How to solve AttributeError: 'Tensor' object has no attribute 'zero_grad' in pytorch

Question

Still working through video tutorial https://www.youtube.com/watch?v=weQ5pShEVic&list=PLbMqOoYQ3Mxw1Sl5iAAV4SJmvnAGAhFvK&index=2 about pytorch but hit another error.

lossFunc = torch.nn.MSELoss()
for i in range(epoch):
        output = net(x)
        loss = lossFunc(output, y)
        loss.zero_grad()
        loss.backward()
 
        
        for f in net.parameters():
            f.data.sub_(learning_rate = f.grad.data)
print(output, loss)

Created the network, loss function and wanted to iterate before backpropogattion

but get this error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
/var/folders/v_/yq26pm194xj5ckqy8p_njwc00000gn/T/ipykernel_9995/2476130544.py in <module>
      3         output = net(x)
      4         loss = lossFunc(output, y)
----> 5         loss.zero_grad()
      6         loss.backward()
      7 

AttributeError: 'Tensor' object has no attribute 'zero_grad'

What gives?

Asked By: elksie5000

||

Source

Answer 1

You should use zero grad for your optimizer.

optimizer = torch.optim.Adam(net.parameters(), lr=0.001)
lossFunc = torch.nn.MSELoss()
for i in range(epoch):
    optimizer.zero_grad()
    output = net(x)
    loss = lossFunc(output, y)
    loss.backward()
    optimizer.step()

Answered By: core_not_dumped

How to solve AttributeError: 'Tensor' object has no attribute 'zero_grad' in pytorch

Question:

Answers: