【发布时间】：2018-09-23 08:14:31
【问题描述】：
我试图在 pytorch 中跟踪整个数据集的误差并计算整个数据集的误差。我在 cifar10 pytorch 0.3.1 中写了以下（可重现的示例并且完全包含）：
import torch
from torch.autograd import Variable
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from math import inf
from pdb import set_trace as st
def error_criterion(outputs,labels):
max_vals, max_indices = torch.max(outputs,1)
train_error = (max_indices != labels).sum().data[0]/max_indices.size()[0]
return train_error
def evalaute_mdl_data_set(loss,error,net,dataloader,enable_cuda,iterations=inf):
'''
Evaluate the error of the model under some loss and error with a specific data set.
'''
running_loss,running_error = 0,0
for i,data in enumerate(dataloader):
if i >= iterations:
break
inputs, labels = extract_data(enable_cuda,data,wrap_in_variable=True)
outputs = net(inputs)
running_loss += loss(outputs,labels).data[0]
running_error += error(outputs,labels)
return running_loss/(i+1),running_error/(i+1)
def extract_data(enable_cuda,data,wrap_in_variable=False):
inputs, labels = data
if enable_cuda:
inputs, labels = inputs.cuda(), labels.cuda() #TODO potential speed up?
if wrap_in_variable:
inputs, labels = Variable(inputs), Variable(labels)
return inputs, labels
def train_and_track_stats(enable_cuda, nb_epochs, trainloader,testloader, net,optimizer,criterion,error_criterion, iterations=inf):
''' Add stats before training '''
train_loss_epoch, train_error_epoch = evalaute_mdl_data_set(criterion, error_criterion, net, trainloader, enable_cuda, iterations)
test_loss_epoch, test_error_epoch = evalaute_mdl_data_set(criterion, error_criterion, net, testloader, enable_cuda, iterations)
print(f'[-1, -1], (train_loss: {train_loss_epoch}, train error: {train_error_epoch}) , (test loss: {test_loss_epoch}, test error: {test_error_epoch})')
##
''' Start training '''
print('about to start training')
for epoch in range(nb_epochs): # loop over the dataset multiple times
running_train_loss,running_train_error = 0.0,0.0
for i,data_train in enumerate(trainloader):
''' zero the parameter gradients '''
optimizer.zero_grad()
''' train step = forward + backward + optimize '''
inputs, labels = extract_data(enable_cuda,data_train,wrap_in_variable=True)
outputs = net(inputs)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
running_train_loss += loss.data[0]
running_train_error += error_criterion(outputs,labels)
''' End of Epoch: collect stats'''
train_loss_epoch, train_error_epoch = running_train_loss/(i+1), running_train_error/(i+1)
#train_loss_epoch, train_error_epoch = evalaute_mdl_data_set(criterion,error_criterion,net,trainloader,enable_cuda,iterations)
test_loss_epoch, test_error_epoch = evalaute_mdl_data_set(criterion,error_criterion,net,testloader,enable_cuda,iterations)
print(f'[{epoch}, {i+1}], (train_loss: {train_loss_epoch}, train error: {train_error_epoch}) , (test loss: {test_loss_epoch}, test error: {test_error_epoch})')
return train_loss_epoch, train_error_epoch, test_loss_epoch, test_error_epoch
class Flatten(torch.nn.Module):
def forward(self, input):
return input.view(input.size(0), -1)
def main():
enable_cuda = True
print('running main')
num_workers = 0
''' Get Data set '''
batch_size_test = 10000
batch_size_train = 10000
data_path = './data'
transform = [transforms.ToTensor(),transforms.Normalize( (0.5, 0.5, 0.5), (0.5, 0.5, 0.5) )]
transform = transforms.Compose(transform)
trainset = torchvision.datasets.CIFAR10(root=data_path, train=True,download=False, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size_train,shuffle=True, num_workers=num_workers)
testset = torchvision.datasets.CIFAR10(root=data_path, train=False,download=False, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size_test,shuffle=False, num_workers=num_workers)
''' Get model '''
net = torch.nn.Sequential(
torch.nn.Conv2d(3,13,5), #(in_channels, out_channels, kernel_size),
Flatten(),
torch.nn.Linear(28*28*13, 13),
torch.nn.Linear(13, 10)
)
net.cuda()
''' Train '''
nb_epochs = 10
lr = 0.1
err_criterion = error_criterion
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=lr, momentum=0.0)
train_and_track_stats(enable_cuda, nb_epochs, trainloader,testloader, net,optimizer,criterion,err_criterion, iterations=inf)
''' Done '''
print('Done')
if __name__ == '__main__':
main()
当我运行它时，我收到以下错误：
python my_cifar10.py
running main
[-1, -1], (train_loss: 2.3172860145568848, train error: 0.0054) , (test loss: 2.317185878753662, test error: 0.0038)
about to start training
[0, 5], (train_loss: 2.22599835395813, train error: 0.015160000000000002) , (test loss: 2.0623881816864014, test error: 0.0066)
[1, 5], (train_loss: 2.014406657218933, train error: 0.00896) , (test loss: 1.9619578123092651, test error: 0.0195)
[2, 5], (train_loss: 1.9428715705871582, train error: 0.01402) , (test loss: 1.918603539466858, test error: 0.0047)
[3, 5], (train_loss: 1.9434458494186402, train error: 0.01192) , (test loss: 1.9194672107696533, test error: 0.0125)
[4, 5], (train_loss: 1.8804980754852294, train error: 0.00794) , (test loss: 1.8549214601516724, test error: 0.004)
[5, 5], (train_loss: 1.8573726177215577, train error: 0.010159999999999999) , (test loss: 1.8625996112823486, test error: 0.0158)
[6, 5], (train_loss: 1.8454653739929199, train error: 0.01524) , (test loss: 1.8155865669250488, test error: 0.0122)
[7, 5], (train_loss: 1.8140610456466675, train error: 0.01066) , (test loss: 1.808283805847168, test error: 0.0101)
[8, 5], (train_loss: 1.8036894083023072, train error: 0.00832) , (test loss: 1.799634575843811, test error: 0.007)
[9, 5], (train_loss: 1.8023016452789307, train error: 0.0077399999999999995) , (test loss: 1.8030155897140503, test error: 0.0114)
Done
显然它一定是错误的，因为对于一个非常小而简单的模型（1 conv 2 fcs），测试误差几乎为零。
代码看起来很简单，我不知道出了什么问题。这几天我一直在做事情和改变事情。有什么新建议可以尝试吗？
【问题讨论】：
-
我无法重现该错误。我只是在 cpu 上测试并得到
[-1, -1], (train_loss: 2.3224891445520894, train error: 0.8735254156010229) , (test loss: 2.3210103873965107, test error: 0.8691653481012658) about to start training [0, 391], (train_loss: 1.8232954066732656, train error: 0.6340393222506394) , (test loss: 1.7517535837390754, test error: 0.6045292721518988) ...。损失是有道理的：如果你有 10 个随机猜测的类，你会在大约 1/10 的时间里是对的，从而导致交叉熵损失为ln(-1/10) = 2.3。所以错误应该在error_criterion函数中。 -
我之前无法重现，因为我将批量大小更改为 128
-
@McLawrence 这似乎是字节张量或其他东西的溢出错误。相关帖子：discuss.pytorch.org/t/… 感谢您的帮助！
标签： python machine-learning neural-network conv-neural-network pytorch