Add missing higher_is_better attribute to metrics (#584)

Co-authored-by: Jirka <jirka.borovec@seznam.cz>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

CHANGELOG.md

@@ -25,7 +25,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added support for float targets in `nDCG` metric ([#437](https://github.com/PyTorchLightning/metrics/pull/437))
 - Added `average` argument to `AveragePrecision` metric for reducing multi-label and multi-class problems ([#477](https://github.com/PyTorchLightning/metrics/pull/477))
 - Added `MultioutputWrapper` ([#510](https://github.com/PyTorchLightning/metrics/pull/510))
-- Added metric sweeping `higher_is_better` as constant attribute ([#544](https://github.com/PyTorchLightning/metrics/pull/544))
+- Added metric sweeping:
+  - `higher_is_better` as constant attribute ([#544](https://github.com/PyTorchLightning/metrics/pull/544))
+  - `higher_is_better` to rest of codebase ([#584](https://github.com/PyTorchLightning/metrics/pull/584))
 - Added simple aggregation metrics: `SumMetric`, `MeanMetric`, `CatMetric`, `MinMetric`, `MaxMetric` ([#506](https://github.com/PyTorchLightning/metrics/pull/506))
 - Added pairwise submodule with metrics ([#553](https://github.com/PyTorchLightning/metrics/pull/553))
   - `pairwise_cosine_similarity`

torchmetrics/audio/snr.py

@@ -71,6 +71,7 @@ class SNR(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = True
     sum_snr: Tensor
     total: Tensor
 

torchmetrics/classification/accuracy.py

@@ -165,6 +165,7 @@ class Accuracy(StatScores):
 
     """
     is_differentiable = False
+    higher_is_better = True
     correct: Tensor
     total: Tensor
 

torchmetrics/classification/auroc.py

@@ -105,6 +105,7 @@ class AUROC(Metric):
 
     """
     is_differentiable = False
+    higher_is_better = True
     preds: List[Tensor]
     target: List[Tensor]
 

torchmetrics/classification/calibration_error.py

@@ -60,6 +60,7 @@ class CalibrationError(Metric):
             default: None (which selects the entire world)
     """
     DISTANCES = {"l1", "l2", "max"}
+    higher_is_better = False
     confidences: List[Tensor]
     accuracies: List[Tensor]
 

torchmetrics/classification/cohen_kappa.py

@@ -78,6 +78,7 @@ class labels.
 
     """
     is_differentiable = False
+    higher_is_better = True
     confmat: Tensor
 
     def __init__(

torchmetrics/classification/f_beta.py

@@ -270,6 +270,7 @@ class F1(FBeta):
     """
 
     is_differentiable = False
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/classification/hamming_distance.py

@@ -68,6 +68,7 @@ class HammingDistance(Metric):
 
     """
     is_differentiable = False
+    higher_is_better = False
     correct: Tensor
     total: Tensor
 

torchmetrics/classification/hinge.py

@@ -85,6 +85,7 @@ class Hinge(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = False
     measure: Tensor
     total: Tensor
 

torchmetrics/classification/iou.py

@@ -78,6 +78,7 @@ class IoU(ConfusionMatrix):
 
     """
     is_differentiable = False
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/classification/kl_divergence.py

@@ -62,6 +62,7 @@ class KLDivergence(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = False
     # TODO: canot be used because if scripting
     # measures: Union[List[Tensor], Tensor]
     total: Tensor

torchmetrics/classification/matthews_corrcoef.py

@@ -74,6 +74,7 @@ class MatthewsCorrcoef(Metric):
 
     """
     is_differentiable = False
+    higher_is_better = True
     confmat: Tensor
 
     def __init__(

torchmetrics/classification/precision_recall.py

@@ -121,6 +121,7 @@ class Precision(StatScores):
 
     """
     is_differentiable = False
+    higher_is_better = True
 
     def __init__(
         self,
@@ -271,6 +272,7 @@ class Recall(StatScores):
 
     """
     is_differentiable = False
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/classification/specificity.py

@@ -122,6 +122,7 @@ class Specificity(StatScores):
 
     """
     is_differentiable = False
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/image/fid.py

@@ -204,6 +204,7 @@ class FID(Metric):
     """
     real_features: List[Tensor]
     fake_features: List[Tensor]
+    higher_is_better = False
 
     def __init__(
         self,

torchmetrics/image/inception.py

@@ -101,6 +101,7 @@ class IS(Metric):
 
     """
     features: List
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/image/kid.py

@@ -164,6 +164,7 @@ class KID(Metric):
     """
     real_features: List[Tensor]
     fake_features: List[Tensor]
+    higher_is_better = False
 
     def __init__(
         self,

torchmetrics/image/lpip_similarity.py

@@ -88,6 +88,7 @@ class LPIPS(Metric):
     """
 
     is_differentiable = True
+    higher_is_better = False
     real_features: List[Tensor]
     fake_features: List[Tensor]
 

torchmetrics/image/psnr.py

@@ -69,6 +69,7 @@ class PSNR(Metric):
     """
     min_target: Tensor
     max_target: Tensor
+    higher_is_better = False
 
     def __init__(
         self,

torchmetrics/image/ssim.py

@@ -52,6 +52,7 @@ class SSIM(Metric):
 
     preds: List[Tensor]
     target: List[Tensor]
+    higher_is_better = True
 
     def __init__(
         self,

torchmetrics/regression/cosine_similarity.py

@@ -62,6 +62,7 @@ class CosineSimilarity(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = True
     preds: List[Tensor]
     target: List[Tensor]
 

torchmetrics/regression/explained_variance.py

@@ -78,6 +78,7 @@ class ExplainedVariance(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = True
     n_obs: Tensor
     sum_error: Tensor
     sum_squared_error: Tensor

torchmetrics/regression/mean_absolute_error.py

@@ -49,6 +49,7 @@ class MeanAbsoluteError(Metric):
         tensor(0.5000)
     """
     is_differentiable = True
+    higher_is_better = False
     sum_abs_error: Tensor
     total: Tensor
 

torchmetrics/regression/mean_absolute_percentage_error.py

@@ -58,6 +58,7 @@ class MeanAbsolutePercentageError(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = False
     sum_abs_per_error: Tensor
     total: Tensor
 

torchmetrics/regression/mean_squared_error.py

@@ -52,6 +52,7 @@ class MeanSquaredError(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = False
     sum_squared_error: Tensor
     total: Tensor
 

torchmetrics/regression/mean_squared_log_error.py

@@ -53,6 +53,7 @@ class MeanSquaredLogError(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = False
     sum_squared_log_error: Tensor
     total: Tensor
 

torchmetrics/regression/pearson.py

@@ -86,6 +86,7 @@ class PearsonCorrcoef(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = None  # both -1 and 1 are optimal
     preds: List[Tensor]
     target: List[Tensor]
     mean_x: Tensor

torchmetrics/regression/r2.py

@@ -88,6 +88,7 @@ class R2Score(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = True
     sum_squared_error: Tensor
     sum_error: Tensor
     residual: Tensor

torchmetrics/regression/spearman.py

@@ -54,6 +54,7 @@ class SpearmanCorrcoef(Metric):
 
     """
     is_differentiable = False
+    higher_is_better = True
     preds: List[Tensor]
     target: List[Tensor]
 

torchmetrics/regression/symmetric_mean_absolute_percentage_error.py

@@ -55,6 +55,7 @@ class SymmetricMeanAbsolutePercentageError(Metric):
         tensor(0.2290)
     """
     is_differentiable = True
+    higher_is_better = False
     sum_abs_per_error: Tensor
     total: Tensor
 

torchmetrics/regression/tweedie_deviance.py

@@ -75,6 +75,7 @@ class TweedieDevianceScore(Metric):
 
     """
     is_differentiable = True
+    higher_is_better = None  # TODO: both -1 and 1 are optimal
     sum_deviance_score: Tensor
     num_observations: Tensor