The library is a collection of the most used metrics within deep learning in pytorch. The library is hugely inspiret by ignites.Metric module, but is intented to be a standalon library for only computing metrics on pytorch tensor. The goal is to get as many of the sklearn.metrics ported to pytorch.
Availble metrics:
-
Regression metrics
- MeanSquaredError
- MeanAbsoluteError
- RootMeanSquaredError
- ExplainedVariance
- R2Score
- MaxError
- MeanSquaredLogarithmicError
- MeanTweedieDeviance
- MeanPoissonDeviance
- MeanGammaDeviance
- CosineSimilarity
- Correlation
-
Classification metrics
- Accuracy
- BalancedAccuracy
- FilteredAccuracy
- Recall
- Precision
- BalancedAccuracy
- F1
- TopKAccuracy
- ROC
- AUC
- ConfusionMatrix
The library have a number of wrappers that (as the name suggest) can be use in combination with the selection of metrics to offer extra functionality:
- MetricCollection
- RunningAverage
- Sum, Mean, Product
Please see the docs for more information.