手把手教你实现PyTorch的MNIST数据集
概述
MNIST 包含 0~9 的手写数字, 共有 60000 个训练集和 10000 个测试集. 数据的格式为单通道 28*28 的灰度图.
获取数据
def get_data():
"""获取数据"""
# 获取测试集
train = torchvision.datasets.MNIST(root="./data", train=True, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(), # 转换成张量
torchvision.transforms.Normalize((0.1307,), (0.3081,)) # 标准化
]))
train_loader = DataLoader(train, batch_size=batch_size) # 分割测试集
# 获取测试集
test = torchvision.datasets.MNIST(root="./data", train=False, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(), # 转换成张量
torchvision.transforms.Normalize((0.1307,), (0.3081,)) # 标准化
]))
test_loader = DataLoader(test, batch_size=batch_size) # 分割训练
# 返回分割好的训练集和测试集
return train_loader, test_loader
网络模型
class Model(torch.nn.Module):
def __init__(self):
super(Model, self).__init__()
# 卷积层
self.conv1 = torch.nn.Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
self.conv2 = torch.nn.Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
# Dropout层
self.dropout1 = torch.nn.Dropout(0.25)
self.dropout2 = torch.nn.Dropout(0.5)
# 全连接层
self.fc1 = torch.nn.Linear(9216, 128)
self.fc2 = torch.nn.Linear(128, 10)
def forward(self, x):
"""前向传播"""
# [b, 1, 28, 28] => [b, 32, 26, 26]
out = self.conv1(x)
out = F.relu(out)
# [b, 32, 26, 26] => [b, 64, 24, 24]
out = self.conv2(out)
out = F.relu(out)
# [b, 64, 24, 24] => [b, 64, 12, 12]
out = F.max_pool2d(out, 2)
out = self.dropout1(out)
# [b, 64, 12, 12] => [b, 64 * 12 * 12] => [b, 9216]
out = torch.flatten(out, 1)
# [b, 9216] => [b, 128]
out = self.fc1(out)
out = F.relu(out)
# [b, 128] => [b, 10]
out = self.dropout2(out)
out = self.fc2(out)
output = F.log_softmax(out, dim=1)
return output
train 函数
def train(model, epoch, train_loader):
"""训练"""
# 训练模式
model.train()
# 迭代
for step, (x, y) in enumerate(train_loader):
# 加速
if use_cuda:
model = model.cuda()
x, y = x.cuda(), y.cuda()
# 梯度清零
optimizer.zero_grad()
output = model(x)
# 计算损失
loss = F.nll_loss(output, y)
# 反向传播
loss.backward()
# 更新梯度
optimizer.step()
# 打印损失
if step % 50 == 0:
print('Epoch: {}, Step {}, Loss: {}'.format(epoch, step, loss))
test 函数
def test(model, test_loader):
"""测试"""
# 测试模式
model.eval()
# 存放正确个数
correct = 0
with torch.no_grad():
for x, y in test_loader:
# 加速
if use_cuda:
model = model.cuda()
x, y = x.cuda(), y.cuda()
# 获取结果
output = model(x)
# 预测结果
pred = output.argmax(dim=1, keepdim=True)
# 计算准确个数
correct += pred.eq(y.view_as(pred)).sum().item()
# 计算准确率
accuracy = correct / len(test_loader.dataset) * 100
# 输出准确
print("Test Accuracy: {}%".format(accuracy))
main 函数
def main():
# 获取数据
train_loader, test_loader = get_data()
# 迭代
for epoch in range(iteration_num):
print("\n================ epoch: {} ================".format(epoch))
train(network, epoch, train_loader)
test(network, test_loader)
完整代码:
import torch
import torchvision
import torch.nn.functional as F
from torch.utils.data import DataLoader
class Model(torch.nn.Module):
def __init__(self):
super(Model, self).__init__()
# 卷积层
self.conv1 = torch.nn.Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
self.conv2 = torch.nn.Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
# Dropout层
self.dropout1 = torch.nn.Dropout(0.25)
self.dropout2 = torch.nn.Dropout(0.5)
# 全连接层
self.fc1 = torch.nn.Linear(9216, 128)
self.fc2 = torch.nn.Linear(128, 10)
def forward(self, x):
"""前向传播"""
# [b, 1, 28, 28] => [b, 32, 26, 26]
out = self.conv1(x)
out = F.relu(out)
# [b, 32, 26, 26] => [b, 64, 24, 24]
out = self.conv2(out)
out = F.relu(out)
# [b, 64, 24, 24] => [b, 64, 12, 12]
out = F.max_pool2d(out, 2)
out = self.dropout1(out)
# [b, 64, 12, 12] => [b, 64 * 12 * 12] => [b, 9216]
out = torch.flatten(out, 1)
# [b, 9216] => [b, 128]
out = self.fc1(out)
out = F.relu(out)
# [b, 128] => [b, 10]
out = self.dropout2(out)
out = self.fc2(out)
output = F.log_softmax(out, dim=1)
return output
# 定义超参数
batch_size = 64 # 一次训练的样本数目
learning_rate = 0.0001 # 学习率
iteration_num = 5 # 迭代次数
network = Model() # 实例化网络
print(network) # 调试输出网络结构
optimizer = torch.optim.Adam(network.parameters(), lr=learning_rate) # 优化器
# GPU 加速
use_cuda = torch.cuda.is_available()
print("是否使用 GPU 加速:", use_cuda)
def get_data():
"""获取数据"""
# 获取测试集
train = torchvision.datasets.MNIST(root="./data", train=True, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(), # 转换成张量
torchvision.transforms.Normalize((0.1307,), (0.3081,)) # 标准化
]))
train_loader = DataLoader(train, batch_size=batch_size) # 分割测试集
# 获取测试集
test = torchvision.datasets.MNIST(root="./data", train=False, download=True,
transform=torchvision.transforms.Compose([
torchvision.transforms.ToTensor(), # 转换成张量
torchvision.transforms.Normalize((0.1307,), (0.3081,)) # 标准化
]))
test_loader = DataLoader(test, batch_size=batch_size) # 分割训练
# 返回分割好的训练集和测试集
return train_loader, test_loader
def train(model, epoch, train_loader):
"""训练"""
# 训练模式
model.train()
# 迭代
for step, (x, y) in enumerate(train_loader):
# 加速
if use_cuda:
model = model.cuda()
x, y = x.cuda(), y.cuda()
# 梯度清零
optimizer.zero_grad()
output = model(x)
# 计算损失
loss = F.nll_loss(output, y)
# 反向传播
loss.backward()
# 更新梯度
optimizer.step()
# 打印损失
if step % 50 == 0:
print('Epoch: {}, Step {}, Loss: {}'.format(epoch, step, loss))
def test(model, test_loader):
"""测试"""
# 测试模式
model.eval()
# 存放正确个数
correct = 0
with torch.no_grad():
for x, y in test_loader:
# 加速
if use_cuda:
model = model.cuda()
x, y = x.cuda(), y.cuda()
# 获取结果
output = model(x)
# 预测结果
pred = output.argmax(dim=1, keepdim=True)
# 计算准确个数
correct += pred.eq(y.view_as(pred)).sum().item()
# 计算准确率
accuracy = correct / len(test_loader.dataset) * 100
# 输出准确
print("Test Accuracy: {}%".format(accuracy))
def main():
# 获取数据
train_loader, test_loader = get_data()
# 迭代
for epoch in range(iteration_num):
print("\n================ epoch: {} ================".format(epoch))
train(network, epoch, train_loader)
test(network, test_loader)
if __name__ == "__main__":
main()
输出结果:
Model(
(conv1): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1))
(conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
(dropout1): Dropout(p=0.25, inplace=False)
(dropout2): Dropout(p=0.5, inplace=False)
(fc1): Linear(in_features=9216, out_features=128, bias=True)
(fc2): Linear(in_features=128, out_features=10, bias=True)
)
是否使用 GPU 加速: True================ epoch: 0 ================
Epoch: 0, Step 0, Loss: 2.3131277561187744
Epoch: 0, Step 50, Loss: 1.0419045686721802
Epoch: 0, Step 100, Loss: 0.6259541511535645
Epoch: 0, Step 150, Loss: 0.7194482684135437
Epoch: 0, Step 200, Loss: 0.4020516574382782
Epoch: 0, Step 250, Loss: 0.6890509128570557
Epoch: 0, Step 300, Loss: 0.28660136461257935
Epoch: 0, Step 350, Loss: 0.3277580738067627
Epoch: 0, Step 400, Loss: 0.2750288248062134
Epoch: 0, Step 450, Loss: 0.28428223729133606
Epoch: 0, Step 500, Loss: 0.3514065444469452
Epoch: 0, Step 550, Loss: 0.23386947810649872
Epoch: 0, Step 600, Loss: 0.25338059663772583
Epoch: 0, Step 650, Loss: 0.1743898093700409
Epoch: 0, Step 700, Loss: 0.35752204060554504
Epoch: 0, Step 750, Loss: 0.17575909197330475
Epoch: 0, Step 800, Loss: 0.20604261755943298
Epoch: 0, Step 850, Loss: 0.17389622330665588
Epoch: 0, Step 900, Loss: 0.3188241124153137
Test Accuracy: 96.56%================ epoch: 1 ================
Epoch: 1, Step 0, Loss: 0.23558208346366882
Epoch: 1, Step 50, Loss: 0.13511177897453308
Epoch: 1, Step 100, Loss: 0.18823786079883575
Epoch: 1, Step 150, Loss: 0.2644936144351959
Epoch: 1, Step 200, Loss: 0.145077645778656
Epoch: 1, Step 250, Loss: 0.30574971437454224
Epoch: 1, Step 300, Loss: 0.2386859953403473
Epoch: 1, Step 350, Loss: 0.08346735686063766
Epoch: 1, Step 400, Loss: 0.10480977594852448
Epoch: 1, Step 450, Loss: 0.07280707359313965
Epoch: 1, Step 500, Loss: 0.20928426086902618
Epoch: 1, Step 550, Loss: 0.20455852150917053
Epoch: 1, Step 600, Loss: 0.10085935145616531
Epoch: 1, Step 650, Loss: 0.13476189970970154
Epoch: 1, Step 700, Loss: 0.19087043404579163
Epoch: 1, Step 750, Loss: 0.0981522724032402
Epoch: 1, Step 800, Loss: 0.1961515098810196
Epoch: 1, Step 850, Loss: 0.041140712797641754
Epoch: 1, Step 900, Loss: 0.250461220741272
Test Accuracy: 98.03%================ epoch: 2 ================
Epoch: 2, Step 0, Loss: 0.09572553634643555
Epoch: 2, Step 50, Loss: 0.10370486229658127
Epoch: 2, Step 100, Loss: 0.17737184464931488
Epoch: 2, Step 150, Loss: 0.1570713371038437
Epoch: 2, Step 200, Loss: 0.07462178170681
Epoch: 2, Step 250, Loss: 0.18744900822639465
Epoch: 2, Step 300, Loss: 0.09910508990287781
Epoch: 2, Step 350, Loss: 0.08929706364870071
Epoch: 2, Step 400, Loss: 0.07703761011362076
Epoch: 2, Step 450, Loss: 0.10133732110261917
Epoch: 2, Step 500, Loss: 0.1314031481742859
Epoch: 2, Step 550, Loss: 0.10394387692213058
Epoch: 2, Step 600, Loss: 0.11612939089536667
Epoch: 2, Step 650, Loss: 0.17494803667068481
Epoch: 2, Step 700, Loss: 0.11065669357776642
Epoch: 2, Step 750, Loss: 0.061209067702293396
Epoch: 2, Step 800, Loss: 0.14715790748596191
Epoch: 2, Step 850, Loss: 0.03930797800421715
Epoch: 2, Step 900, Loss: 0.18030673265457153
Test Accuracy: 98.46000000000001%================ epoch: 3 ================
Epoch: 3, Step 0, Loss: 0.09266342222690582
Epoch: 3, Step 50, Loss: 0.0414913073182106
Epoch: 3, Step 100, Loss: 0.2152961939573288
Epoch: 3, Step 150, Loss: 0.12287424504756927
Epoch: 3, Step 200, Loss: 0.13468700647354126
Epoch: 3, Step 250, Loss: 0.11967387050390244
Epoch: 3, Step 300, Loss: 0.11301510035991669
Epoch: 3, Step 350, Loss: 0.037447575479745865
Epoch: 3, Step 400, Loss: 0.04699449613690376
Epoch: 3, Step 450, Loss: 0.05472381412982941
Epoch: 3, Step 500, Loss: 0.09839300811290741
Epoch: 3, Step 550, Loss: 0.07964356243610382
Epoch: 3, Step 600, Loss: 0.08182843774557114
Epoch: 3, Step 650, Loss: 0.05514759197831154
Epoch: 3, Step 700, Loss: 0.13785190880298615
Epoch: 3, Step 750, Loss: 0.062480345368385315
Epoch: 3, Step 800, Loss: 0.120387002825737
Epoch: 3, Step 850, Loss: 0.04458726942539215
Epoch: 3, Step 900, Loss: 0.17119190096855164
Test Accuracy: 98.55000000000001%================ epoch: 4 ================
Epoch: 4, Step 0, Loss: 0.08094145357608795
Epoch: 4, Step 50, Loss: 0.05615215748548508
Epoch: 4, Step 100, Loss: 0.07766406238079071
Epoch: 4, Step 150, Loss: 0.07915271818637848
Epoch: 4, Step 200, Loss: 0.1301635503768921
Epoch: 4, Step 250, Loss: 0.12118984013795853
Epoch: 4, Step 300, Loss: 0.073218435049057
Epoch: 4, Step 350, Loss: 0.04517696052789688
Epoch: 4, Step 400, Loss: 0.08493026345968246
Epoch: 4, Step 450, Loss: 0.03904269263148308
Epoch: 4, Step 500, Loss: 0.09386837482452393
Epoch: 4, Step 550, Loss: 0.12583576142787933
Epoch: 4, Step 600, Loss: 0.09053893387317657
Epoch: 4, Step 650, Loss: 0.06912104040384293
Epoch: 4, Step 700, Loss: 0.1502612829208374
Epoch: 4, Step 750, Loss: 0.07162325084209442
Epoch: 4, Step 800, Loss: 0.10512275993824005
Epoch: 4, Step 850, Loss: 0.028180215507745743
Epoch: 4, Step 900, Loss: 0.08492615073919296
Test Accuracy: 98.69%
到此这篇关于手把手教你实现PyTorch的MNIST数据集的文章就介绍到这了,更多相关PyTorch MNIST数据集内容请搜索我们以前的文章或继续浏览下面的相关文章希望大家以后多多支持我们！
相关推荐
-
pytorch GAN伪造手写体mnist数据集方式
一,mnist数据集 形如上图的数字手写体就是mnist数据集. 二,GAN原理(生成对抗网络) GAN网络一共由两部分组成:一个是伪造器(Generator,简称G),一个是判别器(Discrimniator,简称D) 一开始,G由服从某几个分布(如高斯分布)的噪音组成,生成的图片不断送给D判断是否正确,直到G生成的图片连D都判断以为是真的.D每一轮除了看过G生成的假图片以外,还要见数据集中的真图片,以前者和后者得到的损失函数值为依据更新D网络中的权值.因此G和D都在不停地更新权值.以下图为例
-
关于Pytorch的MNIST数据集的预处理详解
关于Pytorch的MNIST数据集的预处理详解 MNIST的准确率达到99.7% 用于MNIST的卷积神经网络(CNN)的实现,具有各种技术,例如数据增强,丢失,伪随机化等. 操作系统:ubuntu18.04 显卡:GTX1080ti python版本:2.7(3.7) 网络架构 具有4层的CNN具有以下架构. 输入层:784个节点(MNIST图像大小) 第一卷积层:5x5x32 第一个最大池层 第二卷积层:5x5x64 第二个最大池层 第三个完全连接层:1024个节点 输出层:10个节点(M
-
pytorch:实现简单的GAN示例(MNIST数据集)
我就废话不多说了,直接上代码吧! # -*- coding: utf-8 -*- """ Created on Sat Oct 13 10:22:45 2018 @author: www """ import torch from torch import nn from torch.autograd import Variable import torchvision.transforms as tfs from torch.utils.dat
-
pytorch实现建立自己的数据集(以mnist为例)
本文将原始的numpy array数据在pytorch下封装为Dataset类的数据集,为后续深度网络训练提供数据. 加载并保存图像信息 首先导入需要的库,定义各种路径. import os import matplotlib from keras.datasets import mnist import numpy as np from torch.utils.data.dataset import Dataset from PIL import Image import scipy.misc
-
详解PyTorch手写数字识别(MNIST数据集)
MNIST 手写数字识别是一个比较简单的入门项目,相当于深度学习中的 Hello World,可以让我们快速了解构建神经网络的大致过程.虽然网上的案例比较多,但还是要自己实现一遍.代码采用 PyTorch 1.0 编写并运行. 导入相关库 import torch import torch.nn as nn import torch.nn.functional as F import torch.optim as optim from torchvision import datasets, t
-
手把手教你实现PyTorch的MNIST数据集
概述 MNIST 包含 0~9 的手写数字, 共有 60000 个训练集和 10000 个测试集. 数据的格式为单通道 28*28 的灰度图. 获取数据 def get_data(): """获取数据""" # 获取测试集 train = torchvision.datasets.MNIST(root="./data", train=True, download=True, transform=torchvision.tran
-
pytorch 把MNIST数据集转换成图片和txt的方法
本文介绍了pytorch 把MNIST数据集转换成图片和txt的方法,分享给大家,具体如下: 1.下载Mnist 数据集 import os # third-party library import torch import torch.nn as nn from torch.autograd import Variable import torch.utils.data as Data import torchvision import matplotlib.pyplot as plt # t
-
Pytorch使用MNIST数据集实现基础GAN和DCGAN详解
原始生成对抗网络Generative Adversarial Networks GAN包含生成器Generator和判别器Discriminator,数据有真实数据groundtruth,还有需要网络生成的"fake"数据,目的是网络生成的fake数据可以"骗过"判别器,让判别器认不出来,就是让判别器分不清进入的数据是真实数据还是fake数据.总的来说是:判别器区分真实数据和fake数据的能力越强越好:生成器生成的数据骗过判别器的能力越强越好,这个是矛盾的,所以只能
-
Pytorch使用MNIST数据集实现CGAN和生成指定的数字方式
CGAN的全拼是Conditional Generative Adversarial Networks,条件生成对抗网络,在初始GAN的基础上增加了图片的相应信息. 这里用传统的卷积方式实现CGAN. import torch from torch.utils.data import DataLoader from torchvision.datasets import MNIST from torchvision import transforms from torch import opti
-
pytorch实现mnist数据集的图像可视化及保存
如何将pytorch中mnist数据集的图像可视化及保存 导出一些库 import torch import torchvision import torch.utils.data as Data import scipy.misc import os import matplotlib.pyplot as plt BATCH_SIZE = 50 DOWNLOAD_MNIST = True 数据集的准备 #训练集测试集的准备 train_data = torchvision.datasets.M
-
PyTorch实现MNIST数据集手写数字识别详情
目录 一.PyTorch是什么? 二.程序示例 1.引入必要库 2.下载数据集 3.加载数据集 4.搭建CNN模型并实例化 5.交叉熵损失函数损失函数及SGD算法优化器 6.训练函数 7.测试函数 8.运行 三.总结 前言: 本篇文章基于卷积神经网络CNN,使用PyTorch实现MNIST数据集手写数字识别. 一.PyTorch是什么? PyTorch 是一个 Torch7 团队开源的 Python 优先的深度学习框架,提供两个高级功能: 强大的 GPU 加速 Tensor 计算(类似 nump
-
用Pytorch训练CNN(数据集MNIST,使用GPU的方法)
听说pytorch使用比TensorFlow简单,加之pytorch现已支持windows,所以今天装了pytorch玩玩,第一件事还是写了个简单的CNN在MNIST上实验,初步体验的确比TensorFlow方便. 参考代码(在莫烦python的教程代码基础上修改)如下: import torch import torch.nn as nn from torch.autograd import Variable import torch.utils.data as Data import tor
-
使用 PyTorch 实现 MLP 并在 MNIST 数据集上验证方式
简介 这是深度学习课程的第一个实验,主要目的就是熟悉 Pytorch 框架.MLP 是多层感知器,我这次实现的是四层感知器,代码和思路参考了网上的很多文章.个人认为,感知器的代码大同小异,尤其是用 Pytorch 实现,除了层数和参数外,代码都很相似. Pytorch 写神经网络的主要步骤主要有以下几步: 1 构建网络结构 2 加载数据集 3 训练神经网络(包括优化器的选择和 Loss 的计算) 4 测试神经网络 下面将从这四个方面介绍 Pytorch 搭建 MLP 的过程. 项目代码地址:la