[Resolved] Inplace operation issue

1

Hi, I searched that I’d need to use clone() to avoid inplace operation issue.
So, I cloned pred_R_imgs but I still have error.
When I commented them out, error would disappear.

Why does clone() work in this case?

pred_R_imgs=Net(O_img_tc)
pred_R_imgs[pred_R_imgs.clone()==0.0]=0.0001
pred_R_imgs[pred_R_imgs.clone()==-0.0]=0.0001
2

I incorrectly used clone()
This made error disappear

pred_R_imgs=Net(O_img_tc)
pred_R_imgs.clone()[pred_R_imgs==0.0]=0.0001
pred_R_imgs.clone()[pred_R_imgs==-0.0]=0.0001
3

Yes, but you haven’t changed pred_R_imgs but only it’s clones.
What you likely want is something like

pred_R_imgs = torch.where(pred_R_imgs.abs()>1e-4 , pred_R_imgs, torch.tensor(1e-4))

Best regards

Thomas

1 Like
4

That’s helpful advice using torch.where() to change value.
Thanks a lot, Thomas.