如何使用Pythorch中的检查点模型文件来测试CIFAR10数据集？

Question

model = SqueezeNext()
model = model.to(device)

def load_checkpoint(model, optimizer, losslogger, filename='SqNxt_23_1x_Cifar.ckpt'):
# Note: Input model & optimizer should be pre-defined.  This routine only updates their states.
start_epoch = 0
if os.path.isfile(filename):
    print("=> loading checkpoint '{}'".format(filename))
    checkpoint = torch.load(filename)
    start_epoch = checkpoint['epoch']
    model.load_state_dict(checkpoint['state_dict'])
    optimizer.load_state_dict(checkpoint['optimizer'])
    losslogger = checkpoint['losslogger']
    print("=> loaded checkpoint '{}' (epoch {})"
              .format(filename, checkpoint['epoch']))
else:
    print("=> no checkpoint found at '{}'".format(filename))


return model, optimizer, start_epoch, losslogger

model, optimizer, start_epoch, losslogger = load_checkpoint(model, optimizer, losslogger)

---

TypeError: Traceback (most recent call last) in () 41 test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=80, num_workers=8, shuffle=False) 42 ---> 43 model = SqueezeNext() 44 model = model.to(device) 45 def load_checkpoint(model, optimizer, losslogger, filename='SqNxt_23_1x_Cifar.ckpt'): TypeError: init() missing 3 required positional arguments: 'width_x', 'blocks', and 'num_classes'

我认为我没有以正确的方式来实施这一点！！在

Answer 1

您的错误不是来自检查点函数。如果我们看看回溯：

> TypeError: Traceback (most recent call last)
> <ipython-input-51-94c8be648862> in <module>()
>      41 test_loader   = torch.utils.data.DataLoader(test_dataset, batch_size=80, num_workers=8, shuffle=False)
>      42 
>  -> 43 model = SqueezeNext()
>      44 model = model.to(device)
>      45 def load_checkpoint(model, optimizer, losslogger, filename='SqNxt_23_1x_Cifar.ckpt'): TypeError: __init__() missing 3
> required positional arguments: 'width_x', 'blocks', and 'num_classes'

我们被告知的底线正在突破第43行：

^{pr2}$

错误是：

> required positional arguments: 'width_x', 'blocks', and 'num_classes'

我假设您使用的是squezenext的this implementation，但是无论使用哪种实现，都没有传递初始化模型所需的所有参数。您需要将代码更改为类似于：

^{4}$

如果不使用该实现，则需要找到SqueezeNext模型的源代码，并查看__init__函数需要哪些参数。你可以试试这个：

import inspect

inspect.signature(SqueezeNext.__init__)

它应该给你签名。在