[R-package] lgb.convert functions should warn on unconverted columns of unsupported types

[jameslamb]

Summary

lgb.convert(), and lgb.convert_with_rules() should tell users in a log message if any columns remain unconverted because they are of a type that the function does not support. The names of those columns should be logged.

Motivation

The R package currently exports four functions that can be used to convert tabular datasets into model-ready form:

• lgb.convert(): converts columns of type "character" and "factor" to "integer"
• lgb.convert2(): similar to lgb.convert(), but returns a set of "rules" describing how non-numeric values were mapped to integer values. Also allows for user-provided rules, useful for cases where you want to be sure the encoding is the same on multiple datasets (e.g. training, test, and validation datasets)

These functions are intended to make it easier to create a model-ready dataset (all numeric or integer). The user expectation is likely that after using calling one of these functions on a dataset, that dataset is ready to use in a model. If that dataset contains columns of other types (not integer or numeric), the user should be notified in a log message.

Column types that these functions are unlikely to support:

• POSIX*
• Date
• list
• data.frame
• data.table

[jameslamb]

Closed in favor of being in #2302. We decided to keep all feature requests in one place.

Welcome to contribute this feature! Please re-open this issue (or post a comment if you are not a topic starter) if you are actively working on implementing this feature.

[jameslamb]

I just updated this description and title now that #3095 has been merged.