Make regression recognize strings/enums/categoricals and convert to dummies AND also make helper function for interaction.

[deanm0000]

It'd be a nice feature if it would recognize str/enum/cat columns and automatically convert them to dummies.

You'd have to have a for loop on the inputs checking their type and if it was one of those types it'd have to convert them to dummies. Since df.to_dummies already exists that shouldn't be too bad.

For the interactions, on the user side I'm thinking it'd be like

from polars_ds import interaction_helper as ih
import polars_ds as pds

df.select(pds.lin_reg_report( ih("color", "height"), "color", "height", target="width", add_bias=True)

The only way I can think to implement that would be to haveih wrap the columns to be interacted in a pl.struct so that then on the rust side we can check if it's a struct column and if it is then it first checks to see if it should make dummies out of the sub-columns and then multiplies the columns in the struct to form a single new column.

But wait there's more, the helper function wouldn't be merely a wrapper for pl.struct, it could have parameters to allow only typing the columns once so instead of the above you could do df.select(pds.lin_reg_report( *ih("color", "height", include_main=True), target="width", add_bias=True)

Additionally it could have product (or maybe call it cartesian) so if the user puts in 3 or more variables it would do all the interactions between the 3 or if set to False it would only do the 3 together.

that would be like

df.select(pds.lin_reg_report( *ih("color", "shape", "height", include_main=True, product=True), target="width", add_bias=True)

which would have regressors: red_round_height, red_round, red_square_height, red_square, red_height, ..., green_height. (I don't want to type any more examples but the colors turn into dummies, the shape turns into dummies and then height stays a float and then all the possible combinations between them become their own interactions.