RuntimeError: input.size(-1)必须等于input_size。应为28，得到0

Question

下面是我使用Pysft编写的代码

class Arguments:
        def __init__(self):
            # self.cuda = False
            self.no_cuda = True
            self.seed = 1
            self.batch_size = 50
            self.test_batch_size = 1000
            self.epochs = 10
            self.lr = 0.01
            self.momentum = 0.5
            self.log_interval = 10

hook = sy.TorchHook(torch)  
bob = sy.VirtualWorker(hook, id="bob")  
alice = sy.VirtualWorker(hook, id="alice") 

这是我的LSTM模型，只使用pytorch就可以成功运行，但不能使用pysyft运行

 class Model(torch.nn.Module):
        def __init__(self):
            super(Model, self).__init__()
            self.rnn = torch.nn.RNN(input_size=28,
            hidden_size=16,
            num_layers=2,
            batch_first=True,
            bidirectional=True)
            self.fc = torch.nn.Linear(32, 10)
        def forward(self, x):
            print(np.shape(x))
            x = x.squeeze()
            x, _ = self.rnn(x)
            x = self.fc(x[:, -1, :])
            return x.view(-1, 10)

def train(args, model, device, federated_train_loader, optimizer, epoch):
    model.train()
    for batch_idx, (data, target) in enumerate(federated_train_loader): 
        model.send(data.location) # <-- NEW: send the model to the right location
        data, target = data.to(device), target.to(device)
        # data, target = data.cuda(), target.cuda()
        optimizer.zero_grad() 
        output = model(data.to(device))
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        model.get() # <-- NEW: get the model back
        if batch_idx % args.log_interval == 0:
            loss = loss.get() # <-- NEW: get the loss back
            losses.append(loss.item())
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * args.batch_size, len(federated_train_loader) * args.batch_size,
                100. * batch_idx / len(federated_train_loader), loss.item()))

当我使用Pysyft运行我的LSTM模型时，如果我使用没有Pysyft的模型，会出现一个mistakes.But，它可以成功运行。我不知道如何解决它？

import torch
import matplotlib.pyplot as plt
from torchvision import datasets, transforms
import torch.nn.functional as F
import time
import numpy as np
import syft as sy

class Arguments:
    def __init__(self):
        self.cuda = False
        self.no_cuda = True
        self.seed = 1
        self.batch_size = 50
        self.test_batch_size = 1000
        self.epochs = 10
        self.lr = 0.01
        self.momentum = 0.5
        self.log_interval = 10

hook = sy.TorchHook(torch)  # <-- NEW: hook PyTorch ie add extra functionalities to support Federated Learning
bob = sy.VirtualWorker(hook, id="bob")  # <-- NEW: define remote worker bob
alice = sy.VirtualWorker(hook, id="alice")  # <-- NEW: and alice

class Model(torch.nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.rnn = torch.nn.RNN(input_size=28,
        hidden_size=16,
        num_layers=2,
        batch_first=True,
        bidirectional=True)
        self.fc = torch.nn.Linear(32, 10)
    def forward(self, x):
        print(np.shape(x))
        x = x.squeeze()
        x, _ = self.rnn(x)
        x = self.fc(x[:, -1, :])
        return x.view(-1, 10)

def train(args, model, device, federated_train_loader, optimizer, epoch):
    model.train()
    for batch_idx, (data, target) in enumerate(federated_train_loader): # <-- now it is a distributed dataset
        model.send(data.location) # <-- NEW: send the model to the right location
        data, target = data.to(device), target.to(device)
        optimizer.zero_grad() 
        output = model(data.to(device))
        loss = F.nll_loss(output, target)
        loss.backward()
        optimizer.step()
        model.get() # <-- NEW: get the model back
        if batch_idx % args.log_interval == 0:
            loss = loss.get() # <-- NEW: get the loss back
            losses.append(loss.item())
            print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
                epoch, batch_idx * args.batch_size, len(federated_train_loader) * args.batch_size,
                100. * batch_idx / len(federated_train_loader), loss.item()))

if __name__ == '__main__':

    args = Arguments()

    use_cuda = not args.no_cuda and torch.cuda.is_available()

    torch.manual_seed(args.seed)

    device = torch.device("cuda" if use_cuda else "cpu")

    kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}

    losses = []
    
    federated_train_loader = sy.FederatedDataLoader(
        datasets.MNIST('../data', train=True, download=True,
                       transform=transforms.Compose([
                           transforms.ToTensor(),
                           transforms.Normalize((0.1307,), (0.3081,))
                       ]))
        .federate((bob, alice)),  # <-- NEW: we distribute the dataset across all the workers, it's now a FederatedDataset
        batch_size=args.batch_size, shuffle=True, **kwargs)
    test_loader = torch.utils.data.DataLoader(
        datasets.MNIST('../data', train=False, transform=transforms.Compose([
            transforms.ToTensor(),
            transforms.Normalize((0.1307,), (0.3081,))
        ])),
        batch_size=args.test_batch_size, shuffle=True, **kwargs)
    model = Model().to(device)
    optimizer = torch.optim.Adam(model.parameters(), lr=args.lr)


    t = time.time()
    for epoch in range(1, args.epochs + 1):
        train(args, model, device, federated_train_loader, optimizer, epoch)

        test(args, model, device, test_loader)
    plt.plot(range(0,160),losses,marker='o')
    plt.xlabel("iterator")
    plt.ylabel("loss")
    plt.show()
    total_time = time.time() - t
    print(total_time)

下面是完整的代码

Answer 1

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torchvision import datasets, transforms
import syft as sy

hook = sy.TorchHook(torch)
bob =  sy.VirtualWorker(hook, id="bob")
alice =  sy.VirtualWorker(hook, id="alice")

class Arguments():
    def __init__(self):
        self.batch_size = 64
        self.test_batch_size = 1000
        self.epochs = 10
        self.lr = 0.01
        self.momentum = 0.5
        self.no_cuda = False
        self.seed = 1
        self.log_interval = 10
        self.save_model = False

args = Arguments()


use_cuda = not args.no_cuda and torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
kwargs = {'num_workers': 1, 'pin_memory': True} if use_cuda else {}



federated_train_loader = sy.FederatedDataLoader( # <-- this is now a FederatedDataLoader 
    datasets.MNIST('../data', train=True, download=True,
                   transform=transforms.Compose([
                       transforms.ToTensor(),
                       transforms.Normalize((0.1307,), (0.3081,))
                   ]))
    .federate((bob, alice)), 
    batch_size=args.batch_size, shuffle=True, **kwargs)

test_loader = torch.utils.data.DataLoader(
    datasets.MNIST('../data', train=False, transform=transforms.Compose([
                       transforms.ToTensor(),
                       transforms.Normalize((0.1307,), (0.3081,))
                   ])),
    batch_size=args.test_batch_size, shuffle=True, **kwargs)


class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 20, 5, 1)
        self.conv2 = nn.Conv2d(20, 50, 5, 1)
        self.fc1 = nn.Linear(4*4*50, 500)
        self.fc2 = nn.Linear(500, 10)

    def forward(self, x):
        x = F.relu(self.conv1(x))
        x = F.max_pool2d(x, 2, 2)
        x = F.relu(self.conv2(x))
        x = F.max_pool2d(x, 2, 2)
        x = x.view(-1, 4*4*50)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return F.log_softmax(x, dim=1)



model = Net()
model = model.to(device)  #pushing the model into available device.
optimizer = optim.SGD(model.parameters(), lr=0.01)


for epoch in range(1, args.epochs + 1):
    # Train the model
    model.train()
    for batch_idx, (data, target) in enumerate(federated_train_loader): # iterate through each worker's dataset
        
        model.send(data.location) #send the model to the right location ; data.location returns the worker name in which the data is present
        
        data, target = data.to(device), target.to(device) # pushing both the data and target labels onto the available device.
        
        optimizer.zero_grad() # 1) erase previous gradients (if they exist)
        output = model(data)  # 2) make a prediction
        loss = F.nll_loss(output, target)  # 3) calculate how much we missed
        loss.backward()  # 4) figure out which weights caused us to miss
        optimizer.step()  # 5) change those weights
        model.get()  # get the model back (with gradients)
        
        if batch_idx % args.log_interval == 0:
            loss = loss.get() #get the loss back
            print('Epoch: {} [Training: {:.0f}%]\tLoss: {:.6f}'.format(epoch, 100. * batch_idx / len(federated_train_loader), loss.item()))
    # Test the model
    model.eval()  
    test_loss = 0
    correct = 0
    with torch.no_grad():
      for data, target in test_loader:
        data, target = data.to(device), target.to(device) 
        output = model(data) # Getting a prediction

        test_loss += F.nll_loss(output, target, reduction='sum').item() #updating test loss
        pred = output.argmax(1, keepdim=True) # get the index of the max log-probability 

        correct += pred.eq(target.view_as(pred)).sum().item() #correct pred in the current test set.

    test_loss /= len(test_loader.dataset) 

    print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.0f}%)\n'.format(test_loss, correct, len(test_loader.dataset), 100. * correct / len(test_loader.dataset)))

    torch.save(model.state_dict(), "mnist_cnn.pt")

我已经在torch 1.x和pysyft 0.2.5中测试了上面的代码，并且它工作正常。(但使用cnn模型)...只需在此处更改数据加载器和模型。