make is_differentiable as attribute (#551)

* setter
* typing
* Apply suggestions from code review

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: thomas chaton <[email protected]>

CHANGELOG.md

@@ -33,6 +33,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `half`, `double`, `float` will no longer change the dtype of the metric states. Use `metric.set_dtype` instead ([#493](https://github.com/PyTorchLightning/metrics/pull/493))
 
 
+- Changed `is_differentiable` from property to a constant attribute ([#551](https://github.com/PyTorchLightning/metrics/pull/551))
+
+
 ### Deprecated
 
 

docs/source/pages/overview.rst

@@ -315,23 +315,17 @@ Metrics and differentiability
 
 Metrics support backpropagation, if all computations involved in the metric calculation
 are differentiable. All modular metrics have a property that determines if a metric is
-differentible or not.
-
-.. code-block:: python
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True/False
+differentiable or not.
 
 However, note that the cached state is detached from the computational
-graph and cannot be backpropagated. Not doing this would mean storing the computational
+graph and cannot be back-propagated. Not doing this would mean storing the computational
 graph for each update call, which can lead to out-of-memory errors.
 In practise this means that:
 
 .. code-block:: python
 
     metric = MyMetric()
-    val = metric(pred, target) # this value can be backpropagated
-    val = metric.compute() # this value cannot be backpropagated
+    val = metric(pred, target) # this value can be back-propagated
+    val = metric.compute() # this value cannot be back-propagated
 
 A functional metric is differentiable if its corresponding modular metric is differentiable.

tests/helpers/testers.py

@@ -145,10 +145,12 @@ def _class_test(
     if not metric_args:
         metric_args = {}
 
-    # Instanciate lightning metric
+    # Instantiate lightning metric
     metric = metric_class(
         compute_on_step=check_dist_sync_on_step or check_batch, dist_sync_on_step=dist_sync_on_step, **metric_args
     )
+    with pytest.raises(RuntimeError):
+        metric.is_differentiable = not metric.is_differentiable
 
     # check that the metric is scriptable
     if check_scriptable:

torchmetrics/audio/pit.py

@@ -69,6 +69,7 @@ class PIT(Metric):
         Signal Process. ICASSP, IEEE, New Orleans, LA, 2017: pp. 241–245. https://doi.org/10.1109/ICASSP.2017.7952154.
     """
 
+    is_differentiable = True
     sum_pit_metric: Tensor
     total: Tensor
 
@@ -110,7 +111,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes average PIT metric."""
         return self.sum_pit_metric / self.total
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/audio/si_sdr.py

@@ -64,6 +64,7 @@ class SI_SDR(Metric):
         and Signal Processing (ICASSP) 2019.
     """
 
+    is_differentiable = True
     sum_si_sdr: Tensor
     total: Tensor
 
@@ -101,7 +102,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes average SI-SDR."""
         return self.sum_si_sdr / self.total
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/audio/si_snr.py

@@ -62,6 +62,7 @@ class SI_SNR(Metric):
         696-700, doi: 10.1109/ICASSP.2018.8462116.
     """
 
+    is_differentiable = True
     sum_si_snr: Tensor
     total: Tensor
 
@@ -97,7 +98,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes average SI-SNR."""
         return self.sum_si_snr / self.total
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/audio/snr.py

@@ -70,6 +70,7 @@ class SNR(Metric):
         and Signal Processing (ICASSP) 2019.
 
     """
+    is_differentiable = True
     sum_snr: Tensor
     total: Tensor
 
@@ -107,7 +108,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes average SNR."""
         return self.sum_snr / self.total
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/classification/accuracy.py

@@ -164,6 +164,7 @@ class Accuracy(StatScores):
         tensor(0.6667)
 
     """
+    is_differentiable = False
     correct: Tensor
     total: Tensor
 
@@ -273,7 +274,3 @@ def compute(self) -> Tensor:
             return _subset_accuracy_compute(self.correct, self.total)
         tp, fp, tn, fn = self._get_final_stats()
         return _accuracy_compute(tp, fp, tn, fn, self.average, self.mdmc_reduce, self.mode)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/auc.py

@@ -43,6 +43,7 @@ class AUC(Metric):
             Callback that performs the ``allgather`` operation on the metric state. When ``None``, DDP
             will be used to perform the ``allgather``.
     """
+    is_differentiable = False
     x: List[Tensor]
     y: List[Tensor]
 
@@ -88,9 +89,3 @@ def compute(self) -> Tensor:
         x = dim_zero_cat(self.x)
         y = dim_zero_cat(self.y)
         return _auc_compute(x, y, reorder=self.reorder)
-
-    @property
-    def is_differentiable(self) -> bool:
-        """AUC metrics is considered as non differentiable so it should have `false` value for `is_differentiable`
-        property."""
-        return False

torchmetrics/classification/auroc.py

@@ -99,6 +99,7 @@ class AUROC(Metric):
         tensor(0.7778)
 
     """
+    is_differentiable = False
     preds: List[Tensor]
     target: List[Tensor]
 
@@ -183,9 +184,3 @@ def compute(self) -> Tensor:
             self.average,
             self.max_fpr,
         )
-
-    @property
-    def is_differentiable(self) -> bool:
-        """AUROC metrics is considered as non differentiable so it should have `false` value for
-        `is_differentiable` property."""
-        return False

torchmetrics/classification/average_precision.py

@@ -84,6 +84,7 @@ class AveragePrecision(Metric):
         [tensor(1.), tensor(1.), tensor(0.2500), tensor(0.2500), tensor(nan)]
     """
 
+    is_differentiable = False
     preds: List[Tensor]
     target: List[Tensor]
 
@@ -144,7 +145,3 @@ def compute(self) -> Union[Tensor, List[Tensor]]:
         if not self.num_classes:
             raise ValueError(f"`num_classes` bas to be positive number, but got {self.num_classes}")
         return _average_precision_compute(preds, target, self.num_classes, self.pos_label, self.average)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/cohen_kappa.py

@@ -77,6 +77,7 @@ class labels.
         tensor(0.5000)
 
     """
+    is_differentiable = False
     confmat: Tensor
 
     def __init__(
@@ -116,9 +117,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes cohen kappa score."""
         return _cohen_kappa_compute(self.confmat, self.weights)
-
-    @property
-    def is_differentiable(self) -> bool:
-        """cohen kappa is not differentiable since the implementation is based on calculating the confusion matrix
-        which in general is not differentiable."""
-        return False

torchmetrics/classification/confusion_matrix.py

@@ -91,7 +91,7 @@ class ConfusionMatrix(Metric):
                 [[0., 1.], [0., 1.]]])
 
     """
-
+    is_differentiable = False
     confmat: Tensor
 
     def __init__(
@@ -139,7 +139,3 @@ def compute(self) -> Tensor:
             this will be a `[n_classes, 2, 2]` tensor
         """
         return _confusion_matrix_compute(self.confmat, self.normalize)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/f_beta.py

@@ -269,6 +269,8 @@ class F1(FBeta):
         tensor(0.3333)
     """
 
+    is_differentiable = False
+
     def __init__(
         self,
         num_classes: Optional[int] = None,
@@ -297,7 +299,3 @@ def __init__(
             process_group=process_group,
             dist_sync_fn=dist_sync_fn,
         )
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/hamming_distance.py

@@ -67,6 +67,7 @@ class HammingDistance(Metric):
         tensor(0.2500)
 
     """
+    is_differentiable = False
     correct: Tensor
     total: Tensor
 
@@ -107,7 +108,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes hamming distance based on inputs passed in to ``update`` previously."""
         return _hamming_distance_compute(self.correct, self.total)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/hinge.py

@@ -84,6 +84,7 @@ class Hinge(Metric):
         tensor([2.2333, 1.5000, 1.2333])
 
     """
+    is_differentiable = True
     measure: Tensor
     total: Tensor
 
@@ -124,7 +125,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
 
     def compute(self) -> Tensor:
         return _hinge_compute(self.measure, self.total)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/classification/iou.py

@@ -77,6 +77,7 @@ class IoU(ConfusionMatrix):
         tensor(0.9660)
 
     """
+    is_differentiable = False
 
     def __init__(
         self,
@@ -104,7 +105,3 @@ def __init__(
     def compute(self) -> Tensor:
         """Computes intersection over union (IoU)"""
         return _iou_from_confmat(self.confmat, self.num_classes, self.ignore_index, self.absent_score, self.reduction)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/kl_divergence.py

@@ -61,6 +61,7 @@ class KLDivergence(Metric):
         tensor(0.0853)
 
     """
+    is_differentiable = True
     # TODO: canot be used because if scripting
     # measures: Union[List[Tensor], Tensor]
     total: Tensor
@@ -106,7 +107,3 @@ def update(self, p: Tensor, q: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         measures = dim_zero_cat(self.measures) if self.reduction is None or self.reduction == "none" else self.measures
         return _kld_compute(measures, self.total, self.reduction)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return True

torchmetrics/classification/matthews_corrcoef.py

@@ -73,6 +73,7 @@ class MatthewsCorrcoef(Metric):
         tensor(0.5774)
 
     """
+    is_differentiable = False
     confmat: Tensor
 
     def __init__(
@@ -108,7 +109,3 @@ def update(self, preds: Tensor, target: Tensor) -> None:  # type: ignore
     def compute(self) -> Tensor:
         """Computes matthews correlation coefficient."""
         return _matthews_corrcoef_compute(self.confmat)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/precision_recall.py

@@ -120,6 +120,7 @@ class Precision(StatScores):
         tensor(0.2500)
 
     """
+    is_differentiable = False
 
     def __init__(
         self,
@@ -168,10 +169,6 @@ def compute(self) -> Tensor:
         tp, fp, _, fn = self._get_final_stats()
         return _precision_compute(tp, fp, fn, self.average, self.mdmc_reduce)
 
-    @property
-    def is_differentiable(self) -> bool:
-        return False
-
 
 class Recall(StatScores):
     r"""
@@ -273,6 +270,7 @@ class Recall(StatScores):
         tensor(0.2500)
 
     """
+    is_differentiable = False
 
     def __init__(
         self,
@@ -320,7 +318,3 @@ def compute(self) -> Tensor:
         """
         tp, fp, _, fn = self._get_final_stats()
         return _recall_compute(tp, fp, fn, self.average, self.mdmc_reduce)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/precision_recall_curve.py

@@ -81,6 +81,7 @@ class PrecisionRecallCurve(Metric):
         [tensor([0.7500]), tensor([0.7500]), tensor([0.0500, 0.7500]), tensor([0.0500, 0.7500]), tensor([0.0500])]
     """
 
+    is_differentiable = False
     preds: List[Tensor]
     target: List[Tensor]
 
@@ -146,7 +147,3 @@ def compute(self) -> Union[Tuple[Tensor, Tensor, Tensor], Tuple[List[Tensor], Li
         if not self.num_classes:
             raise ValueError(f"`num_classes` bas to be positive number, but got {self.num_classes}")
         return _precision_recall_curve_compute(preds, target, self.num_classes, self.pos_label)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/roc.py

@@ -103,6 +103,7 @@ class ROC(Metric):
          tensor([1.1837, 0.1837, 0.1338, 0.1183, 0.1138])]
     """
 
+    is_differentiable = False
     preds: List[Tensor]
     target: List[Tensor]
 
@@ -166,7 +167,3 @@ def compute(self) -> Union[Tuple[Tensor, Tensor, Tensor], Tuple[List[Tensor], Li
         if not self.num_classes:
             raise ValueError(f"`num_classes` bas to be positive number, but got {self.num_classes}")
         return _roc_compute(preds, target, self.num_classes, self.pos_label)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False

torchmetrics/classification/specificity.py

@@ -121,6 +121,7 @@ class Specificity(StatScores):
         tensor(0.6250)
 
     """
+    is_differentiable = False
 
     def __init__(
         self,
@@ -168,7 +169,3 @@ def compute(self) -> Tensor:
         """
         tp, fp, tn, fn = self._get_final_stats()
         return _specificity_compute(tp, fp, tn, fn, self.average, self.mdmc_reduce)
-
-    @property
-    def is_differentiable(self) -> bool:
-        return False