Add more metrics

[anvelezec]

The idea is to add more metrics taking as basis torchmetrics and tf metrics implementation.

Guidelines

• For implementation purposes reference to accuracy. Take into account each metric needs and update and compute method as does torchmetrics implementation.
• Some of metrics can inherit from modules such as Mean, Reduce, Metric while others need an implementation from scratch.
• Must add unit test for numerical equivalence with respect to torchmetrics or tf metrics

Regression metrics

• [ ] R2
• [ ] Root mean squared error
• [ ] Mean squared error
• [ ] Cosine similarity

Classification metrics

• [ ] IOU
• [ ] ROC
• [ ] Specificity
• [ ] AUC
• [ ] Precision and recall
• [ ] Cohen kappa
• [ ] Confusion matrix
• [ ] Hamming distance

If you want to work on a metric, feel free to "claim" it via a comment here so others know you are working on it. Additionally feel free to add as many metrics as desired, even from other tasks such as audio, language, image, etc.

[jmarrietar]

I will take Mean squared error. I will post here any or in discord the challenges or doubts I face.