Pyllars

Added
- Helpers for checking and collecting dask futures
- Validation helpers for non-pydata types
- Simple BoW and numeric feature handler
- Helpers for standard plots
- Helpers for MyGene.py
- Helpers for working with the Gene Ontology
- Helper to create scaler from means and standard deviations

Updated
- k-fold splitter to include validation set
- Transparent file opening for compressed files
- Split out machine learning helpers into a separate module
- Split out statistical helpers into a separate module

Fixed
- Missing data one hot encoder to handle sparse inputs

Added
- Followup table construction for MIMIC
- Helpers for MIMIC waveform database
- Additional validation helpers

Updated
- Dataset manager to optionally encode the target variable

Fixed
- SCIP output parsing for instances where SCIP crashed

Added
- Helper to estimate categorical variable MLEs
- Utilities for working with the SCIP solver
- Helper to check if class attributes have been initialized similar to
`check_is_fitted` in sklearn
- Standard validation helpers (`validation_utils`)
- Helper to calculate many regression, multi-class classification metrics

Updated
- Cross-validation helper to work with unsupervised learning
- Documentation of all modules to use docstrings

Removed
- mysql helpers
- automl helpers. These are now available in the [automl-utils](https://github.com/bmmalone/automl-utils)
package.

Added
- Helper for creating chunks of groups from a data frame. This utility can
make submitting jobs to dask and other parallel schedulers more efficient.

Removed
- Dependency on pystan and the `pickle-stan` script. There were no other uses
of stan within this package.

Added
- Utility for supressing pystan (or other compiled function) output. This
addition is motivated by [an rpbp issue](https://github.com/dieterich-lab/rp-bp/issues/10),
and the solution is basically [copied from facebook's prophet](https://github.com/facebook/prophet/issues/223#issuecomment-326455744).
- DatasetManager class to ease reading data and preparing it for sklearn
- Several changes to the ML helper transformers to make working with mixed
data sets (that is, those with categorical and numerical features) easier
- Several sklearn transformers which are robust to missing data and preserve
the missing data (i.e., `np.nan`s) so downstream processing can account for
the missing values appropriately.
- Utilities for working with text (`misc.nlp_utils`,
`misc.incremental_count_vectorizer`)