Conversation
|
|
||
|
|
||
|
|
||
| class WeightedBCE(nn.Module): |
There was a problem hiding this comment.
I would think that weighted bce means that I can throw weights into it. It's rather BalancedWeightedBCE
|
|
||
| losses = [WeightedBCE(), Dice()] | ||
| weights = [.5, .5] | ||
| comboloss = Weighted(losses, weights) No newline at end of file |
There was a problem hiding this comment.
Do you really need it here? There is not a problem to make it in your notebook:
ComboLoss = Weighted([WeightedBCE(), Dice()], [.5, .5])
Nevertheless, comboloss here is inappropriate naming.
There was a problem hiding this comment.
I think, ye, but I thought I should present the full loss used in my models
There was a problem hiding this comment.
Do you really need it here? There is not a problem to make it in your notebook:
ComboLoss = Weighted([WeightedBCE(), Dice()], [.5, .5])
Nevertheless,
combolosshere is inappropriate naming.
Why is comboloss inappropriate naming? The bce loss is combined with dice loss --> comboloss
There was a problem hiding this comment.
It's unclear combination of what
|
|
||
|
|
||
|
|
||
| class BalancedWeightedBCE(nn.Module): |
There was a problem hiding this comment.
CrossEntropyLoss from batchflow.models.torch.losses.core.py already can do all of that. Maybe, make inherit BCE / make it a special case of it somehow?
3 participants
No description provided.