pytorch学習(8)RNNにGPU cuda()を使用する

cuda()を追加する前に、運転時間は55 sです.cuda()を追加すると、実行時間は6 sです.

#coding=utf-8
import torch
import torch.nn as nn
import torch.utils.data as Data
import torchvision      #      
from torch.autograd import Variable
import time
'''
cuda()   ，      、       。
'''

time_start=time.time()


torch.manual_seed(1)    # reproducible

# Hyper Parameters
EPOCH = 1           #          ,       ,        
BATCH_SIZE = 64
TIME_STEP = 28      # rnn      /     
INPUT_SIZE = 28     # rnn       /       
LR = 0.01           # learning rate
DOWNLOAD_MNIST = False  #          mnist      Fasle


# Mnist     
train_data = torchvision.datasets.MNIST(
    root='./mnist/',    #         
    train=True,  # this is training data
    transform=torchvision.transforms.ToTensor(),    #    PIL.Image or numpy.ndarray  
                                                    # torch.FloatTensor (C x H x W),       normalize   [0.0, 1.0]   
    download=DOWNLOAD_MNIST,          #       ,          
)


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

#     50samples, 1 channel, 28x28 (50, 1, 28, 28)
train_loader = Data.DataLoader(dataset=train_data, batch_size=BATCH_SIZE, shuffle=True)

#       ,          2000 
test_x = Variable(torch.unsqueeze(test_data.test_data, dim=1)).type(torch.FloatTensor)[:2000].cuda()/255.   # shape from (2000, 28, 28) to (2000, 1, 28, 28), value in range(0,1)
test_y = test_data.test_labels[:2000].cuda()

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

        self.rnn = nn.LSTM(     # LSTM      nn.RNN()    
            input_size=INPUT_SIZE,      #           
            hidden_size=64,     # rnn hidden unit
            num_layers=1,       #     RNN layers
            batch_first=True,   # input & output     batch size           e.g. (batch, time_step, input_size)
        )

        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)   LSTM     hidden states, h_n    , h_c    
        # h_c shape (n_layers, batch, hidden_size)
        r_out, (h_n, h_c) = self.rnn(x, None)   # None    hidden state    0  state,       h

        #            r_out   
        #    r_out[:, -1, :]      h_n   
        out = self.out(r_out[:, -1, :])  #         
        return out

rnn = RNN()
rnn.cuda()    #     cuda()
print(rnn)

optimizer = torch.optim.Adam(rnn.parameters(), lr=LR)   # optimize all parameters
loss_func = nn.CrossEntropyLoss()   # the target label is not one-hotted

# training and testing
for epoch in range(EPOCH):
    for step, (x, b_y) in enumerate(train_loader):   # gives batch data
        x = Variable(x).cuda()
        b_y = Variable(b_y).cuda()     #       cuda（）
        b_x = x.view(-1, 28, 28)   # reshape x to (batch, time_step, input_size)

        output = rnn(b_x)               # rnn 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 = rnn(test_x[:10].view(-1, 28, 28))
pred_y = torch.max(test_output, 1)[1].cpu().data.numpy().squeeze()   #    cpu（）,    ：can't convert CUDA tensor to numpy (it doesn't support GPU arrays). Use .cpu() to move the tensor to host memory first.
print(pred_y, 'prediction number')
print(test_y[:10], 'real number')

time_end=time.time()
print('time cost',time_end-time_start,'s')

結果:

RNN (
  (rnn): LSTM(28, 64, batch_first=True)
  (out): Linear (64 -> 10)
)
(array([7, 2, 1, 0, 4, 1, 4, 9, 5, 9]), 'prediction number')
(
 7
 2
 1
 0
 4
 1
 4
 9
 5
 9
[torch.cuda.LongTensor of size 10 (GPU 0)]
, 'real number')
('time cost', 6.428691864013672, 's')