我有这个Pytorch Ensemble模型，但是可以找出问题所在。谢谢你的帮助。
我认为问题出在第一个Fc1上，我是从Packt Deep Learning中的PyTorch的书中获得的，我包括了fit（）的定义，因此您可以看到完整的循环。我之前没有包含它，因为它告诉我看起来很多代码而且没有任何文字...
def fit(epoch,model,data_loader,phase='training',volatile=False):
if phase == 'training':
model.train()
if phase == 'validation':
model.eval()
volatile=True
running_loss = 0.0
running_correct = 0
for batch_idx , (data1,data2,data3,target) in enumerate(data_loader):
if is_cuda:
data1,data2,data3,target = data1.cuda(),data2.cuda(),data3.cuda(),target.cuda()
data1,data2,data3,target = Variable(data1,volatile),Variable(data2,volatile),Variable(data3,volatile),Variable(target)
if phase == 'training':
optimizer.zero_grad()
output = model(data1,data2,data3)
loss = F.cross_entropy(output,target)
running_loss += F.cross_entropy(output,target,size_average=False).data[0]
preds = output.data.max(dim=1,keepdim=True)[1]
running_correct += preds.eq(target.data.view_as(preds)).cpu().sum()
if phase == 'training':
loss.backward()
optimizer.step()
loss = running_loss/len(data_loader.dataset)
accuracy = 100. * running_correct/len(data_loader.dataset)
print(f'{phase} loss is {loss:{5}.{2}} and {phase} accuracy is {running_correct}/{len(data_loader.dataset)}{accuracy:{10}.{4}}')
return loss,accuracy
class LayerActivations():
features=[]
def __init__(self,model):
self.features = []
self.hook = model.register_forward_hook(self.hook_fn)
def hook_fn(self,module,input,output):
#out = F.avg_pool2d(output, kernel_size=8)
self.features.extend(output.view(output.size(0),-1).cpu().data)
def remove(self):
self.hook.remove()
class EnsembleModel(nn.Module):
def __init__(self, out_size, training=True):
super().__init__()
self.fc1 = nn.Linear(8192, 512)
self.fc2 = nn.Linear(131072, 512)
self.fc3 = nn.Linear(82944, 512)
self.fc4 = nn.Linear(512, out_size)
def forward(self, inp1, inp2, inp3):
out1 = self.fc1(F.dropout(inp1, training=self.training))
out2 = self.fc2(F.dropout(inp2, training=self.training))
out3 = self.fc3(F.dropout(inp3, training=self.training))
out = out1 + out2 + out3
out = self.fc4(F.dropout(out, training=self.training))
return out
em = EnsembleModel(2)
if is_cuda:
em = em.cuda()
train_losses, train_accuracy = [], []
val_losses, val_accuracy = [], []
for epoch in range(1, 10):
epoch_loss, epoch_accuracy = fit(epoch, em, trn_feat_loader, phase="training")
val_epoch_loss, val_epoch_accuracy = fit(
epoch, em, val_feat_loader, phase="validation"
)
train_losses.append(epoch_loss)
train_accuracy.append(epoch_accuracy)
val_losses.append(val_epoch_loss)
val_accuracy.append(val_epoch_accuracy)
错误：
RuntimeError Traceback (most recent call last)
def extra_repr(self):
RuntimeError: size mismatch, m1: [64 x 512], m2: [8192 x 512] at /pytorch/aten/src/THC/generic/THCTensorMathBlas.cu:290