy(y)  #将numpy数据转换成tensor
w = Variable(torch.randn(1),requires_grad=True)
b = Variable(torch.randn(1),requires_grad=True)
def linear_regression():
    return x*w.expand_as(x) + b.expand_as(x)
for i in range(100):
    if (w.grad is not None) and (b.grad is not None):  
        w.grad.data.zero_() #因为梯度会累加 ，  所以在更新之前要先将之前的梯度信息归零
        b.grad.data.zero_() 
    pred = linear_regression()
    loss = torch.mean((pred - y) ** 2) 
    loss.backward()#方向传播 ，得到 变量的梯度
    print(loss)
    w.data.add_(- 0.01 * w.grad.data)  #利用上一步计算中得到的w的梯度信息更新w中的data数值
    b.data.add_(- 0.01 * b.grad.data)  #利用上一步计算中得到的b的梯度信息更新b中的data数值

import torch
import torch.nn as nn
import torch.optim
from torch.autograd import Variable
x =    
y =
class Net(nn.Model):
  def __init__(self):
    super(Net,self).__init__()  #自动寻找基类
    self.layers = nn.Linear(1,1)   #输入输出数据的维度
  def forwadr(self,x):
    x = self.layers(x)
    return x     #进行前向传播
net = Net()
optimizer = torch.optim.SGD(0.001)
error = torch.nn.MSELoss()
for i in range(1100):
  pred = net(x)
  loss = error(pred,y)
  optimizer.zero_grad()   #清除梯度信息
  loss.backward()     #生成梯度信息
  optimizer.step()        #应用梯度信息

import tensorflow as tf
from tensorflow.keras import Model , layers
x = 
y = 
class Net(Model):
  def __init__(self):
    super(Net,self).__init__()
    self.layers = layers.Dense(1,activation='relu')
  def forward(self,x):
    x = self.layers(x)
    return x
def square_loss(y_pred,y_true):
  return tf.reduce_mean(tf.pow(y_pred-y_true,2).sum())
optimizer = tf.optimizers.SGD(0.001)
net = Net()
def run_optimization():
  with tf.gradientTape() as g:
    pred = net(x)
    loss = square_loss(pred,y)
  trainable_cariables = net.trainable_variables
  gradients = g.gradient(loss,trainable_cariables)
  optimizer.apply_gradients(zip(gradients,trainable_cariables))