RNN的种类

上图中的红色方块代表输入，蓝色方块代表输出，绿色方块代表

一对一（one to one)

一对一的结构一般指经典的卷积神经网络和全连接神经网络，被誉为香草神经网络，它们处理固定大小的输入，并输出固定大小的输出，主要用于图像分类任务。

下面介绍循环神经网络RNN的几种结构和其应用：

多对一（many to one）

多对一结构用于分类，如情感分类（输入一个句子，判断是积极还是消极的情感）

一对多（one to many)

一对多可用于生成，如音乐生成，还可以用于图像字幕（输入一张图片，输出一个文本句子）

多对多（many to many)

多对多结构可用于机器翻译（输入一个英文句子，输出一个中文句子）和视频分类（视频中的多个帧作为输入，对视频中的每一帧进行分类）

RNN多对一用于图像分类

下面介绍使用RNN进行分类（图像分类）的方法。RNN用于分类任务的结构是多对一结构。假设图像的尺寸是28x28的黑白图片，将每一个张图片作为RNN的输入，每一行作为一个输入特征，所有的行构成一个序列。

RNN的输出只选择最后一行的输出（使用out[:, -1, :]），这就是多对一的精髓。最后再套一个全连接层，使用交叉熵作为判断准则。

RNN图像分类的示例代码

import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt
from torch.utils.data import DataLoaderimport os
os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"# Device configuration
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')# Hyper-parameters 
# input_size = 784  # 28x28
hidden_size = 128
num_classes = 10
num_epochs = 2
batch_size = 100
learning_rate = 0.001input_size = 28
sequence_length = 28
num_layers = 2# MNIST dataset 
train_dataset = torchvision.datasets.MNIST(root='./data', train=True, transform=transforms.ToTensor(),  download=True)test_dataset = torchvision.datasets.MNIST(root='./data', train=False, transform=transforms.ToTensor())# Data loader
train_loader = DataLoader(dataset=train_dataset,batch_size=batch_size, shuffle=True)test_loader = DataLoader(dataset=test_dataset,batch_size=batch_size, shuffle=False)examples = iter(test_loader)
example_data, example_targets = next(examples)# for i in range(6):
#     plt.subplot(2,3,i+1)
#     plt.imshow(example_data[i][0], cmap='gray')
# plt.show()# Fully connected neural network with one hidden layer
class RNN(nn.Module):def __init__(self, input_size, hidden_size, num_layers, num_classes):super(RNN, self).__init__()self.num_layers = num_layersself.hidden_size = hidden_sizeself.rnn = nn.RNN(input_size, hidden_size, num_layers, batch_first=True)#  x: (batch_size, sequence_len, input_size)self.fc = nn.Linear(hidden_size, num_classes)def forward(self, x):h0 = torch.zeros(self.num_layers, x.size(0), self.hidden_size).to(device)out, _ = self.rnn(x, h0)# out: batch_size, seq_length, hidden_size# out: (N, 28, 128)out = out[:, -1, :]# out (N, 128)out = self.fc(out)return outmodel = RNN(input_size, hidden_size, num_layers,num_classes).to(device)# Loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)  # Train the model
n_total_steps = len(train_loader)
for epoch in range(num_epochs):for i, (images, labels) in enumerate(train_loader):  # origin shape: [100, 1, 28, 28]# resized: [100, 28, 28]images = images.reshape(-1, sequence_length, input_size).to(device)labels = labels.to(device)# Forward passoutputs = model(images)loss = criterion(outputs, labels)# Backward and optimizeoptimizer.zero_grad()loss.backward()optimizer.step()if (i+1) % 100 == 0:print (f'Epoch [{epoch+1}/{num_epochs}], Step [{i+1}/{n_total_steps}], Loss: {loss.item():.4f}')# Test the model
# In test phase, we don't need to compute gradients (for memory efficiency)
with torch.no_grad():n_correct = 0n_samples = 0for images, labels in test_loader:images = images.reshape(-1, sequence_length, input_size).to(device)labels = labels.to(device)outputs = model(images)# max returns (value ,index)_, predicted = torch.max(outputs.data, 1)n_samples += labels.size(0)n_correct += (predicted == labels).sum().item()acc = 100.0 * n_correct / n_samplesprint(f'Accuracy of the network on the 10000 test images: {acc} %')