- Add transformation as angurment to train
- Syntax on predict, when, give values for time series
- Update the documentation, we changed is null boolean dimension for not null
- Print stats data on stats generator
- Break demo code into train/predict
- Only create time series col representation if average(count(distinct column)) in column by group by is <=1
- On time series, break test-train automatically
- When model is finished training, test by column, store importance matrix
- Distributions on test and validation sets
- Fix time series include target column for past readings (IMPORTANT)
- Update loss function to be log relative
- Implement logging proxy visualizer
- Support from dataframe directly
- Support stream dataframes (http://matthewrocklin.com/blog/work/2017/10/16/streaming-dataframes-1)