'''
Author: 365JHWZGo
Description: rnn--回归
Date: 2021/11/4 14:29
FilePath: day1104-1.py
'''# 导包
import torch
import torchvision
import torch.autograd as Variable
import matplotlib.pyplot as plt
import numpy as nptorch.manual_seed(1)# 超参数设置
BATCH_SIZE = 1
LR = 0.02
TIME_STEP = 10
INPUT_SIZE = 1
DOWNLOAD_MNIST = False# RNN
class RNN(torch.nn.Module):def __init__(self):super(RNN, self).__init__()self.rnn = torch.nn.RNN(input_size=INPUT_SIZE,hidden_size=32,batch_first=True,num_layers=1)self.out = torch.nn.Linear(32, 1)def forward(self, x, h_state):# x (batch, time_step, input_size)# h_state (n_layers, batch, hidden_size)# r_out (batch, time_step, output_size[hidden_size])r_out, h_state = self.rnn(x, h_state)outs = []for time_step in range(r_out.size(1)):outs.append(self.out(r_out[:, time_step, :]))#r_out[:, time_step, :]=>shape(1,32)->[[1,2,...32]]#self.out(r_out[:, time_step, :])=>shape(1,1)->[[1]]#outs =>  [tensor([[1.]]), tensor([[1.]]), tensor([[1.]])...]return torch.stack(outs, dim=1), h_state#torch.stack(outs, dim=1) => shape(1,10,1)#详解了解：https://blog.csdn.net/qq_44833392/article/details/121146516#[[[32],[32],..]]#每一个第一维中的第二维的全部#详细了解：https://blog.csdn.net/qq_44833392/article/details/121131442?spm=1001.2014.3001.5501#eg： a =# [#     [#        [1],[2],[3],[4]#      ],#      [#        [11],[22],[33],[44]#      ],#       [#          [111],[222],[333],[444]#       ]#   ]# 此时a的shape=(3,4,1)# a[:, 0, :] 截取所有第一维中第二维中的第一个的全部# => [[1],[11],[111]] shape=(3,1)#实例化
rnn = RNN()#优化器
optimizer = torch.optim.Adam(rnn.parameters(),lr = LR)#损失函数
loss_func = torch.nn.MSELoss()#初始化h_state
h_state = Noneif __name__ == '__main__':for step in range(60):start, end = step * np.pi, (step + 1) * np.pi#将一个pi分成均匀分成10份，他们的类型是np.float32steps = np.linspace(start,end,10,dtype=np.float32)x_np = np.sin(steps)y_np = np.cos(steps)#将numpy数据转化为tensor x.shape=>(batch_size,time_step,input_size)=(1,10,1)x = torch.from_numpy(x_np[np.newaxis,:,np.newaxis])#给x_np的行和列均添加一个维度#eg: a = [1,2,3,4,5]#    a[np.newaxis,:,np.newaxis]#  =>[[[1],[2],[3],[4],[5]]] shape=(1,5,1)y = torch.from_numpy(y_np[np.newaxis,:,np.newaxis])prediction , h_state = rnn(x,h_state)loss = loss_func(prediction,y)optimizer.zero_grad()loss.backward()optimizer.step()