pytorch LSTM_regression

batch之间传递state

import torch
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt
torch.manual_seed(900)class RNN(nn.Module):def __init__(self):super(RNN, self).__init__()self.lstm = nn.LSTM(input_size=1,hidden_size=2,num_layers=1,batch_first=True)self.out = nn.Linear(2, 1)def forward(self, x,state):# x(batch,seq_len,input_size) h_state(n_layers,batch,hidden_size) r_out(batch,seq_len,hidden_size)r_out,state= self.lstm(x,state) # (batch,seq_len,hidden_size[1]) -> (batch,seq_len,hidden_size[2])outs = self.out(r_out) # (batch,seq_len,hidden_size[2]) -> (batch,seq_len,hidden_size[1])return outs,state
rnn = RNN()optimizer = torch.optim.Adam(rnn.parameters(), lr=0.02) # 同时更新W_hh，W_ih
mse = nn.MSELoss()
state = (torch.randn(1, 1, 2), torch.randn(1, 1, 2))
for step in range(100):# 构建数据start, end = step * np.pi, (step+1)*np.pisteps = torch.linspace(start, end, 10)x = torch.sin(steps).unsqueeze(0).unsqueeze(2)y = torch.cos(steps).unsqueeze(0).unsqueeze(2)# 学习prediction,state= rnn(x,state)state = (state[0].detach(),state[1].detach())loss = mse(prediction, y)optimizer.zero_grad()loss.backward()optimizer.step()# 可视化结果plt.plot(steps, y.numpy().flatten(), 'r-')plt.plot(steps, prediction.detach().numpy().flatten(), 'b-')plt.draw(); plt.pause(0.05)
plt.show()

batch之间不传递state

import torch
import torch.nn as nn
import numpy as np
import matplotlib.pyplot as plt
torch.manual_seed(1000)class RNN(nn.Module):def __init__(self):super(RNN, self).__init__()self.lstm = nn.LSTM(input_size=1,hidden_size=2,num_layers=1,batch_first=True)self.out = nn.Linear(2, 1)def forward(self, x):# x(batch,seq_len,input_size) h_state(n_layers,batch,hidden_size) r_out(batch,seq_len,hidden_size)r_out,_= self.lstm(x,) # (batch,seq_len,hidden_size[1]) -> (batch,seq_len,hidden_size[2])outs = self.out(r_out) # (batch,seq_len,hidden_size[2]) -> (batch,seq_len,hidden_size[1])return outs
rnn = RNN()optimizer = torch.optim.Adam(rnn.parameters(), lr=0.02) # 同时更新W_hh，W_ih
mse = nn.MSELoss()
for step in range(100):# 构建数据start, end = step * np.pi, (step+1)*np.pisteps = torch.linspace(start, end, 10)x = torch.sin(steps).unsqueeze(0).unsqueeze(2)y = torch.cos(steps).unsqueeze(0).unsqueeze(2)# 学习prediction = rnn(x)loss = mse(prediction, y)optimizer.zero_grad()loss.backward()optimizer.step()# 可视化结果plt.plot(steps, y.numpy().flatten(), 'r-')plt.plot(steps, prediction.detach().numpy().flatten(), 'b-')plt.draw(); plt.pause(0.05)
plt.show()

https://zhuanlan.zhihu.com/p/94757947
https://discuss.pytorch.org/t/lstm-how-to-remember-hidden-and-cell-states-across-different-batches/11957
https://pytorch.org/tutorials/beginner/nlp/sequence_models_tutorial.html

pytorch LSTM_regression相关推荐

pyTorch api
应用 pytorch FC_regression pytorch FC_classification pytorch RNN_regression pytorch LSTM_regression py ...
通过anaconda2安装python2.7和安装pytorch
①由于官网下载anaconda2太慢,最好去byrbt下载,然后安装就行 ②安装完anaconda2会自动安装了python2.7(如终端输入python即进入python模式) 但是可能没有设置环境 ...
记录一次简单、高效、无错误的linux上安装pytorch的过程
1 准备miniconda Miniconda Miniconda 可以理解成Anaconda的免费.浓缩版.它非常小,只包含了conda.python以及它们依赖的一些包.我们可以根据我们的需要再安 ...
各种注意力机制PyTorch实现
给出了整个系列的PyTorch的代码实现,以及使用方法. 各种注意力机制 Pytorch implementation of "Beyond Self-attention: External ...
PyTorch代码调试利器_TorchSnooper
GitHub 项目地址: https://github.com/zasdfgbnm/TorchSnooper 大家可能遇到这样子的困扰:比如说运行自己编写的 PyTorch 代码的时候,PyTorch ...
pytorch常用代码
20211228 https://mp.weixin.qq.com/s/4breleAhCh6_9tvMK3WDaw 常用代码段本文代码基于 PyTorch 1.x 版本,需要用到以下包: impo ...
API pytorch tensorflow
pytorch与tensorflow API速查表方法名称 pytroch tensorflow numpy 裁剪 torch.clamp(x, min, max) tf.clip_by_value ...
tensor转换 pytorch tensorflow
一.tensorflow的numpy与tensor互转 1.数组(numpy)转tensor 利用tf.convert_to_tensor(numpy),将numpy转成tensor >> ...
tensor和模型保存与加载 PyTorch
PyTorch教程-7:PyTorch中保存与加载tensor和模型详解保存和读取Tensor PyTorch中的tensor可以保存成 .pt 或者 .pth 格式的文件,使用torch.save ...

pytorch LSTM_regression

pytorch LSTM_regression相关推荐

最新文章

热门文章