More performance metrics

[jarbet]

My understanding of v1.1 on CRAN is that only 1 performance metrics is implemented for regression (mean absolute error?) and classification (overall accuracy?). It would be great to implement more metrics like:

Numeric outcomes

• R-squared
• mean absolute error

Binary outcomes

• AUC
• PRAUC
• F1
• sensitivity
• specificity
• PPV
• NPV
• Matthew's correlation coefficient

3+ category outcomes

• "one-vs-all" versions of all of the binary metrics, as well as the average across classes

[matloff]

Thanks for the suggestions! Unfortunately, this is not something that would be the best use of my time spent on the qeML, as there are many other things I want to do, big and small, to improve/extend the package. A big problem from my point of view is that I don't think some of those accuracy criteria are useful/appropriate, such as AUC and F1. Yes, yes lots of people use them but I have problems with them, and again, that means I don't want to spend precious time on adding them.

…

On Wed, Nov 20, 2024 at 11:05 AM Jaron Arbet ***@***.***> wrote: My understanding of v1.1 on CRAN is that only 1 performance metrics is implemented for regression (mean absolute error?) and classification (overall accuracy?). It would be great to implement more metrics like: *Numeric outcomes* - R-squared - mean absolute error *Binary outcomes* - AUC - PRAUC - F1 - sensitivity - specificity - PPV - NPV - Matthew's correlation coefficient *3+ category outcomes* - "one-vs-all" versions of all of the binary metrics, as well as the average across classes — Reply to this email directly, view it on GitHub <#16>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABZ34ZLAG5AJZI7WN4A3YZL2BTMQXAVCNFSM6AAAAABSFJBFX2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGY3TMOBYGYYDENI> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>