factor2integer or arbitrary transformations

[kevinykuo]

Could we have a step_factor2int() or something similar to convert factors into integers? Ideally with a from parameter that specifies the indexing (e.g. 0- or 1- based). Use case here is feeding data into embedding layers of keras models.

Alternatively, maybe there can be a step_lambda() that takes a user defined function for random use cases like this?

[topepo]

Could we have a step_factor2int()

This goes against every notation that I hold dear... but I know why you need it for sparks so okay.

Alternatively, maybe there can be a step_lambda() that takes a user defined function for random use cases like this?

We'll be adding some dplyr steps like mutate and that would solve the issue too.

[kevinykuo]

Actually, what we really need is a step_stringindexer() or a step_labelencoder(), which should be less cringy than step_factor2int(). These also exist in sklearn and Spark.

Either way, we'll be in a better state than the current workaround in the wild which is "step_string2factor(ordered = TRUE) %>% step_ordinalscore()".