注意:博主没有重写d2l的源代码文件,而是创建了一个新的python文件,并重写了该方法。
一、代码运行日志
C:\Users\Administrator\anaconda3\envs\limu\python.exe G:/PyCharmProjects/limu-d2l/ch03/softmax_regression.py
Traceback (most recent call last):File "<string>", line 1, in <module>
Traceback (most recent call last):File "<string>", line 1, in <module>File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 116, in spawn_mainFile "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 116, in spawn_mainexitcode = _main(fd, parent_sentinel)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 125, in _mainTraceback (most recent call last):
exitcode = _main(fd, parent_sentinel)File "<string>", line 1, in <module>File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 125, in _main
Traceback (most recent call last):File "<string>", line 1, in <module>
prepare(preparation_data)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 236, in prepareprepare(preparation_data)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 236, in prepare_fixup_main_from_path(data['init_main_from_path'])File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_path_fixup_main_from_path(data['init_main_from_path'])File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_pathmain_content = runpy.run_path(main_path,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 265, in run_pathmain_content = runpy.run_path(main_path,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 265, in run_pathreturn _run_module_code(code, init_globals, run_name,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 97, in _run_module_codereturn _run_module_code(code, init_globals, run_name,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 97, in _run_module_code_run_code(code, mod_globals, init_globals,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 87, in _run_code_run_code(code, mod_globals, init_globals,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 87, in _run_codeFile "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 116, in spawn_mainFile "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 116, in spawn_mainexec(code, run_globals)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 97, in <module>
exec(code, run_globals)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 97, in <module>train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 81, in train_ch03
exitcode = _main(fd, parent_sentinel)
train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 125, in _mainFile "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 81, in train_ch03exitcode = _main(fd, parent_sentinel)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 125, in _maintrain_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)train_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 64, in train_epoch_ch03File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 64, in train_epoch_ch03prepare(preparation_data)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 236, in prepareprepare(preparation_data)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 236, in preparefor X, y in train_iter:for X, y in train_iter:File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 441, in __iter__File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 441, in __iter___fixup_main_from_path(data['init_main_from_path'])_fixup_main_from_path(data['init_main_from_path'])File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_pathFile "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 287, in _fixup_main_from_pathmain_content = runpy.run_path(main_path,main_content = runpy.run_path(main_path,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 265, in run_pathFile "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 265, in run_pathreturn self._get_iterator()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 388, in _get_iteratorreturn self._get_iterator()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 388, in _get_iteratorreturn _run_module_code(code, init_globals, run_name,return _run_module_code(code, init_globals, run_name,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 97, in _run_module_codeFile "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 97, in _run_module_codereturn _MultiProcessingDataLoaderIter(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1042, in __init__return _MultiProcessingDataLoaderIter(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1042, in __init___run_code(code, mod_globals, init_globals,
_run_code(code, mod_globals, init_globals,File "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 87, in _run_codeFile "C:\Users\Administrator\anaconda3\envs\limu\lib\runpy.py", line 87, in _run_codeexec(code, run_globals)
exec(code, run_globals)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 97, in <module>File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 97, in <module>train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 81, in train_ch03
train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 81, in train_ch03w.start()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\process.py", line 121, in startw.start()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\process.py", line 121, in starttrain_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)train_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 64, in train_epoch_ch03File "G:\PyCharmProjects\limu-d2l\ch03\softmax_regression.py", line 64, in train_epoch_ch03self._popen = self._Popen(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 224, in _Popenfor X, y in train_iter:for X, y in train_iter:File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 441, in __iter__File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 441, in __iter__self._popen = self._Popen(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 224, in _Popenreturn _default_context.get_context().Process._Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 327, in _Popenreturn _default_context.get_context().Process._Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 327, in _Popenreturn self._get_iterator()return self._get_iterator()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 388, in _get_iteratorFile "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 388, in _get_iteratorreturn Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\popen_spawn_win32.py", line 45, in __init__return Popen(process_obj) return _MultiProcessingDataLoaderIter(self)return _MultiProcessingDataLoaderIter(self)prep_data = spawn.get_preparation_data(process_obj._name)File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1042, in __init__File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1042, in __init__File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 154, in get_preparation_dataFile "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\popen_spawn_win32.py", line 45, in __init__prep_data = spawn.get_preparation_data(process_obj._name)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 154, in get_preparation_data
_check_not_importing_main()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 134, in _check_not_importing_main_check_not_importing_main()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 134, in _check_not_importing_mainraise RuntimeError('''
RuntimeError: An attempt has been made to start a new process before thecurrent process has finished its bootstrapping phase.This probably means that you are not using fork to start yourchild processes and you have forgotten to use the proper idiomin the main module:if __name__ == '__main__':freeze_support()...The "freeze_support()" line can be omitted if the programis not going to be frozen to produce an executable.w.start()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\process.py", line 121, in startw.start()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\process.py", line 121, in startraise RuntimeError('''
RuntimeError: An attempt has been made to start a new process before thecurrent process has finished its bootstrapping phase.This probably means that you are not using fork to start yourchild processes and you have forgotten to use the proper idiomin the main module:if __name__ == '__main__':freeze_support()...The "freeze_support()" line can be omitted if the programis not going to be frozen to produce an executable.self._popen = self._Popen(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 224, in _Popenself._popen = self._Popen(self)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 224, in _Popenreturn _default_context.get_context().Process._Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 327, in _Popen
return _default_context.get_context().Process._Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\context.py", line 327, in _Popenreturn Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\popen_spawn_win32.py", line 45, in __init__return Popen(process_obj)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\popen_spawn_win32.py", line 45, in __init__prep_data = spawn.get_preparation_data(process_obj._name)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 154, in get_preparation_dataprep_data = spawn.get_preparation_data(process_obj._name)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 154, in get_preparation_data_check_not_importing_main()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 134, in _check_not_importing_main_check_not_importing_main()File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\spawn.py", line 134, in _check_not_importing_mainraise RuntimeError('''raise RuntimeError('''
RuntimeError: RuntimeErrorAn attempt has been made to start a new process before thecurrent process has finished its bootstrapping phase.This probably means that you are not using fork to start yourchild processes and you have forgotten to use the proper idiomin the main module:if __name__ == '__main__':freeze_support()...The "freeze_support()" line can be omitted if the programis not going to be frozen to produce an executable.: An attempt has been made to start a new process before thecurrent process has finished its bootstrapping phase.This probably means that you are not using fork to start yourchild processes and you have forgotten to use the proper idiomin the main module:if __name__ == '__main__':freeze_support()...The "freeze_support()" line can be omitted if the programis not going to be frozen to produce an executable.
Traceback (most recent call last):File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1132, in _try_get_datadata = self._data_queue.get(timeout=timeout)File "C:\Users\Administrator\anaconda3\envs\limu\lib\multiprocessing\queues.py", line 108, in getraise Empty
_queue.EmptyThe above exception was the direct cause of the following exception:Traceback (most recent call last):File "G:/PyCharmProjects/limu-d2l/ch03/softmax_regression.py", line 97, in <module>train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)File "G:/PyCharmProjects/limu-d2l/ch03/softmax_regression.py", line 81, in train_ch03train_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)File "G:/PyCharmProjects/limu-d2l/ch03/softmax_regression.py", line 64, in train_epoch_ch03for X, y in train_iter:File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 633, in __next__data = self._next_data()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1328, in _next_dataidx, data = self._get_data()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1294, in _get_datasuccess, data = self._try_get_data()File "C:\Users\Administrator\anaconda3\envs\limu\lib\site-packages\torch\utils\data\dataloader.py", line 1145, in _try_get_dataraise RuntimeError('DataLoader worker (pid(s) {}) exited unexpectedly'.format(pids_str)) from e
RuntimeError: DataLoader worker (pid(s) 14032, 23312, 21048, 1952) exited unexpectedlyProcess finished with exit code 1
二、问题分析
这个错误是由于在使用多进程 DataLoader 时出现的问题,通常与 Windows 操作系统相关。在 Windows 上,使用多进程的 DataLoader 可能会导致一些问题,这与 Windows 的进程模型不太兼容。
三、解决方案(使用单进程 DataLoader)
在 Windows 上,将 DataLoader 的 num_workers
参数设置为 0,以使用单进程 DataLoader。这会禁用多进程加载数据,虽然可能会导致数据加载速度变慢,但通常可以解决与多进程 DataLoader 相关的问题。
d2l.load_data_fashion_mnist(batch_size)源代码
def get_dataloader_workers():"""Use 4 processes to read the data.Defined in :numref:`sec_utils`"""return 4def load_data_fashion_mnist(batch_size, resize=None):"""Download the Fashion-MNIST dataset and then load it into memory.Defined in :numref:`sec_utils`"""trans = [transforms.ToTensor()]if resize:trans.insert(0, transforms.Resize(resize))trans = transforms.Compose(trans)mnist_train = torchvision.datasets.FashionMNIST(root="../data", train=True, transform=trans, download=True)mnist_test = torchvision.datasets.FashionMNIST(root="../data", train=False, transform=trans, download=True)return (torch.utils.data.DataLoader(mnist_train, batch_size, shuffle=True,num_workers=get_dataloader_workers()),torch.utils.data.DataLoader(mnist_test, batch_size, shuffle=False,num_workers=get_dataloader_workers()))
在代码中使用修改后的load_data_fashion_mnist函数
def load_data_fashion_mnist(batch_size, resize=None, num_workers=4):"""下载Fashion-MNIST数据集,然后将其加载到内存中"""trans = [transforms.ToTensor()]if resize:trans.index(0, transforms.Resize(resize))trans = transforms.Compose(trans)mnist_train = torchvision.datasets.FashionMNIST(root='../data', train=True, transform=trans, download=True)mnist_test = torchvision.datasets.FashionMNIST(root='../data', train=False, transform=trans, download=True)return (data.DataLoader(mnist_train, batch_size, shuffle=True, num_workers=num_workers),data.DataLoader(mnist_test, batch_size, shuffle=False, num_workers=num_workers))
train_iter, test_iter = fashion_mnist.load_data_fashion_mnist(batch_size, num_workers=0)
四、为什么将 DataLoader 的 num_workers 参数设置为 0,是使用的单进程,而不是零进程呢?
在 PyTorch 的 DataLoader 中,num_workers
参数控制了用于加载数据的子进程数量。当 num_workers
被设置为 0 时,实际上是表示不使用任何子进程来加载数据,即单进程加载数据。
为什么不是零进程?这是因为 DataLoader 需要至少一个进程来加载数据,这个进程被称为主进程。主进程负责数据加载和分发给训练的进程。当 num_workers
设置为 0 时,只有主进程用于加载和处理数据,没有额外的子进程。这是一种单进程的数据加载方式。
如果将 num_workers
设置为 1,则会有一个额外的子进程来加载数据,总共会有两个进程:一个主进程和一个数据加载子进程。这种设置可以在某些情况下提高数据加载的效率,特别是当数据加载耗时较长时,子进程可以并行地加载数据,从而加速训练过程。
五、完整训练代码
import torch
from d2l import torch as d2l
import fashion_mnistbatch_size = 256
train_iter, test_iter = fashion_mnist.load_data_fashion_mnist(batch_size, num_workers=0)# 初始化模型参数
num_inputs = 784 # 每个输入图像的通道数为1, 高度和宽度均为28像素
num_outputs = 10W = torch.normal(0, 0.01, size=(num_inputs, num_outputs), requires_grad=True)
b = torch.zeros(num_outputs, requires_grad=True)# 定义softmax操作
def softmax(X):"""矩阵中的非常大或非常小的元素可能造成数值上溢或者下溢解决方案: P84 3.7.2 重新审视softmax的实现"""X_exp = torch.exp(X)partition = X_exp.sum(1, keepdim=True)return X_exp / partition# 定义模型
def net(X):return softmax(torch.matmul(X.reshape((-1, W.shape[0])), W) + b)# 定义损失函数
def cross_entropy(y_hat, y):return - torch.log(y_hat[range(len(y_hat)), y])# 分类精度
def accuracy(y_hat, y):"""计算预测正确的数量"""if len(y_hat.shape) > 1 and y_hat.shape[1] > 1:y_hat = y_hat.argmax(axis=1)cmp = y_hat.type(y.dtype) == yreturn float(cmp.type(y.dtype).sum())def evaluate_accuracy(net, data_iter):"""计算在制定数据集上模型的精度"""if isinstance(net, torch.nn.Module):net.eval()metric = d2l.Accumulator(2)with torch.no_grad():for X, y in data_iter:metric.add(accuracy(net(X), y), y.numel())return metric[0] / metric[1]# 训练
def train_epoch_ch03(net, train_iter, loss, updater):if isinstance(net, torch.nn.Module):net.train()# 训练损失总和, 训练准确度总和, 样本数metric = d2l.Accumulator(3)for X, y in train_iter:y_hat = net(X)l = loss(y_hat, y)if isinstance(updater, torch.optim.Optimizer):updater.zero_grad()l.mean().backward()updater.step()else:l.sum().backward()updater(X.shape[0])metric.add(float(l.sum()), accuracy(y_hat, y), y.numel())# 返回训练损失和训练精度return metric[0] / metric[2], metric[1] / metric[2]def train_ch03(net, train_iter, test_iter, loss, num_epochs, updater):for epoch in range(num_epochs):train_loss, train_acc = train_epoch_ch03(net, train_iter, loss, updater)test_acc = evaluate_accuracy(net, test_iter)print(f'epoch {epoch + 1}, train_loss {train_loss:f}, train_acc {train_acc:f}, test_acc {test_acc:f}')assert train_loss < 0.5, train_lossassert train_acc <= 1 and train_acc > 0.7, train_accassert test_acc <= 1 and test_acc > 0.7, test_acclr = 0.1def updater(batch_size):return d2l.sgd([W, b], lr, batch_size)num_epochs = 10
train_ch03(net, train_iter, test_iter, cross_entropy, num_epochs, updater)