Pytorch GRU error RuntimeError : size mismatch, m1: [1600 x 3], m2: [50 x 20]
Pytorch GRU error RuntimeError : size mismatch, m1: [1600 x 3], m2: [50 x 20]
目前，我正在尝试为 LSTM 和 GRU 制作训练模型。 LSTM 运行良好，但是一旦我切换到 GRU 训练，就会弹出错误，例如大小不匹配错误。
这是我的代码
path = "new_z_axis"
device = "cuda:0"
in_size = 3
h_size = 50
n_layers = 3
fc = 20
out = 1
batch_size = 16
seq = 100
epoch = 100
learning_rate = 1e-3
ratio = 0.8
checkpoint = os.path.join("checkpoints","model_"+path+"_"+str(in_size)+".pth")
class GRUNet(nn.Module):
def __init__(self,in_size,h_size,n_layers,fc_out,out_size,dropout=0.5):
super(GRUNet, self).__init__()
self.gru = nn.GRU(input_size=in_size,hidden_size=h_size,num_layers=n_layers,dropout=dropout,bias=False)
self.fc = nn.Linear(in_features=h_size,out_features=fc_out,bias=False)
self.relu = nn.ReLU(inplace=True)
self.out = nn.Linear(in_features=fc_out,out_features=out_size,bias=False)
self.tanh = nn.Tanh()
def forward(self, x, hidden):
out, hidden = self.gru(x, hidden)
x = self.fc(x)
x = self.relu(x)
x = self.out(x)
x = self.tanh(x)
return x, hidden
class MyLstm(nn.Module):
def __init__(self,in_size,h_size,n_layers,fc_out,out_size,dropout=0.5):
super(MyLstm, self).__init__()
self.lstm = nn.LSTM(input_size=in_size,hidden_size=h_size,num_layers=n_layers,dropout=dropout,bias=False)
self.fc = nn.Linear(in_features=h_size,out_features=fc_out,bias=False)
self.relu = nn.ReLU(inplace=True)
self.out = nn.Linear(in_features=fc_out,out_features=out_size,bias=False)
self.tanh = nn.Tanh()
def forward(self,x,hidden):
x, hidden = self.lstm(x,hidden)
# x = x[-1:]
x = self.fc(x)
x = self.relu(x)
x = self.out(x)
x = self.tanh(x)
return x, hidden
def train(model,train_list,val_list,path,seq,epoch,batch_size,criterion,optimizer,model_type):
for e in range(epoch):
train_data = load_data(train_list,batch_size)
a_loss = 0
a_size = 0
model.train()
for x,y in train_data:
x,y = x.to(device),y.to(device)
bs = x.size()[1]
# hidden = (hidden[0].detach(),hidden[1].detach())
# print(x.size(),hidden[0].size())
if model_type == "GRU":
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = h1
hidden = hidden.data
else:
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
h2 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = (h1,h2)
hidden = tuple([e.data for e in hidden])
model.zero_grad()
print (len(hidden))
pred,hidden = model(x,hidden)
loss = criterion(pred,y)
loss.backward()
nn.utils.clip_grad_norm_(model.parameters(),5)
optimizer.step()
a_loss += loss.detach()
a_size += bs
# print(e,a_loss/a_size*1e+6)
model.eval()
with torch.no_grad():
val_data = load_data(val_list,batch_size)
b_loss = 0
b_size = 0
for x,y in val_data:
x,y = x.to(device),y.to(device)
bs = x.size()[1]
if model_type == "GRU":
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = h1
hidden = hidden.data
else:
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
h2 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = (h1,h2)
hidden = tuple([e.data for e in hidden])
pred,hidden = model(x,hidden)
loss = criterion(pred,y)
b_loss += loss.detach()
b_size += bs
print("epoch: {} - train_loss: {} - val_loss: {}".format(e+1,float(a_loss.item()/a_size*1e+6),b_loss.item()/b_size*1e+6))
train(modelGRU,train_list,val_list,path,seq,epoch,batch_size,criterionGRU,optimizerGRU,model_type="GRU")
这是我得到的错误
-------------------------------------------------- ------------------------- RuntimeError Traceback (most recent call last) in
---- > 1 train ( modelGRU , train_list , val_list , path , seq , epoch , batch_size , criterionGRU , optimizerGRU , model_type = "GRU"
)
in train (model, train_list, val_list,
path, seq, epoch, batch_size, criterion, optimizer, model_type)
61 model . zero_grad ( )
62 print ( len ( hidden ) )
---> 63 pred , hidden = model ( x , hidden )
64 loss = criterion ( pred , y )
65 loss .backward ( )
~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in call
(self, *input, **kwargs)
539 result = self . _slow_forward ( * input , ** kwargs )
540 else :
--> 541 result = self . forward ( * input , ** kwargs )
542 for hook in self . _forward_hooks . values ( ) :
543 hook_result = hook ( self , input , result )
in forward (self, x, hidden)
11 def forward ( self , x , hidden ) :
12 out , hidden = self . gru ( x , hidden )
---> 13 x = self . fc ( x )
14 x = self . relu ( x )
15 x =self . out ( x )
~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in call
(self, *input, **kwargs)
539 result = self . _slow_forward ( * input , ** kwargs )
540 else :
--> 541 result = self . forward ( * input , ** kwargs )
542 for hook in self . _forward_hooks . values ( ) :
543 hook_result = hook ( self , input , result )
~ \ Anaconda3 \ lib \ site-packages \ torch \ nn \ modules \ linear.py
in forward (self, input)
85
86 def forward ( self , input ) :
---> 87 return F . Linear ( the Input , Self . Weight , Self . negative bias ) 88 89 def extra_repr ( Self ) :
~\Anaconda3\lib\site-packages\torch\nn\functional.py in linear (input,
weight, bias) 1370 ret = torch . addmm ( bias , input ,
weight . t ( ) ) 1371 else :
-> 1372 output = input . matmul ( weight . t ( ) ) 1373 if bias is not None : 1374 output += bias
RuntimeError : size mismatch, m1: [1600 x 3], m2: [50 x 20] at
C:/w/1/s/tmp_conda_3.7_104508/conda/conda-bld/pytorch_1572950778684/work/aten/src\THC/
generic/THCTensorMathBlas.cu:290
有什么建议吗？
谢谢
这可能与您没有将 nn.GRU 的输出传递给 GRUNet 的 forward 函数中的第一个线性层有关：
def forward(self, x, hidden):
out, hidden = self.gru(x, hidden)
x = self.fc(out)
目前，我正在尝试为 LSTM 和 GRU 制作训练模型。 LSTM 运行良好，但是一旦我切换到 GRU 训练，就会弹出错误，例如大小不匹配错误。
这是我的代码
path = "new_z_axis"
device = "cuda:0"
in_size = 3
h_size = 50
n_layers = 3
fc = 20
out = 1
batch_size = 16
seq = 100
epoch = 100
learning_rate = 1e-3
ratio = 0.8
checkpoint = os.path.join("checkpoints","model_"+path+"_"+str(in_size)+".pth")
class GRUNet(nn.Module):
def __init__(self,in_size,h_size,n_layers,fc_out,out_size,dropout=0.5):
super(GRUNet, self).__init__()
self.gru = nn.GRU(input_size=in_size,hidden_size=h_size,num_layers=n_layers,dropout=dropout,bias=False)
self.fc = nn.Linear(in_features=h_size,out_features=fc_out,bias=False)
self.relu = nn.ReLU(inplace=True)
self.out = nn.Linear(in_features=fc_out,out_features=out_size,bias=False)
self.tanh = nn.Tanh()
def forward(self, x, hidden):
out, hidden = self.gru(x, hidden)
x = self.fc(x)
x = self.relu(x)
x = self.out(x)
x = self.tanh(x)
return x, hidden
class MyLstm(nn.Module):
def __init__(self,in_size,h_size,n_layers,fc_out,out_size,dropout=0.5):
super(MyLstm, self).__init__()
self.lstm = nn.LSTM(input_size=in_size,hidden_size=h_size,num_layers=n_layers,dropout=dropout,bias=False)
self.fc = nn.Linear(in_features=h_size,out_features=fc_out,bias=False)
self.relu = nn.ReLU(inplace=True)
self.out = nn.Linear(in_features=fc_out,out_features=out_size,bias=False)
self.tanh = nn.Tanh()
def forward(self,x,hidden):
x, hidden = self.lstm(x,hidden)
# x = x[-1:]
x = self.fc(x)
x = self.relu(x)
x = self.out(x)
x = self.tanh(x)
return x, hidden
def train(model,train_list,val_list,path,seq,epoch,batch_size,criterion,optimizer,model_type):
for e in range(epoch):
train_data = load_data(train_list,batch_size)
a_loss = 0
a_size = 0
model.train()
for x,y in train_data:
x,y = x.to(device),y.to(device)
bs = x.size()[1]
# hidden = (hidden[0].detach(),hidden[1].detach())
# print(x.size(),hidden[0].size())
if model_type == "GRU":
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = h1
hidden = hidden.data
else:
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
h2 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = (h1,h2)
hidden = tuple([e.data for e in hidden])
model.zero_grad()
print (len(hidden))
pred,hidden = model(x,hidden)
loss = criterion(pred,y)
loss.backward()
nn.utils.clip_grad_norm_(model.parameters(),5)
optimizer.step()
a_loss += loss.detach()
a_size += bs
# print(e,a_loss/a_size*1e+6)
model.eval()
with torch.no_grad():
val_data = load_data(val_list,batch_size)
b_loss = 0
b_size = 0
for x,y in val_data:
x,y = x.to(device),y.to(device)
bs = x.size()[1]
if model_type == "GRU":
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = h1
hidden = hidden.data
else:
h1 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
h2 = torch.zeros((n_layers,bs,h_size)).to("cuda:0")
hidden = (h1,h2)
hidden = tuple([e.data for e in hidden])
pred,hidden = model(x,hidden)
loss = criterion(pred,y)
b_loss += loss.detach()
b_size += bs
print("epoch: {} - train_loss: {} - val_loss: {}".format(e+1,float(a_loss.item()/a_size*1e+6),b_loss.item()/b_size*1e+6))
train(modelGRU,train_list,val_list,path,seq,epoch,batch_size,criterionGRU,optimizerGRU,model_type="GRU")
这是我得到的错误
-------------------------------------------------- ------------------------- RuntimeError Traceback (most recent call last) in ---- > 1 train ( modelGRU , train_list , val_list , path , seq , epoch , batch_size , criterionGRU , optimizerGRU , model_type = "GRU" )
in train (model, train_list, val_list, path, seq, epoch, batch_size, criterion, optimizer, model_type) 61 model . zero_grad ( ) 62 print ( len ( hidden ) ) ---> 63 pred , hidden = model ( x , hidden ) 64 loss = criterion ( pred , y ) 65 loss .backward ( )
~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in call (self, *input, **kwargs) 539 result = self . _slow_forward ( * input , ** kwargs ) 540 else : --> 541 result = self . forward ( * input , ** kwargs ) 542 for hook in self . _forward_hooks . values ( ) : 543 hook_result = hook ( self , input , result )
in forward (self, x, hidden) 11 def forward ( self , x , hidden ) : 12 out , hidden = self . gru ( x , hidden ) ---> 13 x = self . fc ( x ) 14 x = self . relu ( x ) 15 x =self . out ( x )
~\Anaconda3\lib\site-packages\torch\nn\modules\module.py in call (self, *input, **kwargs) 539 result = self . _slow_forward ( * input , ** kwargs ) 540 else : --> 541 result = self . forward ( * input , ** kwargs ) 542 for hook in self . _forward_hooks . values ( ) : 543 hook_result = hook ( self , input , result )
~ \ Anaconda3 \ lib \ site-packages \ torch \ nn \ modules \ linear.py in forward (self, input) 85
86 def forward ( self , input ) : ---> 87 return F . Linear ( the Input , Self . Weight , Self . negative bias ) 88 89 def extra_repr ( Self ) :~\Anaconda3\lib\site-packages\torch\nn\functional.py in linear (input, weight, bias) 1370 ret = torch . addmm ( bias , input , weight . t ( ) ) 1371 else : -> 1372 output = input . matmul ( weight . t ( ) ) 1373 if bias is not None : 1374 output += bias
RuntimeError : size mismatch, m1: [1600 x 3], m2: [50 x 20] at C:/w/1/s/tmp_conda_3.7_104508/conda/conda-bld/pytorch_1572950778684/work/aten/src\THC/ generic/THCTensorMathBlas.cu:290
有什么建议吗？ 谢谢
这可能与您没有将 nn.GRU 的输出传递给 GRUNet 的 forward 函数中的第一个线性层有关：
def forward(self, x, hidden):
out, hidden = self.gru(x, hidden)
x = self.fc(out)