UCI-HAR数据集CNN分类

import torch.utils.data as Data
import numpy as np
import torchtrain_x_list = "x_train.npy"
train_y_list = "y_train.npy"
test_x_list = "x_test.npy"
test_y_list = "y_test.npy"#加载数据集
class HAR(Data.Dataset):def __init__(self, filename_x, filename_y):self.filename_x = filename_xself.filename_y = filename_ydef HAR_data(self):data_x_raw = np.load(self.filename_x)print(data_x_raw.shape)data_x = data_x_raw.reshape(-1, 1, data_x_raw.shape[1],data_x_raw.shape[2])  # (N, C, H, W) (7352, 1, 128, 9)# data_x = np.expand_dims(data_x_raw, 1)print(data_x.shape)data_y = np.load(self.filename_y)torch_dataset = Data.TensorDataset(torch.from_numpy(data_x), torch.from_numpy(data_y))return torch_datasetif __name__ == "__main__":data_train = HAR(train_x_list, train_y_list)har_train_tensor = data_train.HAR_data()data_test = HAR(test_x_list, test_y_list)har_test_tensor = data_test.HAR_data()train_loader = Data.DataLoader(dataset=har_train_tensor, batch_size=128, shuffle=True, num_workers=5,)test_loader = Data.DataLoader(dataset=har_test_tensor, batch_size=128, shuffle=True, num_workers=5,)# for step, (batch_x, batch_y) in enumerate(train_loader):

'''
-*- coding: utf-8 -*-
@Time : 2021/8/7 20:55
@Author : Small_Volcano
@File : UCI_HAR_CNN.py
'''
import copy
import time
import torch
import torch.nn as nn
import torch.utils.data as Data
from torch.optim import Adam
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from torchsummary import summarytrain_x_list = "x_train.npy"
train_y_list = "y_train.npy"
#测试数据集
test_x_list = "x_test.npy"
test_y_list = "y_test.npy"class HAR(Data.Dataset):def __init__(self, filename_x, filename_y):self.filename_x = filename_xself.filename_y = filename_ydef HAR_data(self):"""更改x的维度,加载x和y"""data_x_raw = np.load(self.filename_x)#print(data_x_raw.shape)                                                        #为什么是1通道 128*9data_x = data_x_raw.reshape(-1, 1, data_x_raw.shape[1], data_x_raw.shape[2])  # (N, C, H, W) (7352, 1, 128, 9)# data_x = np.expand_dims(data_x_raw, 1)#print(data_x.shape)data_y = np.load(self.filename_y)print("datay{}".format(data_y))torch_dataset = Data.TensorDataset(torch.from_numpy(data_x), torch.from_numpy(data_y))#造数据集return torch_datasetdata_train = HAR(train_x_list, train_y_list)#这一步是做什么？
har_train_tensor = data_train.HAR_data()    #创造训练验证数据集
#print(har_train_tensor)
#测试集数据
data_test = HAR(test_x_list, test_y_list)
har_test_tensor = data_test.HAR_data()train_loader = Data.DataLoader(dataset=har_train_tensor,batch_size=128,shuffle=True,num_workers=0, )
#设置一个测试集加载器
test_loader = Data.DataLoader(dataset=har_test_tensor,batch_size=1,shuffle=True,num_workers=0, )#搭建卷积神经网络
class Net(nn.Module):def __init__(self):super(Net, self).__init__()#定义第一个卷积层self.conv1 = nn.Sequential(nn.Conv2d(in_channels=1,out_channels=12,          #输出高度12kernel_size=3,            #卷积核尺寸3*3stride=1,padding=1),               #(1*128*9)-->(12*128*9)nn.ReLU(),nn.MaxPool2d(kernel_size=2,stride=2) #(12*128*9)-->(12*64*4))#定义第二个卷积层self.conv2 = nn.Sequential(nn.Conv2d(in_channels=12,out_channels=32,kernel_size=3,stride=1,padding=1),               #(12*64*4)-->(32*64*4)nn.ReLU(),nn.MaxPool2d(kernel_size=2,stride=2) #池化后：(32*32*2))self.conv3 = nn.Sequential(nn.Conv2d(in_channels=32,out_channels=64,kernel_size=3,stride=1,padding=1),                #(32*32*2)-->(64*32*2)nn.ReLU())#定义全连接层self.classifier = nn.Sequential(nn.Linear(64*32*2,128),              #长方体变平面nn.ReLU(),nn.Dropout(p = 0.5),nn.Linear(128,6))#定义网络的前向传播路径def forward(self,x):x = self.conv1(x)#print(x.shape)x = self.conv2(x)#print(x.shape)x = self.conv3(x)#print(x.shape)x = x.view(x.shape[0],-1) #展平多维的卷积图层output = self.classifier(x)return output#输出网络结构
net = Net()     #创建实例
print(net)
#定义网络的训练过程函数
def train_model(model,traindataloader,train_rate,criterion,optimizer,num_epochs=25):#train_rate：训练集中训练数量的百分比#计算训练使用的batch数量batch_num = len(traindataloader)train_batch_num = round(batch_num * train_rate) #前train_rate（80%）的batch进行训练#复制最好模型的参数best_model_wts = copy.deepcopy(model.state_dict())best_acc = 0.0train_loss_all = []train_acc_all =    []val_loss_all = []val_acc_all = []since = time.time()for epoch in range(num_epochs):print('Epoch {}/{}'.format(epoch,num_epochs-1)) #格式化字符串print('-' * 10)#每个epoch有两个训练阶段train_loss = 0.0train_corrects = 0train_num = 0val_loss = 0.0val_corrects = 0val_num = 0for step,(b_x,b_y) in enumerate(traindataloader,1): #取标签和样本b_y = b_y.long()if step < train_batch_num:                      #前train_rate（80%）的batch进行训练model.train()                               #设置模型为训练模式，对Droopou有用output = model(b_x)# print(b_x)#取得模型预测结果pre_lab = torch.argmax(output,1)            #横向获得最大值位置#b_y = torch.Tensor(b_y).long()             #修改BUG#print(b_y)loss = criterion(output,b_y)                #每个样本的lossoptimizer.zero_grad()loss.backward()optimizer.step()                            #修改权值train_loss += loss.item() * b_x.size(0)#print(pre_lab)#print(b_y.data)train_corrects += torch.sum(pre_lab == b_y.data) #训练正确个数train_num += b_x.size(0)else:model.eval()                                    #设置模型为验证模式output = model(b_x)pre_lab = torch.argmax(output,1)loss = criterion(output,b_y)val_loss += loss.item() * b_x.size(0)val_corrects += torch.sum(pre_lab == b_y.data)val_num += b_x.size(0)#计算训练集和验证集上的损失和精度train_loss_all.append(train_loss / train_num)           #一个epoch上的losstrain_acc_all.append(train_corrects.double().item() / train_num)val_loss_all.append(val_loss / val_num)val_acc_all.append(val_corrects.double().item() / val_num)print('{} Train Loss: {:.4f} Train Acc: {:.4f}'.format(epoch,train_loss_all[-1],train_acc_all[-1])) #此处-1没搞明白print('{} Val Loss: {:.4f} Val Acc: {:.4f}'.format(epoch,val_loss_all[-1],val_acc_all[-1]))#拷贝模型最高精度下的参数if val_acc_all[-1] > best_acc:best_acc = val_acc_all[-1]best_model_wts = copy.deepcopy(model.state_dict())torch.save(model.state_dict(),"UCI_HAR_model")torch.save(optimizer.state_dict(),"UCI_HAR_optimizer")time_use = time.time() - sinceprint("Train and val complete in {:.0f}m {:.0f}s".format(time_use // 60,time_use % 60)) #训练用时#使用最好模型的参数model.load_state_dict(best_model_wts)#组成数据表格train_process打印train_process = pd.DataFrame(data={"epoch":range(num_epochs),"train_loss_all":train_loss_all,"val_loss_all":val_loss_all,"train_acc_all":train_acc_all,"val_acc_all":val_acc_all})return model,train_process
#对模型进行训练
optimizer = Adam(net.parameters(),lr=0.0003)            #优化器
criterion = nn.CrossEntropyLoss()                       #使用交叉熵作为损失函数
net,train_process = train_model(net,train_loader,0.8,   #使用训练集的20%作为验证criterion,optimizer,num_epochs=25)
#可视化模型训练过程中
plt.figure(figsize=(12,4))
plt.subplot(1,2,1)
plt.plot(train_process.epoch,train_process.train_loss_all,"ro-",label="Train loss")
plt.plot(train_process.epoch,train_process.val_loss_all,"bs-",label="Val loss")
plt.legend()
plt.xlabel("epoch")
plt.ylabel("Loss")
plt.subplot(1,2,2)
plt.plot(train_process.epoch,train_process.train_acc_all,"ro-",label="Train acc")
plt.plot(train_process.epoch,train_process.val_acc_all,"bs-",label="Val acc")
plt.xlabel("epoch")
plt.ylabel("acc")
plt.legend()
plt.show()#对测试集进行预测,计算模型的泛化能力
def test(model,testdataloader,criterion):test_loss_all = []test_acc_all = []test_loss = 0.0test_corrects = 0test_num = 0for step,(input,target) in enumerate(testdataloader):   #取标签和样本target = target.long()#target = torch.Tensor(target).long()model.eval()                                       #设置模型为训练模式，对Droopou有用output = model(input)# print(b_x)#取得模型预测结果pre_lab = torch.argmax(output,1)                    #横向获得最大值位置loss = criterion(output,target)                     #每个样本的losstest_loss += loss.item() * input.size(0)            #此处的b_x.size(0)=batch_size。此处相当于一个batch的loss？计算的是整体训练的loss#print(pre_lab)#print(input.data)test_corrects += torch.sum(pre_lab == target.data)  #测试正确个数test_num += input.size(0)test_loss_all.append(test_loss / test_num)test_acc_all.append(test_corrects.double().item() / test_num)print('Test all Loss: {:.4f} Test Acc: {:.4f}'.format(test_loss_all[-1], test_acc_all[-1]))test = test(net,test_loader,criterion)"""
Epoch 0/24
----------
0 Train Loss: 1.3879 Train Acc: 0.4361
0 Val Loss: 0.8158 Val Acc: 0.6156
Train and val complete in 0m 8s
Epoch 1/24
----------
1 Train Loss: 0.8039 Train Acc: 0.5974
1 Val Loss: 0.6446 Val Acc: 0.7487
Train and val complete in 0m 15s
Epoch 2/24
----------
2 Train Loss: 0.6284 Train Acc: 0.7115
2 Val Loss: 0.5581 Val Acc: 0.7198
Train and val complete in 0m 22s
Epoch 3/24
----------
3 Train Loss: 0.5377 Train Acc: 0.7696
3 Val Loss: 0.4765 Val Acc: 0.8003
Train and val complete in 0m 29s
Epoch 4/24
----------
4 Train Loss: 0.4640 Train Acc: 0.7990
4 Val Loss: 0.3666 Val Acc: 0.8580
Train and val complete in 0m 36s
Epoch 5/24
----------
5 Train Loss: 0.3987 Train Acc: 0.8370
5 Val Loss: 0.3171 Val Acc: 0.8894
Train and val complete in 0m 43s
Epoch 6/24
----------
6 Train Loss: 0.3426 Train Acc: 0.8681
6 Val Loss: 0.2961 Val Acc: 0.8863
Train and val complete in 0m 49s
Epoch 7/24
----------
7 Train Loss: 0.3154 Train Acc: 0.8786
7 Val Loss: 0.2763 Val Acc: 0.8976
Train and val complete in 0m 56s
Epoch 8/24
----------
8 Train Loss: 0.2811 Train Acc: 0.9019
8 Val Loss: 0.2236 Val Acc: 0.9033
Train and val complete in 1m 3s
Epoch 9/24
----------
9 Train Loss: 0.2505 Train Acc: 0.9042
9 Val Loss: 0.2151 Val Acc: 0.9234
Train and val complete in 1m 10s
Epoch 10/24
----------
10 Train Loss: 0.2375 Train Acc: 0.9109
10 Val Loss: 0.1955 Val Acc: 0.9165
Train and val complete in 1m 17s
Epoch 11/24
----------
11 Train Loss: 0.2131 Train Acc: 0.9200
11 Val Loss: 0.1771 Val Acc: 0.9353
Train and val complete in 1m 23s
Epoch 12/24
----------
12 Train Loss: 0.1995 Train Acc: 0.9231
12 Val Loss: 0.1647 Val Acc: 0.9366
Train and val complete in 1m 30s
Epoch 13/24
----------
13 Train Loss: 0.1917 Train Acc: 0.9281
13 Val Loss: 0.1392 Val Acc: 0.9454
Train and val complete in 1m 37s
Epoch 14/24
----------
14 Train Loss: 0.1835 Train Acc: 0.9306
14 Val Loss: 0.1347 Val Acc: 0.9491
Train and val complete in 1m 44s
Epoch 15/24
----------
15 Train Loss: 0.1822 Train Acc: 0.9316
15 Val Loss: 0.1394 Val Acc: 0.9479
Train and val complete in 1m 52s
Epoch 16/24
----------
16 Train Loss: 0.1633 Train Acc: 0.9403
16 Val Loss: 0.1280 Val Acc: 0.9529
Train and val complete in 1m 59s
Epoch 17/24
----------
17 Train Loss: 0.1611 Train Acc: 0.9389
17 Val Loss: 0.1619 Val Acc: 0.9221
Train and val complete in 2m 6s
Epoch 18/24
----------
18 Train Loss: 0.1471 Train Acc: 0.9448
18 Val Loss: 0.1444 Val Acc: 0.9403
Train and val complete in 2m 14s
Epoch 19/24
----------
19 Train Loss: 0.1442 Train Acc: 0.9465
19 Val Loss: 0.1342 Val Acc: 0.9472
Train and val complete in 2m 22s
Epoch 20/24
----------
20 Train Loss: 0.1497 Train Acc: 0.9422
20 Val Loss: 0.1150 Val Acc: 0.9585
Train and val complete in 2m 31s
Epoch 21/24
----------
21 Train Loss: 0.1388 Train Acc: 0.9477
21 Val Loss: 0.1221 Val Acc: 0.9460
Train and val complete in 2m 39s
Epoch 22/24
----------
22 Train Loss: 0.1410 Train Acc: 0.9436
22 Val Loss: 0.1288 Val Acc: 0.9460
Train and val complete in 2m 47s
Epoch 23/24
----------
23 Train Loss: 0.1417 Train Acc: 0.9477
23 Val Loss: 0.1131 Val Acc: 0.9585
Train and val complete in 2m 55s
Epoch 24/24
----------
24 Train Loss: 0.1320 Train Acc: 0.9483
24 Val Loss: 0.1275 Val Acc: 0.9472
Train and val complete in 3m 3s
Test all Loss: 0.3446 Test Acc: 0.8683
"""

UCI-HAR数据集CNN分类相关推荐

UCI 机器学习数据集（分类）
113 Data Sets UCI数据集 Name Data Types Default Task Attribute Types # Instances # Attributes Year Ab ...
matlab粗集系统评估wine,C4.5 决策树算法对UCI wine数据集分类的实现(MATLAB)
一.功能实现与样本分析在数据挖掘领域,能够利用相应的算法对数据集进行训练,即对样本的特征进行分析,从而概括出相同类别的样本之间存在的内在特征联系,进一步对未知类别的样本进行预测,判断出该样本所属的类 ...
C4.5 决策树算法对UCI wine数据集分类的实现(MATLAB)
1.功能实现与样本分析在数据挖掘领域,可以利用相应的算法对数据集进行训练,即对样本的特征进行分析,从而归纳出相同类别的样本之间存在的内在特征联系,进一步对未知类别的样本进行预测,判断出该样本所属的类 ...
wine葡萄酒数据集KNNSVM分类实验
声明:本篇文章是本人课程作业的内容,只提供平时学习参考使用,请勿转载. 介绍:数据挖掘来源:kaibo_lei_ZZU 本片文章是使用分类算法KNN,和SVM支持向量机分类算法,对Wine数据集进行 ...
手写汉字数字识别详细过程（构建数据集+CNN神经网络+Tensorflow）
手写汉字数字识别(构建数据集+CNN神经网络) 期末,P老师布置了一个大作业,自己构建数据集实现手写汉字数字的识别.太捞了,记录一下过程.大概花了一个下午加半个晚上,主要是做数据集花时间. 一.构建数 ...
paddle2.0高层API实现自定义数据集文本分类中的情感分析任务
paddle2.0高层API实现自定义数据集文本分类中的情感分析任务本文包含了: - 自定义文本分类数据集继承 - 文本分类数据处理 - 循环神经网络RNN, LSTM - ·seq2vec· - ...
Python实现哈里斯鹰优化算法(HHO)优化卷积神经网络分类模型(CNN分类算法)项目实战
说明:这是一个机器学习实战项目(附带数据+代码+文档+视频讲解),如需数据+代码+文档+视频讲解可以直接到文章最后获取. 1.项目背景 2019年Heidari等人提出哈里斯鹰优化算法(Harris ...
R语言图形用户界面数据挖掘包Rattle：加载UCI糖尿病数据集、并启动Rattle图形用户界面、数据集变量重命名，为数据集结果变量添加标签、数据划分（训练集、测试集、验证集）、随机数设置
R语言图形用户界面数据挖掘包Rattle:加载UCI糖尿病数据集.并启动Rattle图形用户界面.数据集变量重命名,为数据集结果变量添加标签.数据划分(训练集.测试集.验证集).随机数设置目录
CNN分类，ResNet V1 ,ResNet V2,ResNeXt,DenseNet
一.CNN分类 1.基于空间利用的CNN 2.基于深度的CNN 3.基于多路径的CNN 4.基于宽度的多连接 5.基于特征图的CNN 6.基于通道的CNN 7.基于注意力的CNN 二,ResNet V ...
《pytorch》对CIFAR数据集的分类
今天复习了下pytorch,又看了下它的一个实例,复习了下对CIFAR数据集的分类. 学习链接:https://zhuanlan.zhihu.com/p/39667289 直接上源码,注释均在其中 i ...

UCI-HAR数据集CNN分类

UCI-HAR数据集CNN分类相关推荐

最新文章

热门文章