only auto set ignored_index in multiclass

Signed-off-by: Zhiyuan Chen <[email protected]>

danling/metrics/functional.py

@@ -21,6 +21,7 @@
 from collections.abc import Sequence
 
 import torch
+from chanfig.utils import NULL, Null
 from lazy_imports import try_import
 from torch import Tensor
 
@@ -40,12 +41,14 @@ def auroc(
     num_labels: int | None = None,
     num_classes: int | None = None,
     task_weights: Tensor | None = None,
-    ignored_index: int = -100,
+    ignored_index: int | None | NULL = Null,
     **kwargs,
 ):
     te.check()
     if num_classes and num_labels:
         raise ValueError("Only one of num_classes or num_labels can be specified, but not both")
+    if ignored_index is Null:
+        ignored_index = -100 if num_classes else None
     input, target = preprocess(input, target, ignored_index=ignored_index)
     if num_labels is None and num_classes is None:
         return tef.binary_auroc(input=input, target=target, weight=weight, **kwargs)
@@ -66,12 +69,14 @@ def auprc(
     num_labels: int | None = None,
     num_classes: int | None = None,
     task_weights: Tensor | None = None,
-    ignored_index: int = -100,
+    ignored_index: int | None | NULL = Null,
     **kwargs,
 ):
     te.check()
     if num_classes and num_labels:
         raise ValueError("Only one of num_classes or num_labels can be specified, but not both")
+    if ignored_index is Null:
+        ignored_index = -100 if num_classes else None
     input, target = preprocess(input, target, ignored_index=ignored_index)
     if num_labels is None and num_classes is None:
         return tef.binary_auprc(input=input, target=target, **kwargs)
@@ -92,12 +97,14 @@ def accuracy(
     average: str | None = "micro",
     num_labels: int | None = None,
     num_classes: int | None = None,
-    ignored_index: int = -100,
+    ignored_index: int | None | NULL = Null,
     **kwargs,
 ):
     te.check()
     if num_classes and num_labels:
         raise ValueError("Only one of num_classes or num_labels can be specified, but not both")
+    if ignored_index is Null:
+        ignored_index = -100 if num_classes else None
     input, target = preprocess(input, target, ignored_index=ignored_index)
     if num_labels is None and num_classes is None:
         return tef.binary_accuracy(input=input, target=target, threshold=threshold, **kwargs)
@@ -114,7 +121,7 @@ def mcc(
     threshold: float = 0.5,
     num_labels: int | None = None,
     num_classes: int | None = None,
-    ignored_index: int = -100,
+    ignored_index: int | None | NULL = Null,
 ):
     tm.check()
     if num_classes and num_labels:
@@ -124,6 +131,8 @@ def mcc(
         task = "multiclass"
     if num_labels:
         task = "multilabel"
+    if ignored_index is Null:
+        ignored_index = -100 if num_classes else None
     input, target = preprocess(input, target, ignored_index=ignored_index)
     try:
         return tmf.matthews_corrcoef(