Improve built-in index_string data format #68
Description
Description
Option 1: save both mappings in the analysis. Could make it a different transformer/aggregator? Probably not
Option 2: At startup time, predict.py can create the mapping of string -> index to speed up transformations (keep index -> string array also, for reverse transformations). Is it custom for just this analysis? Or is there an API for this (e.g. "preprocess")?
Motivation
Faster real-time predictions