softmax回归分类2

最新推荐文章于 2023-06-08 23:22:08 发布
最新推荐文章于 2023-06-08 23:22:08 发布
阅读量749 收藏
点赞数
分类专栏： pytorch实践 文章标签： 回归 分类 pytorch
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/jiadongfengyahoo/article/details/122312194
版权
博客对比了使用全连接层和多层感知机在FashionMNIST数据集上的训练效果。全连接层在100个训练周期后的测试准确率为82%，而加入ReLU激活函数的多层感知机在相同条件下测试准确率从82%逐步提升到约83%，显示出一定的性能提升。
摘要由CSDN通过智能技术生成 
import torch
from torch import nn
from torch.nn import init
import numpy as np
import sys
sys.path.append("..")
from utils.SlowFast import SlowFast1
from utils.SlowFast import get_fashion_data

使用全连接层替换slowfast实现

num_inputs = 784
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)
        init.normal_(self.linear.weight, mean = 0, std = 0.01)
        init.constant_(self.linear.bias, val = 0)
    def forward(self, x): # x shape: (batch, 1, 28, 28)
        y = self.linear(x.view(x.shape[0], -1))
        return y

net = LinearNet(num_inputs, num_outputs)
loss = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr = 0.01)
slowfast1 = SlowFast1(image_size = 28*28, class_nums = 10)
num_epochs = 100
batch_size = 1000
train_iter, test_iter = get_fashion_data(img_dir='./Datasets/FashionMNIST', batch_size = batch_size)

slowfast1.train(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)

epoch 1, loss 0.0018, train acc 0.570, test acc 0.652
epoch 2, loss 0.0014, train acc 0.666, test acc 0.659
epoch 3, loss 0.0012, train acc 0.673, test acc 0.671
epoch 4, loss 0.0011, train acc 0.686, test acc 0.679
epoch 5, loss 0.0010, train acc 0.699, test acc 0.693
epoch 6, loss 0.0009, train acc 0.712, test acc 0.705
epoch 7, loss 0.0009, train acc 0.724, test acc 0.718
epoch 8, loss 0.0009, train acc 0.733, test acc 0.726
epoch 9, loss 0.0008, train acc 0.742, test acc 0.732
epoch 10, loss 0.0008, train acc 0.748, test acc 0.740
epoch 11, loss 0.0008, train acc 0.754, test acc 0.743
epoch 12, loss 0.0008, train acc 0.760, test acc 0.747
epoch 13, loss 0.0008, train acc 0.764, test acc 0.750
epoch 14, loss 0.0007, train acc 0.767, test acc 0.755
epoch 15, loss 0.0007, train acc 0.771, test acc 0.757
epoch 16, loss 0.0007, train acc 0.774, test acc 0.762
epoch 17, loss 0.0007, train acc 0.777, test acc 0.764
epoch 18, loss 0.0007, train acc 0.779, test acc 0.766
epoch 19, loss 0.0007, train acc 0.782, test acc 0.769
epoch 20, loss 0.0007, train acc 0.785, test acc 0.772
epoch 21, loss 0.0007, train acc 0.787, test acc 0.773
epoch 22, loss 0.0007, train acc 0.789, test acc 0.774
epoch 23, loss 0.0007, train acc 0.791, test acc 0.777
epoch 24, loss 0.0007, train acc 0.792, test acc 0.779
epoch 25, loss 0.0007, train acc 0.794, test acc 0.780
epoch 26, loss 0.0007, train acc 0.795, test acc 0.781
epoch 27, loss 0.0006, train acc 0.796, test acc 0.783
epoch 28, loss 0.0006, train acc 0.798, test acc 0.785
epoch 29, loss 0.0006, train acc 0.799, test acc 0.786
epoch 30, loss 0.0006, train acc 0.800, test acc 0.787
epoch 31, loss 0.0006, train acc 0.801, test acc 0.788
epoch 32, loss 0.0006, train acc 0.802, test acc 0.788
epoch 33, loss 0.0006, train acc 0.803, test acc 0.789
epoch 34, loss 0.0006, train acc 0.804, test acc 0.791
epoch 35, loss 0.0006, train acc 0.805, test acc 0.793
epoch 36, loss 0.0006, train acc 0.806, test acc 0.793
epoch 37, loss 0.0006, train acc 0.806, test acc 0.794
epoch 38, loss 0.0006, train acc 0.807, test acc 0.795
epoch 39, loss 0.0006, train acc 0.808, test acc 0.796
epoch 40, loss 0.0006, train acc 0.809, test acc 0.795
epoch 41, loss 0.0006, train acc 0.810, test acc 0.797
epoch 42, loss 0.0006, train acc 0.810, test acc 0.798
epoch 43, loss 0.0006, train acc 0.811, test acc 0.799
epoch 44, loss 0.0006, train acc 0.811, test acc 0.801
epoch 45, loss 0.0006, train acc 0.812, test acc 0.801
epoch 46, loss 0.0006, train acc 0.813, test acc 0.801
epoch 47, loss 0.0006, train acc 0.813, test acc 0.803
epoch 48, loss 0.0006, train acc 0.814, test acc 0.803
epoch 49, loss 0.0006, train acc 0.815, test acc 0.804
epoch 50, loss 0.0006, train acc 0.815, test acc 0.805
epoch 51, loss 0.0006, train acc 0.815, test acc 0.805
epoch 52, loss 0.0006, train acc 0.816, test acc 0.805
epoch 53, loss 0.0006, train acc 0.817, test acc 0.805
epoch 54, loss 0.0006, train acc 0.817, test acc 0.806
epoch 55, loss 0.0006, train acc 0.818, test acc 0.807
epoch 56, loss 0.0006, train acc 0.818, test acc 0.806
epoch 57, loss 0.0006, train acc 0.819, test acc 0.807
epoch 58, loss 0.0006, train acc 0.819, test acc 0.807
epoch 59, loss 0.0006, train acc 0.820, test acc 0.807
epoch 60, loss 0.0006, train acc 0.820, test acc 0.808
epoch 61, loss 0.0006, train acc 0.820, test acc 0.808
epoch 62, loss 0.0006, train acc 0.821, test acc 0.809
epoch 63, loss 0.0006, train acc 0.821, test acc 0.808
epoch 64, loss 0.0006, train acc 0.821, test acc 0.809
epoch 65, loss 0.0006, train acc 0.822, test acc 0.810
epoch 66, loss 0.0006, train acc 0.822, test acc 0.810
epoch 67, loss 0.0005, train acc 0.822, test acc 0.811
epoch 68, loss 0.0005, train acc 0.823, test acc 0.810
epoch 69, loss 0.0005, train acc 0.823, test acc 0.810
epoch 70, loss 0.0005, train acc 0.824, test acc 0.811
epoch 71, loss 0.0005, train acc 0.824, test acc 0.811
epoch 72, loss 0.0005, train acc 0.824, test acc 0.811
epoch 73, loss 0.0005, train acc 0.824, test acc 0.812
epoch 74, loss 0.0005, train acc 0.825, test acc 0.812
epoch 75, loss 0.0005, train acc 0.825, test acc 0.812
epoch 76, loss 0.0005, train acc 0.826, test acc 0.812
epoch 77, loss 0.0005, train acc 0.826, test acc 0.812
epoch 78, loss 0.0005, train acc 0.826, test acc 0.812
epoch 79, loss 0.0005, train acc 0.826, test acc 0.813
epoch 80, loss 0.0005, train acc 0.827, test acc 0.813
epoch 81, loss 0.0005, train acc 0.827, test acc 0.813
epoch 82, loss 0.0005, train acc 0.827, test acc 0.813
epoch 83, loss 0.0005, train acc 0.827, test acc 0.814
epoch 84, loss 0.0005, train acc 0.827, test acc 0.815
epoch 85, loss 0.0005, train acc 0.827, test acc 0.815
epoch 86, loss 0.0005, train acc 0.828, test acc 0.815
epoch 87, loss 0.0005, train acc 0.828, test acc 0.815
epoch 88, loss 0.0005, train acc 0.828, test acc 0.816
epoch 89, loss 0.0005, train acc 0.829, test acc 0.816
epoch 90, loss 0.0005, train acc 0.828, test acc 0.816
epoch 91, loss 0.0005, train acc 0.829, test acc 0.816
epoch 92, loss 0.0005, train acc 0.829, test acc 0.816
epoch 93, loss 0.0005, train acc 0.829, test acc 0.817
epoch 94, loss 0.0005, train acc 0.829, test acc 0.816
epoch 95, loss 0.0005, train acc 0.829, test acc 0.816
epoch 96, loss 0.0005, train acc 0.830, test acc 0.817
epoch 97, loss 0.0005, train acc 0.830, test acc 0.817
epoch 98, loss 0.0005, train acc 0.830, test acc 0.818
epoch 99, loss 0.0005, train acc 0.830, test acc 0.817
epoch 100, loss 0.0005, train acc 0.830, test acc 0.817

准确率跟上一节一致，也是82%左右

使用多层感知机进行训练

class LinearMultiNet(nn.Module):
    def __init__(self, num_inputs, num_outputs):
        super(LinearMultiNet, self).__init__()
        num_hidden = 50
        self.linear = nn.Linear(num_inputs, num_hidden)
        self.linear2 = nn.Linear(num_hidden, num_outputs)
        self.net = nn.Sequential(
            nn.Linear(num_inputs, num_hidden),
            nn.ReLU(),
            nn.Linear(num_hidden, num_outputs),
            )
        for params in net.parameters():
            init.normal_(params, mean=0, std=0.01)
    def forward(self, x): # x shape: (batch, 1, 28, 28)
        y = self.net(x.view(x.shape[0], -1))
        return y

net = LinearMultiNet(num_inputs, num_outputs)
loss = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(), lr = 0.01)
slowfast1 = SlowFast1(image_size = 28*28, class_nums = 10)
num_epochs = 100
batch_size = 1000
train_iter, test_iter = get_fashion_data(img_dir='./Datasets/FashionMNIST', batch_size = batch_size)
slowfast1.train(net, train_iter, test_iter, loss, num_epochs, batch_size, None, None, optimizer)

epoch 1, loss 0.0022, train acc 0.216, test acc 0.327
epoch 2, loss 0.0019, train acc 0.466, test acc 0.592
epoch 3, loss 0.0017, train acc 0.620, test acc 0.623
epoch 4, loss 0.0014, train acc 0.640, test acc 0.642
epoch 5, loss 0.0013, train acc 0.656, test acc 0.655
epoch 6, loss 0.0011, train acc 0.665, test acc 0.660
epoch 7, loss 0.0011, train acc 0.671, test acc 0.664
epoch 8, loss 0.0010, train acc 0.679, test acc 0.676
epoch 9, loss 0.0009, train acc 0.688, test acc 0.684
epoch 10, loss 0.0009, train acc 0.696, test acc 0.688
epoch 11, loss 0.0009, train acc 0.704, test acc 0.696
epoch 12, loss 0.0008, train acc 0.711, test acc 0.702
epoch 13, loss 0.0008, train acc 0.717, test acc 0.709
epoch 14, loss 0.0008, train acc 0.724, test acc 0.715
epoch 15, loss 0.0008, train acc 0.732, test acc 0.719
epoch 16, loss 0.0008, train acc 0.737, test acc 0.728
epoch 17, loss 0.0007, train acc 0.742, test acc 0.732
epoch 18, loss 0.0007, train acc 0.746, test acc 0.737
epoch 19, loss 0.0007, train acc 0.751, test acc 0.741
epoch 20, loss 0.0007, train acc 0.756, test acc 0.747
epoch 21, loss 0.0007, train acc 0.759, test acc 0.751
epoch 22, loss 0.0007, train acc 0.763, test acc 0.755
epoch 23, loss 0.0007, train acc 0.767, test acc 0.757
epoch 24, loss 0.0007, train acc 0.770, test acc 0.761
epoch 25, loss 0.0007, train acc 0.774, test acc 0.765
epoch 26, loss 0.0007, train acc 0.777, test acc 0.768
epoch 27, loss 0.0006, train acc 0.780, test acc 0.771
epoch 28, loss 0.0006, train acc 0.783, test acc 0.775
epoch 29, loss 0.0006, train acc 0.786, test acc 0.777
epoch 30, loss 0.0006, train acc 0.789, test acc 0.779
epoch 31, loss 0.0006, train acc 0.792, test acc 0.782
epoch 32, loss 0.0006, train acc 0.794, test acc 0.784
epoch 33, loss 0.0006, train acc 0.796, test acc 0.786
epoch 34, loss 0.0006, train acc 0.798, test acc 0.788
epoch 35, loss 0.0006, train acc 0.800, test acc 0.791
epoch 36, loss 0.0006, train acc 0.802, test acc 0.792
epoch 37, loss 0.0006, train acc 0.804, test acc 0.794
epoch 38, loss 0.0006, train acc 0.806, test acc 0.794
epoch 39, loss 0.0006, train acc 0.807, test acc 0.798
epoch 40, loss 0.0006, train acc 0.808, test acc 0.798
epoch 41, loss 0.0006, train acc 0.810, test acc 0.798
epoch 42, loss 0.0006, train acc 0.812, test acc 0.800
epoch 43, loss 0.0006, train acc 0.813, test acc 0.801
epoch 44, loss 0.0006, train acc 0.813, test acc 0.803
epoch 45, loss 0.0006, train acc 0.815, test acc 0.803
epoch 46, loss 0.0006, train acc 0.815, test acc 0.804
epoch 47, loss 0.0005, train acc 0.816, test acc 0.805
epoch 48, loss 0.0005, train acc 0.817, test acc 0.807
epoch 49, loss 0.0005, train acc 0.817, test acc 0.808
epoch 50, loss 0.0005, train acc 0.819, test acc 0.808
epoch 51, loss 0.0005, train acc 0.820, test acc 0.810
epoch 52, loss 0.0005, train acc 0.821, test acc 0.810
epoch 53, loss 0.0005, train acc 0.822, test acc 0.810
epoch 54, loss 0.0005, train acc 0.822, test acc 0.811
epoch 55, loss 0.0005, train acc 0.823, test acc 0.812
epoch 56, loss 0.0005, train acc 0.824, test acc 0.814
epoch 57, loss 0.0005, train acc 0.825, test acc 0.814
epoch 58, loss 0.0005, train acc 0.826, test acc 0.815
epoch 59, loss 0.0005, train acc 0.826, test acc 0.816
epoch 60, loss 0.0005, train acc 0.826, test acc 0.815
epoch 61, loss 0.0005, train acc 0.827, test acc 0.816
epoch 62, loss 0.0005, train acc 0.828, test acc 0.817
epoch 63, loss 0.0005, train acc 0.828, test acc 0.817
epoch 64, loss 0.0005, train acc 0.829, test acc 0.818
epoch 65, loss 0.0005, train acc 0.830, test acc 0.819
epoch 66, loss 0.0005, train acc 0.830, test acc 0.819
epoch 67, loss 0.0005, train acc 0.830, test acc 0.820
epoch 68, loss 0.0005, train acc 0.831, test acc 0.820
epoch 69, loss 0.0005, train acc 0.831, test acc 0.820
epoch 70, loss 0.0005, train acc 0.832, test acc 0.820
epoch 71, loss 0.0005, train acc 0.832, test acc 0.821
epoch 72, loss 0.0005, train acc 0.832, test acc 0.820
epoch 73, loss 0.0005, train acc 0.832, test acc 0.822
epoch 74, loss 0.0005, train acc 0.833, test acc 0.822
epoch 75, loss 0.0005, train acc 0.834, test acc 0.822
epoch 76, loss 0.0005, train acc 0.834, test acc 0.821
epoch 77, loss 0.0005, train acc 0.834, test acc 0.822
epoch 78, loss 0.0005, train acc 0.835, test acc 0.822
epoch 79, loss 0.0005, train acc 0.835, test acc 0.823
epoch 80, loss 0.0005, train acc 0.836, test acc 0.823
epoch 81, loss 0.0005, train acc 0.836, test acc 0.822
epoch 82, loss 0.0005, train acc 0.836, test acc 0.823
epoch 83, loss 0.0005, train acc 0.837, test acc 0.824
epoch 84, loss 0.0005, train acc 0.837, test acc 0.824
epoch 85, loss 0.0005, train acc 0.837, test acc 0.824
epoch 86, loss 0.0005, train acc 0.838, test acc 0.824
epoch 87, loss 0.0005, train acc 0.838, test acc 0.825
epoch 88, loss 0.0005, train acc 0.838, test acc 0.825
epoch 89, loss 0.0005, train acc 0.839, test acc 0.825
epoch 90, loss 0.0005, train acc 0.839, test acc 0.825
epoch 91, loss 0.0005, train acc 0.840, test acc 0.826
epoch 92, loss 0.0005, train acc 0.840, test acc 0.827
epoch 93, loss 0.0005, train acc 0.840, test acc 0.827
epoch 94, loss 0.0005, train acc 0.840, test acc 0.827
epoch 95, loss 0.0005, train acc 0.841, test acc 0.828
epoch 96, loss 0.0005, train acc 0.841, test acc 0.828
epoch 97, loss 0.0005, train acc 0.841, test acc 0.828
epoch 98, loss 0.0005, train acc 0.841, test acc 0.829
epoch 99, loss 0.0005, train acc 0.842, test acc 0.828
epoch 100, loss 0.0005, train acc 0.841, test acc 0.829

也是没有多大的提升

古风子
关注
专栏目录
Softmax 回归 vs. k 个二元分类器
qq_26154071的博客
09-27 575
Softmax 回归 vs. k 个二元分类器  (2014-04-03 08:37:41) 转载▼ 标签：  分类器   logistic   softmax   it 分类： 统计分析 如果你在开发一个音乐分类的应用，需要对k种类型的音乐进行识别，那么是选择使用 softmax 分类器呢，还是使用
逻辑回归二分类与softmax多分类方法实现
Air_x的博客
05-22 1894
一般来说，回归不使用与分类问题中，因为回归是连续型模型，而且数据噪音会影响模型结果，因此，可以使用一些函数对回归结果进行归类，达到分类的结果
Softmax回归模型及其代码实现
weixin_44881806的博客
10-08 1111
一、Softamx回归模型的原理 softmax回归跟线性回归一样，都是将输入特征与权重做线性叠加。与线性回归的一个主要不同在于，softmax回归的输出值个数等于标签里的类别数。比如线性回归中很常见的问题----预测房屋价格，我们通常会选取面积（平方米）和房龄（年）作为影响价格的两个因素，在这里面积和房龄作为特征而输出只有一个价格的预测值。而对于softmax，它更适用于做分类预测，比如在垃圾邮件分类问题中，我们往往需要选取几个主要的特征作为系统的输入，然后经过softmax分类算法得到...
从头学pytorch(四)　softmax回归实现
山有山的高度，海有海的胸怀。
12-26 1370
FashionMNIST数据集共70000个样本,60000个train,10000个test．共计10种类别. 通过如下方式下载． mnist_train = torchvision.datasets.FashionMNIST(root='/home/sc/disk/keepgoing/learn_pytorch/Datasets/FashionMNIST', ...
shap 模型_使用shap loss值调试监控模型
weixin_26630173的博客
09-23 1938
shap 模型Responsible AI has been a very hot topic in recent years. Accountability and explainability now become the necessary components of your machine learning models, particularly when the models mak...
【softmax分类】基于梯度下降softmax回归实现minist数据分类matlab代码.zip
12-26
在这个案例中，我们看到它与梯度下降法结合，用于训练一个softmax回归模型，以对MNIST数据集进行分类。 MNIST数据集是一个广泛使用的手写数字识别数据集，包含60,000个训练样本和10,000个测试样本。每个样本是28x28...
【softmax分类】梯度下降softmax回归minist数据分类【含Matlab源码 1645期】.zip
最新发布
06-19
【softmax分类】是一种广泛应用于多分类问题的机器学习模型，它是逻辑回归的扩展形式，能够为每个类别提供一个概率输出。在本项目中，我们将会深入探讨如何利用softmax分类算法结合梯度下降法对MNIST手写数字数据集...
深度学习|2|softmax回归pytorch实现iris鸢尾花数据的softmax回归分类任务
weixin_51672245的博客
06-08 717
由统计学家和植物学家Ronald Fisher在1936年收集并发布。该数据集中包含了150个样本，其中每个样本代表了一朵鸢尾花（iris flower），并且包含了四个特征（sepal length（花萼长度）、sepal width（花萼宽度）、petal length（花瓣长度）和petal width（花瓣宽度））以及对应的类别标签（iris setosa、iris versicolor和iris virginica）。样本数量：150条类别数量：3类每类样本：50条特征维度：4。
logistic回归分类与softmax回归
奔跑的小仙女
10-21 9128
目录 Logistic回归 逻辑回归的定义式： 损失函数 梯度下降 Logistic回归防止过拟合： Softmax回归： loss函数 逻辑回归与Softmax回归的联系 与神经网络的关系 logistic回归（多分类）和softmax的关系： YOLOV3中的逻辑分类应用 Logistic回归 Logistic回归（LR）：是一种常用的处理二分类问题的模型。 二分类...
梯度下降法求解回归问题.rar_sky8gx_softmax _softmax分类器_softmax回归_回归问题
07-13
在"softmax回归"中，softmax函数与线性回归相结合，预测多个类别的概率。文件"mysoftmax_gd_test_fisheriris.m"和"mysoftmax_gd_test_MNIST.m"很可能是测试代码，分别用于鸢尾花数据集（Fisher's Iris）和手写数字...
二分类、多分类、多标签、softmax、sigmoid、pytorch实现
jzlin1997的博客
11-24 5791
二分类、多分类、多标签、softmax、sigmoid、pytorch实现 参考：多标签分类与BCEloss - 简书 (jianshu.com) 25、二分类、多分类与多标签问题的区别 - Andy_George - 博客园 (cnblogs.com) Sigmoid函数_百度百科 (baidu.com) (7条消息) Sigmoid函数_saltriver的专栏-CSDN博客_sigmoid BCELoss — PyTorch 1.7.0 documentation CrossEntropy
Sigmoid和Softmax在二分类中使用的区别与实现
whocare的博客
04-01 4240
对于二分类有三种实现方式： 1. nn.Linear(input, 1) + sigmoid + BCELoss 全连接输出维度为1，使用sigmoid将输出映射到0,1之间 self.outputs = nn.Linear(input, 1) def forward(self, x): # other layers omitted x = self.outputs(x) return torch.sigmoid(x) 那么在这种情况下，我们使用torch.
深度学习笔记1：池化 全连接 激活函数 softmax
qq_41978139的博客
11-09 2441
1. 池化 池化层的输入一般来源于上一个卷积层，主要作用是提供了很强的鲁棒性（例如max-pooling是取一小块区域中的最大值，此时若此区域中的其他值略有变化，或者图像稍有平移，pooling后的结果仍不变），并且减少了参数的数量，防止过拟合现象的发生。池化层一般没有参数，所...
二分类问题，应该选择sigmoid还是softmax函数
小姑仔的博客
03-14 6297
二分类的激活函数通常选择sigmoid或者softmax函数，对于二分类的问题，sigmoid函数和softmax函数可以等价 首先记隐层的输出为h。 如果使用sigmoid的话，输出层有一个结点，值为θh\theta hθh。然后类别1，2的预测概率分别为： p1=sigmoid(θh)=11+e−θhp_{1} = sigmoid(\theta h) = \frac{1}{1+e^{-\theta h}}p1​=sigmoid(θh)=1+e−θh1​ p2=1−sigmoid(θh)=11+eθhp_
torch.nn.softmax(dim),理解Softmax分类器
weixin_46568462的博客
11-17 3284
例如一列数组[1,2,3,4,5,6,7,8,9,10],代入到softamx计算公式之中，我们可以从公式中看出元素的数值越大，softmax算出的值也就越大，对应在图像处理中也就是概率越大。dim=2时，对应维度数值为4，因此每次只有4个元素进行Softmax计算。dim=0时，对应维度数值为2，因此每次只有两个元素进行Softmax计算。dim=1时，对应维度数值为3，因此每次只有3个元素进行Softmax计算。上述两式子合成一个: nn.Softmax(dim=2)(a)
分类任务常用激活函数之nn.Softmax函数参数dim理解（Pytorch）
NorthSmile的博客
08-29 4189
2）dim=1表示该操作将在输入张量的第1维进行计算，而此张量第1维长度为3，故计算实际是沿着第1维方向，将整个张量划分为2×4个切片，每个切片包含三个元素，并分别对每个切片进行softmax计算，使得这三个元素和为1.3）dim=2表示该操作将在输入张量的第2维进行计算，而此张量第2维长度为4，故计算实际是沿着第2维方向，将整个张量划分为2×3个切片，每个切片包含四个元素，并分别对每个切片进行softmax计算，使得这四个元素和为1.该函数使得输入中所有数值变为。.........
【深度学习框架简单梳理】电影评论二分类问题
Shieffer的博客
01-30 1327
文章目录0x00 前言0x01 整体概述0x02 细节展开2.1 数据预处理2.1.1 获取数据2.1.2 处理数据2.2 神经网络模型搭建2.2.1 模型架构2.2.2 层全连接层激活函数2.2.3 优化器和损失函数优化器损失函数2.3 训练验证2.4 优化改进0x03 小结 0x00 前言   最近开始学习深度学习，所以打算以写博客的形式来记录自己的学习过程。   本篇博客的目的是梳理熟悉深度学习的整体框架、神经网络模型及代码的构建以及加深深度学习整个过程的理解。基于目的，选择了较为简单的电影评论分
全连接神经网络的二分类问题
weixin_46824122的博客
04-25 5911
引言 全连接神经网络在分类、回归预测问题上表现出极强的生命力。相比于传统的机器学习算法，神经网路的学习能力大大推动了人工智能往更智能更高效方向发展。人工神经网络（Artificial Neural Networks，简写为ANNs）也简称为神经网络（NNs）或称作连接模型（Connection Model），它是一种模仿动物神经网络行为特征，进行分布式并行信息处理的算法数学模型。这种网络依靠系统的...
基于Pytorch全连接神经网络实现多分类
bigData1994pb的博客
06-09 4045
为了方便开发者应用，PyTorch专门开发了一个视觉工具包torchvision，主要包含以下三个部分： 1.model
softmax回归：多分类模型详解
"这篇文档是关于softmax回归的介绍，它是一种多分类的机器学习模型，常用于处理如MNIST手写数字识别等多类别的问题。文档内容涵盖了softmax回归的基本概念、代价函数、模型参数化特点、权重衰减、与logistic回归的...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值