Metrics

Aggregation Metrics

AUC	Computes Area Under the Curve (AUC) using the trapezoidal rule.
Cat	Concatenate all input tensors along dimension dim.
Max	Calculate the maximum value of all elements in all the input tensors.
Mean	Calculate the weighted mean value of all elements in all the input tensors.
Min	Calculate the minimum value of all elements in all the input tensors.
Sum	Calculate the weighted sum value of all elements in all the input tensors.
Throughput	Calculate the throughput value which is the number of elements processed per second.

None

FrechetAudioDistance

Computes the Fréchet distance between predicted and target audio waveforms.

Classification Metrics

BinaryAccuracy	Compute binary accuracy score, which is the frequency of input matching target.
BinaryAUPRC	Compute AUPRC, also called Average Precision, which is the area under the Precision-Recall Curve, for binary classification.
BinaryAUROC	Compute AUROC, which is the area under the ROC Curve, for binary classification.
BinaryBinnedAUROC	Compute AUROC, which is the area under the ROC Curve, for binary classification.
BinaryBinnedAUPRC	Compute Binned AUPRC, which is the area under the binned version of the Precision Recall Curve, for binary classification.
BinaryBinnedPrecisionRecallCurve	Compute precision recall curve with given thresholds.
BinaryConfusionMatrix	Compute binary confusion matrix, a 2 by 2 tensor with counts ( (true positive, false negative) , (false positive, true negative) ) See also MulticlassConfusionMatrix
BinaryF1Score	Compute binary f1 score, which is defined as the harmonic mean of precision and recall.
BinaryNormalizedEntropy	Compute the normalized binary cross entropy between predicted input and ground-truth binary target.
BinaryPrecision	Compute the precision score for binary classification tasks, which is calculated as the ratio of the true positives and the sum of true positives and false positives.
BinaryPrecisionRecallCurve	Returns precision-recall pairs and their corresponding thresholds for binary classification tasks.
BinaryRecall	Compute the recall score for binary classification tasks, which is calculated as the ratio of the true positives and the sum of true positives and false negatives.
BinaryRecallAtFixedPrecision	Returns the highest possible recall value give the minimum precision for binary classification tasks.
MulticlassAccuracy	Compute accuracy score, which is the frequency of input matching target.
MulticlassAUPRC	Compute AUPRC, also called Average Precision, which is the area under the Precision-Recall Curve, for multiclass classification.
MulticlassAUROC	Compute AUROC, which is the area under the ROC Curve, for multiclass classification in a one vs rest fashion.
MulticlassBinnedAUPRC	Compute Binned AUPRC, which is the area under the binned version of the Precision Recall Curve, for multiclass classification.
MulticlassBinnedAUROC	Compute AUROC, which is the area under the ROC Curve, for multiclass classification.
MulticlassBinnedPrecisionRecallCurve	Compute precision recall curve with given thresholds.
MulticlassConfusionMatrix	Compute multi-class confusion matrix, a matrix of dimension num_classes x num_classes where each element at position (i,j) is the number of examples with true class i that were predicted to be class j.
MulticlassF1Score	Compute f1 score, which is defined as the harmonic mean of precision and recall.
MulticlassPrecision	Compute the precision score, the ratio of the true positives and the sum of true positives and false positives.
MulticlassPrecisionRecallCurve	Returns precision-recall pairs and their corresponding thresholds for multi-class classification tasks.
MulticlassRecall	Compute the recall score, the ratio of the true positives and the sum of true positives and false negatives.
MultilabelAccuracy	Compute multilabel accuracy score, which is the frequency of input matching target.
MultilabelAUPRC	Compute AUPRC, also called Average Precision, which is the area under the Precision-Recall Curve, for multilabel classification.
MultilabelBinnedAUPRC	Compute Binned AUPRC, which is the area under the binned version of the Precision Recall Curve, for multilabel classification.
MultilabelBinnedPrecisionRecallCurve	Compute precision recall curve with given thresholds.
MultilabelPrecisionRecallCurve	Returns precision-recall pairs and their corresponding thresholds for multi-label classification tasks.
MultilabelRecallAtFixedPrecision	Returns the highest possible recall value given the minimum precision for each label and their corresponding thresholds for multi-label classification tasks.
TopKMultilabelAccuracy	Compute multilabel accuracy score, which is the frequency of the top k label predicted matching target.

Image Metrics

FrechetInceptionDistance
PeakSignalNoiseRatio	Compute the PSNR (Peak Signal to Noise Ratio) between two images.
StructuralSimilarity	Compute the structural similarity index (SSIM) between two sets of images.

Ranking Metrics

ClickThroughRate	Compute the click through rate given click events.
HitRate	Compute the hit rate of the correct class among the top predicted classes.
ReciprocalRank	Compute the reciprocal rank of the correct class among the top predicted classes.
RetrievalPrecision	Compute the retrieval precision.
RetrievalRecall	Compute the retrieval recall.
WeightedCalibration	Compute weighted calibration metric.

Regression Metrics

MeanSquaredError	Compute Mean Squared Error, which is the mean of squared error of input and target.
R2Score	Compute R-squared score, which is the proportion of variance in the dependent variable that can be explained by the independent variable.

Statistical Metrics

Text Metrics

BLEUScore	Compute BLEU score (https://en.wikipedia.org/wiki/BLEU) given translations and references.
Perplexity	Perplexity measures how well a model predicts sample data.
WordErrorRate	Compute the word error rate of the predicted word sequence(s) with the reference word sequence(s).
WordInformationLost	Word Information Lost (WIL) is a metric of the performance of an automatic speech recognition system.
WordInformationPreserved	Compute the word information preserved of the predicted word sequence(s) with the reference word sequence(s).

Windowed Metrics

WindowedBinaryAUROC	The windowed version of BinaryAUROC.
WindowedBinaryNormalizedEntropy	The windowed version of BinaryNormalizedEntropy that provides both windowed and liftime values.
WindowedClickThroughRate	The windowed version of ClickThroughRate that provides both windowed and lifetime values.
WindowedMeanSquaredError	The windowed version of Mean Squared Error that provides both windowed and liftime values.
WindowedWeightedCalibration	Compute weighted calibration metric.