Newmetric: NRMSE

CHANGELOG.md

@@ -12,7 +12,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
--
+- Added `NormalizedRootMeanSquaredError` metric to regression subpackage ([#2442](https://github.com/Lightning-AI/torchmetrics/pull/2442))
 
 
 ### Changed

docs/source/conf.py

@@ -447,6 +447,9 @@ def linkcode_resolve(domain, info) -> Optional[str]:  # noqa: ANN001
     "https://aclanthology.org/W17-4770",
     # A wavelet transform method to merge Landsat TM and SPOT panchromatic data
     "https://www.ingentaconnect.com/content/tandf/tres/1998/00000019/00000004/art00013",
+    # Improved normalization of time-lapse seismic data using normalized root mean square repeatability data ...
+    # ... to improve automatic production and seismic history matching in the Nelson field
+    "https://onlinelibrary.wiley.com/doi/abs/10.1111/1365-2478.12109",
     # todo: these links seems to be unstable, referring to .devcontainer
     "https://code.visualstudio.com",
     "https://code.visualstudio.com/.*",

docs/source/links.rst

@@ -171,6 +171,7 @@
 .. _FLORES-200: https://arxiv.org/abs/2207.04672
 .. _averaging curve objects: https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html
 .. _SCC: https://www.ingentaconnect.com/content/tandf/tres/1998/00000019/00000004/art00013
+.. _Normalized Root Mean Squared Error: https://onlinelibrary.wiley.com/doi/abs/10.1111/1365-2478.12109
 .. _Generalized Dice Score: https://arxiv.org/abs/1707.03237
 .. _Hausdorff Distance: https://en.wikipedia.org/wiki/Hausdorff_distance
 .. _averaging curve objects: https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

docs/source/regression/normalized_root_mean_squared_error.rst

@@ -0,0 +1,21 @@
+.. customcarditem::
+   :header: Normalized Root Mean Squared Error (NRMSE)
+   :image: https://pl-flash-data.s3.amazonaws.com/assets/thumbnails/tabular_classification.svg
+   :tags: Regression
+
+.. include:: ../links.rst
+
+##########################################
+Normalized Root Mean Squared Error (NRMSE)
+##########################################
+
+Module Interface
+________________
+
+.. autoclass:: torchmetrics.NormalizedRootMeanSquaredError
+    :exclude-members: update, compute
+
+Functional Interface
+____________________
+
+.. autofunction:: torchmetrics.functional.normalized_root_mean_squared_error

requirements/_devel.txt

@@ -20,3 +20,4 @@
 -r classification_test.txt
 -r nominal_test.txt
 -r segmentation_test.txt
+-r regression_test.txt

requirements/regression_test.txt

@@ -0,0 +1 @@
+permetrics==2.0.0

src/torchmetrics/__init__.py

@@ -114,6 +114,7 @@
     MeanSquaredError,
     MeanSquaredLogError,
     MinkowskiDistance,
+    NormalizedRootMeanSquaredError,
     PearsonCorrCoef,
     R2Score,
     RelativeSquaredError,
@@ -158,25 +159,23 @@
 )
 
 __all__ = [
-    "functional",
-    "Accuracy",
     "AUROC",
+    "Accuracy",
     "AveragePrecision",
     "BLEUScore",
     "BootStrapper",
+    "CHRFScore",
     "CalibrationError",
     "CatMetric",
-    "ClasswiseWrapper",
     "CharErrorRate",
-    "CHRFScore",
-    "ConcordanceCorrCoef",
+    "ClasswiseWrapper",
     "CohenKappa",
+    "ConcordanceCorrCoef",
     "ConfusionMatrix",
     "CosineSimilarity",
     "CramersV",
     "CriticalSuccessIndex",
     "Dice",
-    "TweedieDevianceScore",
     "ErrorRelativeGlobalDimensionlessSynthesis",
     "ExactMatch",
     "ExplainedVariance",
@@ -187,8 +186,8 @@
     "HammingDistance",
     "HingeLoss",
     "JaccardIndex",
-    "KendallRankCorrCoef",
     "KLDivergence",
+    "KendallRankCorrCoef",
     "LogCoshError",
     "MatchErrorRate",
     "MatthewsCorrCoef",
@@ -201,23 +200,25 @@
     "Metric",
     "MetricCollection",
     "MetricTracker",
-    "MinkowskiDistance",
     "MinMaxMetric",
     "MinMetric",
+    "MinkowskiDistance",
     "ModifiedPanopticQuality",
+    "MultiScaleStructuralSimilarityIndexMeasure",
     "MultioutputWrapper",
     "MultitaskWrapper",
-    "MultiScaleStructuralSimilarityIndexMeasure",
+    "NormalizedRootMeanSquaredError",
     "PanopticQuality",
+    "PeakSignalNoiseRatio",
     "PearsonCorrCoef",
     "PearsonsContingencyCoefficient",
     "PermutationInvariantTraining",
     "Perplexity",
     "Precision",
     "PrecisionAtFixedRecall",
     "PrecisionRecallCurve",
-    "PeakSignalNoiseRatio",
     "R2Score",
+    "ROC",
     "Recall",
     "RecallAtFixedPrecision",
     "RelativeAverageSpectralError",
@@ -228,37 +229,38 @@
     "RetrievalMRR",
     "RetrievalNormalizedDCG",
     "RetrievalPrecision",
-    "RetrievalRecall",
-    "RetrievalRPrecision",
     "RetrievalPrecisionRecallCurve",
+    "RetrievalRPrecision",
+    "RetrievalRecall",
     "RetrievalRecallAtFixedPrecision",
-    "ROC",
     "RootMeanSquaredErrorUsingSlidingWindow",
     "RunningMean",
     "RunningSum",
+    "SQuAD",
     "SacreBLEUScore",
-    "SignalDistortionRatio",
     "ScaleInvariantSignalDistortionRatio",
     "ScaleInvariantSignalNoiseRatio",
+    "SensitivityAtSpecificity",
+    "SignalDistortionRatio",
     "SignalNoiseRatio",
     "SpearmanCorrCoef",
     "Specificity",
     "SpecificityAtSensitivity",
-    "SensitivityAtSpecificity",
     "SpectralAngleMapper",
     "SpectralDistortionIndex",
-    "SQuAD",
-    "StructuralSimilarityIndexMeasure",
     "StatScores",
+    "StructuralSimilarityIndexMeasure",
     "SumMetric",
     "SymmetricMeanAbsolutePercentageError",
     "TheilsU",
     "TotalVariation",
     "TranslationEditRate",
     "TschuprowsT",
+    "TweedieDevianceScore",
     "UniversalImageQualityIndex",
     "WeightedMeanAbsolutePercentageError",
     "WordErrorRate",
     "WordInfoLost",
     "WordInfoPreserved",
+    "functional",
 ]

src/torchmetrics/functional/__init__.py

@@ -100,6 +100,7 @@
     mean_squared_error,
     mean_squared_log_error,
     minkowski_distance,
+    normalized_root_mean_squared_error,
     pearson_corrcoef,
     r2_score,
     relative_squared_error,
@@ -146,14 +147,13 @@
     "calibration_error",
     "char_error_rate",
     "chrf_score",
-    "concordance_corrcoef",
     "cohen_kappa",
+    "concordance_corrcoef",
     "confusion_matrix",
     "cosine_similarity",
     "cramers_v",
     "cramers_v_matrix",
     "critical_success_index",
-    "tweedie_deviance_score",
     "dice",
     "error_relative_global_dimensionless_synthesis",
     "exact_match",
@@ -177,63 +177,65 @@
     "mean_squared_log_error",
     "minkowski_distance",
     "multiscale_structural_similarity_index_measure",
+    "normalized_root_mean_squared_error",
     "pairwise_cosine_similarity",
     "pairwise_euclidean_distance",
     "pairwise_linear_similarity",
     "pairwise_manhattan_distance",
     "pairwise_minkowski_distance",
     "panoptic_quality",
+    "peak_signal_noise_ratio",
     "pearson_corrcoef",
     "pearsons_contingency_coefficient",
     "pearsons_contingency_coefficient_matrix",
     "permutation_invariant_training",
     "perplexity",
     "pit_permutate",
     "precision",
+    "precision_at_fixed_recall",
     "precision_recall_curve",
-    "peak_signal_noise_ratio",
     "r2_score",
     "recall",
+    "recall_at_fixed_precision",
     "relative_average_spectral_error",
     "relative_squared_error",
     "retrieval_average_precision",
     "retrieval_fall_out",
     "retrieval_hit_rate",
     "retrieval_normalized_dcg",
     "retrieval_precision",
+    "retrieval_precision_recall_curve",
     "retrieval_r_precision",
     "retrieval_recall",
     "retrieval_reciprocal_rank",
-    "retrieval_precision_recall_curve",
     "roc",
     "root_mean_squared_error_using_sliding_window",
     "rouge_score",
     "sacre_bleu_score",
-    "signal_distortion_ratio",
     "scale_invariant_signal_distortion_ratio",
     "scale_invariant_signal_noise_ratio",
+    "sensitivity_at_specificity",
+    "signal_distortion_ratio",
     "signal_noise_ratio",
     "spearman_corrcoef",
     "specificity",
+    "specificity_at_sensitivity",
+    "spectral_angle_mapper",
     "spectral_distortion_index",
     "squad",
-    "structural_similarity_index_measure",
     "stat_scores",
+    "structural_similarity_index_measure",
     "symmetric_mean_absolute_percentage_error",
     "theils_u",
     "theils_u_matrix",
     "total_variation",
     "translation_edit_rate",
     "tschuprows_t",
     "tschuprows_t_matrix",
+    "tweedie_deviance_score",
     "universal_image_quality_index",
-    "spectral_angle_mapper",
     "weighted_mean_absolute_percentage_error",
     "word_error_rate",
     "word_information_lost",
     "word_information_preserved",
-    "precision_at_fixed_recall",
-    "recall_at_fixed_precision",
-    "sensitivity_at_specificity",
-    "specificity_at_sensitivity",
 ]

src/torchmetrics/functional/regression/__init__.py

@@ -23,6 +23,7 @@
 from torchmetrics.functional.regression.mape import mean_absolute_percentage_error
 from torchmetrics.functional.regression.minkowski import minkowski_distance
 from torchmetrics.functional.regression.mse import mean_squared_error
+from torchmetrics.functional.regression.nrmse import normalized_root_mean_squared_error
 from torchmetrics.functional.regression.pearson import pearson_corrcoef
 from torchmetrics.functional.regression.r2 import r2_score
 from torchmetrics.functional.regression.rse import relative_squared_error
@@ -39,13 +40,14 @@
     "kendall_rank_corrcoef",
     "kl_divergence",
     "log_cosh_error",
-    "mean_squared_log_error",
     "mean_absolute_error",
-    "mean_squared_error",
-    "pearson_corrcoef",
     "mean_absolute_percentage_error",
     "mean_absolute_percentage_error",
+    "mean_squared_error",
+    "mean_squared_log_error",
     "minkowski_distance",
+    "normalized_root_mean_squared_error",
+    "pearson_corrcoef",
     "r2_score",
     "relative_squared_error",
     "spearman_corrcoef",