Preston Parry

documentation goals

2

- [ ] have an examples folder - [ ] basic regressor - [ ] basic classifier - [ ] advanced options - [ ] same, but in an ipython...

gs issues

1

parallelism doesn't work. probably need to bring back copy_reg stuff. we didn't catch this because our test suite runs single-threaded, because nested parallelism fails. the user has to pass in...

this human was killing it: https://www.kaggle.com/c/porto-seguro-safe-driver-prediction/discussion/44629

- [ ] what percent of all predictions are actually over/under each percentile - [ ] how tight each band is - [ ] how this compares to X*training rmse...

we'll probably want to add in some logging, or make that the return value, in place of what we currently have for return_X_transformed. and, i guess in that case, we...

we also need to add a test back in (now that we've split up our test suite) that has parallelism involved, so we can catch errors like this.

these failures seem to have been caused by some unknown feature scaling interaction, that doesn't affect deep learning or linear models by themselves... https://travis-ci.org/ClimbsRocks/auto_ml/jobs/312215168

it spikes once between when we log the final training_params, and when we start data cleaning. then, it seems to spike once more after we start fitting data cleaning, and...

- [ ] lgbm quantile regression - [ ] dfv speedups - [ ] test suite refactor - [ ] nanstd - [ ] that's it. - [ ] specifically...