fl16180 / nowcast Goto Github PK

(mainly a note to self of what features to add)

1. allow specifying the timestamp variable name rather than requiring "Timestamp" to be used
2. allow renaming columns of datasets during registration. Often disparate datasets will have non-informative or overlapping names which causes confusion and are annoying to rename beforehand.
3. either support multiple target variables or better handle no target variables because sometimes the target isn't well-defined/non-existent e.g. for semi/unsupervised problems. The current workaround is to just define everything as "predictor" and call config.prepared_data instead of config.data