Start from scratch pytorch (four) softmax return to achieve

FashionMNIST data set samples were 70,000, 60,000 train, 10000 th test. A total of 10 categories.

Download the following manner.

mnist_train = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST', 
                                                train=True, download=True, 
                                                transform=transforms.ToTensor())
mnist_test = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST', 
                                               train=False, download=True, 
                                               transform=transforms.ToTensor())

softmax realized from zero

• Loading
• Model initialization parameters
• Model definition
• Loss of function definition
• Optimization definition
• training

Loading

import torch
import torchvision
import torchvision.transforms as transforms

mnist_train = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST',
                                                train=True, download=True,
                                                transform=transforms.ToTensor())
mnist_test = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST',
                                               train=False, download=True,
                                               transform=transforms.ToTensor())
batch_size = 256
num_workers = 4  # 多进程同时读取
train_iter = torch.utils.data.DataLoader(
    mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(
    mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)

Model initialization parameters

num_inputs = 784  # 图像是28 X 28的图像,共784个特征
num_outputs = 10

W = torch.tensor(np.random.normal(
    0, 0.01, (num_inputs, num_outputs)), dtype=torch.float)
b = torch.zeros(num_outputs, dtype=torch.float)

W.requires_grad_(requires_grad=True)
b.requires_grad_(requires_grad=True)

Model definition

REMEMBER: Behavior dim dimension along the direction of 0, as 1. addition dimension along the column direction, i.e. the sum of the elements of each row.

def softmax(X):  # X.shape=[样本数,类别数]
    Ｘ_exp = X.exp()
    partion = X_exp.sum(dim=1, keepdim=True)  # 沿着列方向求和,即对每一行求和
    #print(partion.shape)
    return X_exp/partion  # 广播机制,partion被扩展成与Ｘ_exp同shape的,对应位置元素做除法

def net(X):
    # 通过`view`函数将每张原始图像改成长度为`num_inputs`的向量
    return softmax(torch.mm(X.view(-1, num_inputs), W) + b)

Loss of function definition

Number of samples is assumed that the training data set as \ (n-\) , the cross-entropy loss function defined as
\ [\ ell (\ boldsymbol { \ Theta}) = \ frac {1} {n} \ sum_ {i = 1} ^ n H \ left (\ boldsymbol y ^ {(i)}, \ boldsymbol {\ hat y} ^ {(i)} \ right), \]

Wherein \ (\ boldsymbol {\ Theta} \) representative of the model parameters. Likewise, if only one label each sample, then the cross entropy loss can be abbreviated as \ (\ ell (\ boldsymbol { \ Theta}) = - (1 / n) \ sum_ {i = 1} ^ n \ log \ hat ^ {Y {Y_ (I)}} ^ {(I)} \) .

def cross_entropy(y_hat, y):
    y_hat_prob = y_hat.gather(1, y.view(-1, 1))  # ,沿着列方向,即选取出每一行下标为y的元素
    return -torch.log(y_hat_prob)

https://pytorch.org/docs/stable/torch.html

Gather () along dimension dim, of the selected element index index

Optimization algorithm definition

def sgd(params, lr, batch_size):
    for param in params:
        param.data -= lr * param.grad / batch_size  # 注意这里更改param时用的param.data

Accuracy evaluation function

def evaluate_accuracy(data_iter, net):
    acc_sum, n = 0.0, 0
    for X, y in data_iter:
        acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
        n += y.shape[0]
    return acc_sum / n

training

• Read batch_size samples
• Forward propagation, calculating the predicted value
• Compared with the real value to calculate the loss
• Back-propagation, a gradient calculation
• Update the parameters
  so on ad infinitum.

num_epochs, lr = 5, 0.1
def train():
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
        for X, y in train_iter:
            #print(X.shape,y.shape)
            y_hat = net(X)
            l = cross_entropy(y_hat, y).sum()  # 求loss
            l.backward()  # 反向传播,计算梯度
            sgd([W, b], lr, batch_size)  # 根据梯度,更新参数

            Ｗ.grad.data.zero_()  # 清空梯度
            b.grad.data.zero_()

            train_l_sum += l.item()
            train_acc_sum += (y_hat.argmax(dim=1) == y).sum().item()
            n += y.shape[0]
        test_acc = evaluate_accuracy(test_iter, net)
        print('epoch %d, loss %.4f, train_acc %.3f，test_acc %.3f'
              % (epoch + 1, train_l_sum / n, train_acc_sum/n, test_acc))

train()

Output is as follows:

epoch 1, loss 0.7848, train_acc 0.748，test_acc 0.793
epoch 2, loss 0.5704, train_acc 0.813，test_acc 0.811
epoch 3, loss 0.5249, train_acc 0.825，test_acc 0.821
epoch 4, loss 0.5011, train_acc 0.832，test_acc 0.821
epoch 5, loss 0.4861, train_acc 0.837，test_acc 0.829

-------

softmax simple realization

• Loading
• Model definition and model initialization parameters
• Loss of function definition
• Optimization definition
• training

Data read

import torch
from torch import nn
from torch.nn import init
import numpy as np
import sys
import torchvision
import torchvision.transforms as transforms

mnist_train = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST',
                                                train=True, download=True,
                                                transform=transforms.ToTensor())
mnist_test = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST',
                                               train=False, download=True,
                                               transform=transforms.ToTensor())
batch_size = 256
num_workers = 4  # 多进程同时读取
train_iter = torch.utils.data.DataLoader(
    mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
test_iter = torch.utils.data.DataLoader(
    mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)

Model definition and model initialization parameters

num_inputs = 784  # 图像是28 X 28的图像,共784个特征
num_outputs = 10

class LinearNet(nn.Module):
    def __init__(self,num_inputs,num_outputs):
        super(LinearNet,self).__init__()
        self.linear = nn.Linear(num_inputs,num_outputs)
    
    def forward(self,x):  #x.shape=(batch,1,28,28)
        return self.linear(x.view(x.shape[0],-1))　#输入shape应该是[,784]

net = LinearNet(num_inputs,num_outputs)

torch.nn.init.normal_(net.linear.weight,mean=0,std=0.01)
torch.nn.init.constant_(net.linear.bias,val=0)

没有什么要特别注意的,注意一点,由于self.linear的input size为[,784]，所以要对x做一次变形x.view(x.shape[0],-1)

损失函数定义

torch里的这个损失函数是包括了softmax计算概率和交叉熵计算的．

loss = nn.CrossEntropyLoss()

优化器定义

optimizer = torch.optim.SGD(net.parameters(),lr=0.01)

训练

精度评测函数

def evaluate_accuracy(data_iter, net):
    acc_sum, n = 0.0, 0
    for X, y in data_iter:
        acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
        n += y.shape[0]
    return acc_sum / n

训练

• 读入batch_size个样本
• 前向传播,计算预测值
• 与真值相比,计算loss
• 反向传播,计算梯度
• 更新各个参数
  如此循环往复.

num_epochs = 5
def train():
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n = 0.0, 0.0, 0
        for X,y in train_iter:
            y_hat=net(X)   #前向传播
            l = loss(y_hat,y).sum()#计算loss
            l.backward()#反向传播

            optimizer.step()#参数更新
            optimizer.zero_grad()#清空梯度

            train_l_sum += l.item()
            train_acc_sum += (y_hat.argmax(dim=1) == y).sum().item()
            n += y.shape[0]

        test_acc = evaluate_accuracy(test_iter, net)
        print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f'
                % (epoch + 1, train_l_sum / n, train_acc_sum / n, test_acc))

train()

输出

epoch 1, loss 0.0054, train acc 0.638, test acc 0.681
epoch 2, loss 0.0036, train acc 0.716, test acc 0.724
epoch 3, loss 0.0031, train acc 0.749, test acc 0.745
epoch 4, loss 0.0029, train acc 0.767, test acc 0.759
epoch 5, loss 0.0028, train acc 0.780, test acc 0.770