-
-
Notifications
You must be signed in to change notification settings - Fork 623
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add probability to Accuracy #1354
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@vcarpani thanks for the PR and sorry for the delay. I didn't see that it is not draft anymore.
I left few comments to improve it and we have to discuss a bit on probably misleading terms...
ignite/metrics/accuracy.py
Outdated
def __init__( | ||
self, | ||
output_transform: Callable = lambda x: x, | ||
is_multilabel: bool = False, | ||
device: Union[str, torch.device] = torch.device("cpu"), | ||
mode: Mode = Mode.LABELS, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not sure about correctness of name LABELS
for multi-class case where we require to pass probas or logits and then take argmax.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
What about UNCHANGED
or RAW_INPUT
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, something like RAW_INPUT
could work. I'm thinking if we could not generalize this new options to all possible inputs and metric type: binary, multiclass, multilabel...
@vfdev-5 the last commit should solve your comments |
I'll try to implement that behaviour somewhere next week ;) |
@vfdev-5, just wanted to update you on this one, I've been quite busy recently, but I started working on it |
@vcarpani thanks for the head-up :) |
I would like to add that "logits" in the context of multiclass problems usually mean the input to the softmax (not the sigmoid) function. I am not sure if there are any use cases for the current |
Fixes #1089
Description:
Accuracy
according to the suggestions from the issue.Still WIP, to be done:
Check list: