`log_dict` is missing `prefix` parameter that is mentioned in docs about torchmetrics HOT 11 CLOSED

Hello @SkafteNicki I like the idea of your implementation and I think it could add as well an optional argument to the clone method.

I am thinking something like ...

...
def clone(self, prefix = None):
    cloned = super().clone()
    if prefix is not None:
        cloned.prefix = prefix
    return cloned
...

Since it would make it a lot easier to make clones of large collections (train and test metrics for instance ...)
let me know when you feel would be a good time to work on it (although i feel like you got it covered, thanks a lot)

from torchmetrics.

@jspaezp please checkout the PR I have opened #70 to see if anything is missing :]

from torchmetrics.

Fix the documentation. Does MetricCollection have a mechanism to set a prefix? If it doesn't, then the argument should be added there

from torchmetrics.

@carmocca are you in favor of adding the argument to log_dict or fix the documentation?

from torchmetrics.

@carmocca I agree this would be a nice addition and the implementation seems easy ...

-def __init__(self, metrics: Union[List[Metric], Tuple[Metric], Dict[str, Metric]]):
+def __init__(self, metrics: Union[List[Metric], Tuple[Metric], Dict[str, Metric]], prefix: Union[Literal[None],str]=None):
         super().__init__()
         if isinstance(metrics, dict):
+            prefix = prefix if prefix is not None else ""
             # Check all values are metrics
             for name, metric in metrics.items():
                 if not isinstance(metric, Metric):
                     raise ValueError(
                         f"Value {metric} belonging to key {name}"
                         " is not an instance of `pl.metrics.Metric`"
                     )
-                self[name] = metric
+                self[prefix + name] = metric

possibly add a warning as well if a list (and not a dict) is passed so you know that it will get ignored.
let me know if you want the PR. (I could also update the docs)

Kindest wishes,
Sebastian

from torchmetrics.

@jspaezp your solution looks very similar to the one I had in mind:
Lightning-AI/pytorch-lightning@master...SkafteNicki:fix_prefix_arg
Feel free to submit a PR but please wait a couple of days for an update that will come to the metrics package (cannot disclose exactly what right now) :]

from torchmetrics.

I'm just updating to the latest version of pytorch lightning and I ran into this issue also. Do we have an ETA for when it will be released?

from torchmetrics.

Hi! thanks for your contribution!, great first issue!

from torchmetrics.

@SkafteNicki looks great! Amazing work!

@uwaisiqbal (just mentioning you so you know it is getting merged)

from torchmetrics.

Awesome thanks for the update and for the amazing work!

from torchmetrics.

When is the update with the prefix arg going to be released? Also, the documentation for the MetricCollections is out of date where it mentions the prefix arg can be used when cloning the MetricCollection https://torchmetrics.readthedocs.io/en/latest/pages/overview.html#metriccollection

from torchmetrics.