Maybe a new patch release?

[ioangatop]

Motivation

A new release will help since there were some refactoring in the segmentation metrics.

Thanks a lot 🙏

[Borda]

sure I can do one today :)

[SkafteNicki]

Closing as v1.4.2 was released yesterday :]

[ioangatop]

hmmm that is kinda weird as the new release for example does not have the following argument from main - how is this not included 🤔 ?

torchmetrics/src/torchmetrics/segmentation/generalized_dice.py

Line 122 in fd11f33

input_format: Literal["one-hot", "index"] = "one-hot",

[ioangatop]

@Borda do you maybe know why this happen? thank you 🙏

[Borda]

@Borda do you maybe know why this happen? thank you 🙏

what do you mean?

[ioangatop]

For example, lets get the GeneralizedDiceScore; on the release it seems that it has an older code from the main at that time. For example:

# on main; 3 weeks ago
class GeneralizedDiceScore(Metric):
    ...
    def __init__(
        self,
        num_classes: int,
        include_background: bool = True,
        per_class: bool = False,
        weight_type: Literal["square", "simple", "linear"] = "square",
        input_format: Literal["one-hot", "index"] = "one-hot",                # <-- HERE
        **kwargs: Any,
    ) -> None:

# on release 1.4.2; 2 weeks ago
class GeneralizedDiceScore(Metric):
    ...
    def __init__(
        self,
        num_classes: int,
        include_background: bool = True,
        per_class: bool = False,
        weight_type: Literal["square", "simple", "linear"] = "square",
                                                                             # <-- HERE missing arg
        **kwargs: Any,
    ) -> None:

[Borda]

Thanks for your concern. We have released patch release recently which by nature does not expect API changes, correct?

[ioangatop]

@Borda make sense yeah... I guess I wanted to have a minor release? and is this will be possible? :D