Confusion Matrix: ValueError: Unexpected keyword arguments: nan_strategy

[mfoglio]

After updating from torchmetrics 0.10.0 to 1.3.1 I get the following error when initializing a confusion matrix:

  File "/Users/user/Documents/ai-models/src/metrics/classifier.py", line 85, in __init__
    self.confusion_matrix_metrics = nn.ModuleDict({
  File "/Users/user/Documents/ai-models/src/metrics/classifier.py", line 86, in <dictcomp>
    dataset_split: nn.ModuleDict({
  File "/Users/user/Documents/ai-models/src/metrics/classifier.py", line 87, in <dictcomp>
    attribute.name: torchmetrics.ConfusionMatrix(
  File "/Users/user/Documents/ai-models/venv/lib/python3.8/site-packages/torchmetrics/classification/confusion_matrix.py", line 526, in __new__
    return MulticlassConfusionMatrix(num_classes, **kwargs)
  File "/Users/user/Documents/ai-models/venv/lib/python3.8/site-packages/torchmetrics/classification/confusion_matrix.py", line 265, in __init__
    super().__init__(**kwargs)
  File "/Users/user/Documents/ai-models/venv/lib/python3.8/site-packages/torchmetrics/metric.py", line 147, in __init__
    raise ValueError(f"Unexpected keyword arguments: {', '.join(kwargs_)}")
ValueError: Unexpected keyword arguments: `nan_strategy`

I initialized the confusion matrix in the following way:

torchmetrics.ConfusionMatrix(
                    task='multiclass',
                    num_classes=10,
                    ignore_index=-1,
                    normalize='true',
                    nan_strategy='ignore'
                )

Is the attribute nan_strategy deprecated or was I simply passing a useless argument before?

[Borda]

Hello, looking at the version you refer to and I don't think we have this argument ever implemented

torchmetrics/src/torchmetrics/classification/confusion_matrix.py

Lines 396 to 407 in 2a055f5

	def __new__(
	cls,
	num_classes: Optional[int] = None,
	normalize: Optional[str] = None,
	threshold: float = 0.5,
	multilabel: bool = False,
	task: Optional[Literal["binary", "multiclass", "multilabel"]] = None,
	num_labels: Optional[int] = None,
	ignore_index: Optional[int] = None,
	validate_args: bool = True,
	**kwargs: Any,
	) -> Metric: