pytorch基础

数据类型

dtype，8种
16-bit floating point torch.float16 or torch.half
32-bit floating point torch.float32 or torch.float
64-bit floating point torch.float64 or torch.double

8-bit integer (unsigned) torch.uint8
8-bit integer (signed) torch.int8
16-bit integer (signed) torch.int16 or torch.short
32-bit integer (signed) torch.int32 or torch.int
64-bit integer (signed) torch.int64 or torch.long

pytorch与numpy相互转换：

#tensor
import torch
import numpy as np

np_data = np.arange(6).reshape((2, 3))
torch_data = torch.from_numpy(np_data)
tensor2array = torch_data.numpy()
print(
    '\nnumpy array:', np_data,         
    '\ntorch tensor:', torch_data,      
    '\ntensor to array:', tensor2array, 
)

pytorch函数名称与numpy一致：

# abs 绝对值计算
data = [-1, -2, 1, 2]
tensor = torch.FloatTensor(data)  
print(
    '\nabs',
    '\nnumpy: ', np.abs(data),          
    '\ntorch: ', torch.abs(tensor)     
)

# sin   三角函数 sin
print(
    '\nsin',
    '\nnumpy: ', np.sin(data),      
    '\ntorch: ', torch.sin(tensor)  
)

# mean  均值
print(
    '\nmean',
    '\nnumpy: ', np.mean(data),       
    '\ntorch: ', torch.mean(tensor)    
)

pytorch函数名称 与numpy不一致的函数：

data = [[1,2], [3,4]]
tensor = torch.FloatTensor(data)  
# 正确
print(
    '\nmatrix multiplication (matmul)',
    '\nnumpy: ', np.matmul(data, data),    
    '\ntorch: ', torch.mm(tensor, tensor)   
)

变量

变量是储存张量的篮子。

import torch
from torch.autograd import Variable 

tensor = torch.FloatTensor([[1,2],[3,4]])
#将张量放入变量中
variable = Variable(tensor, requires_grad=True)

print(tensor)

print(variable)

t_out = torch.mean(tensor*tensor)       
v_out = torch.mean(variable*variable)   
print(t_out)
print(v_out)    

反向传播：

v_out.backward()    # 模拟 v_out 的误差反向传递

print(variable.grad)    # 初始 Variable 的梯度

打印结果：

print(variable)     #  Variable 形式

print(variable.data)    # tensor 形式

print(variable.data.numpy())    # numpy 形式

数据集

在torchvision.datasets内置有如下数据集：

• MNIST
• COCO（用于图像标注和目标检测）(Captioning and Detection)
• LSUN Classification
• ImageFolder
• Imagenet-12
• CIFAR10 and CIFAR100
• STL10

datasets.MNIST(
	root, # 根目录
	train=True, # 训练集
	transform=None, # 对数据集进行改变
	target_transform=None, # 对label做改变
	download=False # 是否需要下载
)

#自定义数据集
#继承torch.utils.data.Dataset
Class dateset(Dataset)
 def __len__(self):
 def __getitem__(self, idx):
# len(dataset)
# dataset[i]

迭代数据集

#简单迭代
for i in range(len(dataset)):
  dataset[i]

# 多功能迭代
torch.utils.data.DataLoader
dataloader = DataLoader(dataset, batch_size=4,
            shuffle=True, num_workers=4)

网络层

import torch.nn as nn # class,有内部变量
import torch.nn.functional as F # function，没有内部变量，需要输入参数

线性层

class torch.nn.Linear(in_features, out_features, bias=True)
torch.nn.functional.linear(input, weight, bias=None)

卷积层

class torch.nn.Conv1d(in_channels, out_channels, kernel_size, 
    stride=1, padding=0, dilation=1, groups=1, bias=True)
# dilation: 用于控制内核点之间的距离 - 空洞卷积
# groups: 控制输入和输出之间的连接，参数变小
# 输入: 28*28*2 kenel: 3*3*64 输出：26*26*64   参数量：2*3*3*64
# group1: 28*28*1 3*3*32 26*26*32   1*3*3*32
# group2: 28*28*1 3*3*32 26*26*32   1*3*3*32
torch.nn.functional.conv1d(input, weight, bias=None, 
      stride=1, padding=0, dilation=1, groups=1)
class torch.nn.Conv2d
# 1d 100*1*batchsize->3*1*64->100*64*batchsize
# 2d 224*224*3*batchsize->3*3*3*64->224*224*64*batchsize
# 3d 3*3*3 * 64
class torch.nn.Conv3d

# 
class torch.nn.ConvTranspose1d(in_channels, out_channels, 
        kernel_size, stride=1, padding=0, output_padding=0, groups=1, bias=True)
torch.nn.functional.conv_transpose1d(input, weight, bias=None, 
          stride=1, padding=0, output_padding=0, groups=1)
class torch.nn.ConvTranspose2d
class torch.nn.ConvTranspose3d

池化层

class torch.nn.MaxPool1d(kernel_size, stride=None, 
        padding=0, dilation=1, return_indices=False, ceil_mode=False)
# return_indices - 如果等于True，会返回输出最大值的序号
# ceil_mode - 如果等于True，计算输出信号大小的时候，会使用向上取整，代替默认的向下取整的操作
torch.nn.functional.max_pool1d(input, kernel_size, stride=None, padding=0, 
       dilation=1, ceil_mode=False, return_indices=False)
class torch.nn.MaxPool2d
class torch.nn.MaxPool3d   
class torch.nn.MaxUnpool1d(kernel_size, stride=None, padding=0) 
torch.nn.functional.max_unpool1d(input, indices, 
            kernel_size, stride=None, padding=0, output_size=None)
#2d
#3d

class torch.nn.AvgPool1d(kernel_size, stride=None, padding=0, 
            ceil_mode=False, count_include_pad=True)
torch.nn.functional.avg_pool1d(input, kernel_size, stride=None, padding=0, 
            ceil_mode=False, count_include_pad=True)
#2d
#3d

class torch.nn.FractionalMaxPool2d(kernel_size, output_size=None, output_ratio=None, 
                return_indices=False, _random_samples=None)
#对输入信号提供2维的幂平均池化操作
class torch.nn.LPPool2d(norm_type, kernel_size, stride=None, ceil_mode=False)
torch.nn.functional.lp_pool2d(input, norm_type, kernel_size, stride=None, ceil_mode=False)

#自适应最大池化,当不知道输入是什么的时候使用
class torch.nn.AdaptiveMaxPool1d(output_size, return_indices=False)
torch.nn.functional.adaptive_max_pool1d(input, output_size, return_indices=False)
#2d
# 自适应平均池化
class torch.nn.AdaptiveAvgPool1d(output_size)
torch.nn.functional.adaptive_avg_pool1d(input, output_size)
#2d

激活函数

常用ReLU与LeakyReLU

class torch.nn.ReLU(inplace=False)
torch.nn.functional.relu(input, inplace=False)
class torch.nn.ReLU6(inplace=False)#输出值最大为6
torch.nn.functional.relu6(input, inplace=False)
class torch.nn.ELU(alpha=1.0, inplace=False)
torch.nn.functional.elu(input, alpha=1.0, inplace=False)
class torch.nn.PReLU(num_parameters=1, init=0.25)
torch.nn.functional.prelu(input, weight)
class torch.nn.LeakyReLU(negative_slope=0.01, inplace=False)
torch.nn.functional.leaky_relu(input, negative_slope=0.01, inplace=False)

# 大于threshold，则设置为value 
class torch.nn.Threshold(threshold, value, inplace=False) 
torch.nn.functional.threshold(input, threshold, value, inplace=False)
class torch.nn.Tanh()
torch.nn.functional.tanh(input)
class torch.nn.Hardtanh(min_value=-1, max_value=1, inplace=False)
torch.nn.functional.hardtanh(input, min_val=-1.0, max_val=1.0, inplace=False)
class torch.nn.Sigmoid()
torch.nn.functional.sigmoid(input)
class torch.nn.LogSigmoid()
torch.nn.functional.logsigmoid(input)
class torch.nn.Softplus(beta=1, threshold=20)
torch.nn.functional.softplus(input, beta=1, threshold=20)
class torch.nn.Softshrink(lambd=0.5)
torch.nn.functional.softshrink(input, lambd=0.5)
class torch.nn.Softmin()
#多分类
torch.nn.functional.softmin(input)
class torch.nn.Softmax()
torch.nn.functional.softmax(input)
class torch.nn.LogSoftmax()
torch.nn.functional.log_softmax(input)

class torch.nn.BatchNorm1d(num_features, eps=1e-05, momentum=0.1, affine=True) #affine:是否使用再学习参数
torch.nn.functional.batch_norm(input, running_mean, running_var, 
          weight=None, bias=None, training=False, momentum=0.1, eps=1e-05)
#2d
#3d

RNN

class torch.nn.RNN( args, * kwargs)
#input_size – 输入x的特征数量。
#hidden_size – 隐层的特征数量。
#num_layers – RNN的层数。
#nonlinearity – 指定非线性函数使用tanh还是relu。默认是tanh。
#bias – 如果是False，那么RNN层就不会使用偏置权重 $b_ih$和$b_hh$,默认是True
#dropout – 如果值非零，那么除了最后一层外，其它层的输出都会套上一个dropout层。
#bidirectional – 如果True，将会变成一个双向RNN，默认为False。
class torch.nn.LSTM( args, * kwargs)
class torch.nn.GRU()
#单个RNN细胞
class torch.nn.RNNCell(input_size, hidden_size, bias=True, nonlinearity='tanh')
class torch.nn.LSTMCell(input_size, hidden_size, bias=True)
class torch.nn.GRUCell(input_size, hidden_size, bias=True)

droupout

解决过拟合

class torch.nn.Dropout(p=0.5, inplace=False)
#p - 将元素置0的概率。默认值：0.5
torch.nn.functional.dropout(input, p=0.5, training=False, inplace=False)

Embedding层

class torch.nn.Embedding(num_embeddings, embedding_dim, 
    padding_idx=None, max_norm=None, norm_type=2, scale_grad_by_freq=False, sparse=False)

损失函数

class torch.nn.L1Loss(size_average=True) # 特征选择损失函数
torch.nn.functional.l1_loss(input, target, size_average=True)
class torch.nn.SmoothL1Loss(size_average=True)
torch.nn.functional.smooth_l1_loss(input, target, size_average=True)
class torch.nn.MSELoss(size_average=True) # 回归问题
torch.nn.functional.mse_loss(input, target, size_average=True)
# 多分类
class torch.nn.CrossEntropyLoss(weight=None, size_average=True)
torch.nn.functional.cross_entropy(input, target, weight=None, size_average=True)
# 二分类
class torch.nn.BCELoss(weight=None, size_average=True)
torch.nn.functional.binary_cross_entropy(input, target, weight=None, size_average=True)
#最大似然函数，逻辑回归
class torch.nn.NLLLoss(weight=None, size_average=True)
torch.nn.functional.nll_loss(input, target, weight=None, size_average=True)
class torch.nn.NLLLoss2d(weight=None, size_average=True)
torch.nn.functional.nll_loss2d(input, target, weight=None, size_average=True)
#EM算法，自动编码解码器
class torch.nn.KLDivLoss(weight=None, size_average=True)
torch.nn.functional.kl_div(input, target, size_average=True)

class torch.nn.HingeEmbeddingLoss(size_average=True)
class torch.nn.CosineEmbeddingLoss(margin=0, size_average=True)

填充

torch.nn.functional.pad(input, pad, mode='constant', value=0)

池化与反池化

import torch.nn as nn
import torch.nn.functional as F
#maxunpool
pool = nn.MaxPool2d(2, stride=2, return_indices=True)
unpool = nn.MaxUnpool2d(2, stride=2)
input = Variable(torch.Tensor([[[[ 1,  2,  3,  4],
    [ 5,  6,  7,  8],
    [ 9, 10, 11, 12],
    [13, 14, 15, 16]]]]))
output, indices = pool(input)
print(output,indices)
unpool(output, indices)

优化器

torch.optim
class torch.optim.Optimizer(params, defaults)
# step()
# zero_grad() #梯度清零

class torch.optim.Adadelta(params, lr=1.0, rho=0.9, eps=1e-06, weight_decay=0)
class torch.optim.Adagrad(params, lr=0.01, lr_decay=0, weight_decay=0)
class torch.optim.Adam(params, lr=0.001, betas=(0.9, 0.999), eps=1e-08, weight_decay=0)
class torch.optim.Adamax(params, lr=0.002, betas=(0.9, 0.999), eps=1e-08, weight_decay=0)
class torch.optim.ASGD(params, lr=0.01, lambd=0.0001, alpha=0.75, t0=1000000.0, weight_decay=0)
class torch.optim.LBFGS(params, lr=1, max_iter=20, max_eval=None, tolerance_grad=1e-05, tolerance_change=1e-09, history_size=100, line_search_fn=None)
class torch.optim.RMSprop(params, lr=0.01, alpha=0.99, eps=1e-08, weight_decay=0, momentum=0, centered=False)
class torch.optim.Rprop(params, lr=0.01, etas=(0.5, 1.2), step_sizes=(1e-06, 50))
class torch.optim.SGD(params, lr=, momentum=0, dampening=0, weight_decay=0, nesterov=False)#默认值均为0

模型

两种方法

线性模型

net = torch.nn.Sequential(
    torch.nn.Linear(1, 10),
    torch.nn.ReLU(),
    torch.nn.Linear(10, 1)
)

class LinearNet(torch.nn.Module):
    def __init__(self, n_feature, n_hidden, n_output):
        super(LinearNet, self).__init__()
        self.hidden = torch.nn.Linear(n_feature, n_hidden)
        self.predict = torch.nn.Linear(n_hidden, n_output)

    def forward(self, x):
        x = F.relu(self.hidden(x))
        x = self.predict(x)
        return x

net1 = LinearNet(1, 10, 1) 

卷积模型

class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(  
            nn.Conv2d(
	            in_channels=1,      
	            out_channels=16,   
	            kernel_size=5,     
	            stride=1,         
	            padding=2, 
            ),      
            nn.ReLU(),    
            nn.MaxPool2d(kernel_size=2),   
        )
        self.conv2 = nn.Sequential(  
            nn.Conv2d(16, 32, 5, 1, 2), 
            nn.ReLU(),  
            nn.MaxPool2d(2),  
        )
        self.out = nn.Linear(32 * 7 * 7, 10) 

    def forward(self, x):
        x = self.conv1(x)
        x = self.conv2(x)
        x = x.view(x.size(0), -1)  # reshape
        output = self.out(x)
        return output
cnn = CNN()
print(cnn) 

RNN

class RNN(nn.Module):
    def __init__(self):
        super(RNN, self).__init__()

        self.rnn = nn.LSTM(     
            input_size=28,      
            hidden_size=64,     
            num_layers=1,      
            batch_first=True,   
        )

        self.out = nn.Linear(64, 10)    

    def forward(self, x):   
        # x shape (batch, time_step, input_size)
        # r_out shape (batch, time_step, output_size)
        # h_n shape (n_layers, batch, hidden_size) .
        # h_c shape (n_layers, batch, hidden_size)     
        r_out, (h_n, h_c) = self.rnn(x, None)   
        # 选取最后一个时间点的 r_out 输出
        # 这里 r_out[:, -1, :] 的值也是 h_n 的值
        out = self.out(r_out[:, -1, :])
        return out

rnn = RNN()
print(rnn)

gpu加速

将所有数据和操作加上.cuda()

x = x.cuda() 
y = y.cuda() 
cnn = CNN()
cnn.cuda()  

保存与加载

torch.save(net, 'net.pkl')  #保存整个网络
torch.save(net.state_dict(), 'net_params.pkl')   #只保存网络中的参数

#提取
net = torch.load('net.pkl')
net3 = xxx
net3.load_state_dict(torch.load('net_params.pkl'))

minist

import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision      
import matplotlib.pyplot as plt

# Hyper Parameters
EPOCH = 10          
BATCH_SIZE = 50
LR = 0.001        # 学习率
DOWNLOAD_MNIST = True  # 如果你已经下载好了mnist数据就写上 False

train_data = torchvision.datasets.MNIST(
    root='./mnist/',    # 保存或者提取位置
    train=True,  # this is training data
    transform=torchvision.transforms.ToTensor(), 
    download=DOWNLOAD_MNIST # 没下载就下载, 下载了就不用再下了
)

test_data = torchvision.datasets.MNIST(root='./mnist/', train=False)

train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)

test_x = torch.unsqueeze(test_data.test_data, dim=1).type(torch.FloatTensor)/255.  
test_y = test_data.test_labels

class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(  
            nn.Conv2d(
	            in_channels=1,      
	            out_channels=16,   
	            kernel_size=5,     
	            stride=1,         
	            padding=2, 
            ),      
            nn.ReLU(),    
            nn.MaxPool2d(kernel_size=2),   
        )
        self.conv2 = nn.Sequential(  
            nn.Conv2d(16, 32, 5, 1, 2), 
            nn.ReLU(),  
            nn.MaxPool2d(2),  
        )
        self.out = nn.Linear(32 * 7 * 7, 10) 
      
    def forward(self, x):
        x = self.conv1(x)
        x = self.conv_new(x)
        x = self.conv2(x)
        x = x.view(x.size(0), -1)  
        output = self.out(x)
        return output
        
cnn=CNN()
optimizer = torch.optim.Adam(cnn.parameters(), lr=LR)   # optimize all cnn parameters
loss_func = nn.CrossEntropyLoss()   # the target label is not one-hotted

# training and testing
for epoch in range(EPOCH):
    for step, (b_x, b_y) in enumerate(train_loader):   # 分配 batch data, normalize x when iterate train_loader
        output = cnn(b_x)               # cnn output
        loss = loss_func(output, b_y)   # cross entropy loss
        optimizer.zero_grad()           # clear gradients for this training step
        loss.backward()              # backpropagation, compute gradients
        optimizer.step()              # apply gradients

    test_output = cnn(test_x)
    pred_y = torch.max(test_output, 1)[1].data.numpy()
    accuracy = float((pred_y == test_y.data.numpy()).astype(int).sum()) / float(test_y.size(0))
    print('Epoch: ', epoch, '| train loss: %.4f' % loss.data.numpy(), '| test accuracy: %.2f' % accuracy)

GPU

test_x = torch.unsqueeze(test_data.test_data, dim=1).type(torch.FloatTensor).cuda()/255.  
test_y = test_data.test_labels.cuda()
cnn.cuda()
optimizer = torch.optim.Adam(cnn.parameters(), lr=LR)   # optimize all cnn parameters
loss_func = nn.CrossEntropyLoss()   # the target label is not one-hotted

for epoch in range(EPOCH):
    for step, (b_x, b_y) in enumerate(train_loader):   # 分配 batch data, normalize x when iterate train_loader
        b_x=b_x.cuda()
        b_y=b_y.cuda()
        output = cnn(b_x)               # cnn output
        loss = loss_func(output, b_y)   # cross entropy loss
        optimizer.zero_grad()           # clear gradients for this training step
        loss.backward()              # backpropagation, compute gradients
        optimizer.step()              # apply gradients

    test_output = cnn(test_x)
    pred_y = torch.max(test_output, 1)[1].cuda().data
    accuracy = torch.sum(pred_y == test_y).cuda().type(torch.FloatTensor) / test_y.size(0)
    print('Epoch: ', epoch, '| train loss: %.4f' % loss.data, '| test accuracy: %.2f' % accuracy)