Support different types of prediction

[isabelizimm]

Currently, vetiver supports the predict method when handling models. In other use cases, it maybe important to be able to use predict_proba or predict_log_proba.

[juliasilge]

On the R side, we handle this by passing in an argument about the type of predictions. It would look something like this:

pr() %>% vetiver_pr_predict(v, type = "prob")
## or 
pr() %>% vetiver_pr_predict(v, type = "quantile")
## and so on

[farhadsalimi]

Hi there,

Has there been any progress on this?
I am particulalry interested in predict_proba, is there a work around that you can direct me to meanwhile?

Many thanks

[isabelizimm]

Hi there 👋 -- there has not been worked on yet (for context: nowadays I am spending most of my time in this repository -> https://github.com/posit-dev/positron, but am still around to do maintenance and features as needed :D)

For predict_proba, there are a few current workarounds. The first would be would be able to add another endpoint with VetiverAPI.vetiver_post if you wanted to have both the /predict and /predict_proba options. In this case, you would create a function that returns the predict_proba values and then add it to the API. Another option is to create a custom handler for this prediction type, where the handler_predict would return the output of predict_proba.

Let me know if you run into any troubles or if you have any followup questions!

[andre-sun]

Hi Isabel, has there been any plans to include seamless compatibility with predict_proba like @farhadsalimi has mention? Otherwise, where would be the best place to pose an issue/question I have regarding creating a custom_handler that doesn't seem to be working?

Thanks!