Taylor Miller

## Background There is a ton of logic for classification only properties that is making maintenance burdensome. Extract out two new classes `TrainedRegressor` and `TrainedClassifier` (or other better names) that...

>Very small nitpick: should the prediction object be returned sorted? In case user wants to find the 2nd highest, they can just select the[1] entry? ## What it currently looks...

Implement XGBoost Multiclass

4

@Aylr commented on [Sat Jun 03 2017](https://github.com/HealthCatalyst/healthcareai-r/issues/465) ## Implementation - [ ] try scikit learn wrapper: https://github.com/dmlc/xgboost/blob/master/python-package/xgboost/sklearn.py - [ ] algorithm in AdvancedTrainer - [ ] tests - [ ]...

Implement Logging

1

## Notes This is a nice to have, and may not be in the MVP. ## Details - [ ] implement a logger - [ ] implement a config file...

## Background - this should be added to all categorical columns to help deal with unseen factors in new data - [ ] add to DataFrameImputer and possibly DataFrameDummifier

## Background - The tests for SupervisedModelTrainer and AdvancedSupervisedModelTrainer are needlessly complex. - Most of them are really testing the data preparation pipeline. - These tests should be extracted out...

Found a great example here: http://surprise.readthedocs.io - [ ]

Note you may run into trouble if dependencies are not built as conda packages and available in conda-forge. So, you may have to: - build sklearn yourself `conda skeleton pypi...

...and replace it with: `dict(list(defaults.items()) + list(user.items()))`