Fitgrid

Bug fixes

- Saving in `fit_lmers()` (TPU)

Other

- A number of clarifications in the docs (thanks Anna!)
- Added `fitgrid.__version__`

Features
- add Feather file reader
- add VIF calculator for OLS

Bug fixes
- add missing dependencies to `setup.py`
- improve `dir(grid)` behavior

Other
- massive refactor, move away from using constants to using parameters
- update docs to reflect refactor
- make the tutorial into a notebook
- add design matrix checker, which is a first step towards eliminating redundant model fitting
- update `conda` channel from `aportnoy` to `kutaslab`

Major refactor in this release.

Features

- Add DFBETAS calculator routine for lmer models

Other

- major refactor, model fitters now top level functions, so `fitgrid.lm` instead of `epochs.lm`, add `LMFitGrid` and `LmerFitGrid`, subclasses of `FitGrid`
- Travis environment updates
- expand docs, add page with installation instructions, add page with miscellaneous details of how things are done
- skip single thread context manager tests on Travis, see 86

Bug fixes

- `utils` package added to distribution, was missing because no `__init__.py`

Bug fixes

- set correct evaluation environment for `patsy` in `lm`

Installation

bash
conda install -c aportnoy fitgrid

or
bash
pip install -U fitgrid


Features

- new lmer fitting routine that allows for visual model comparison based on AIC
- grids can now be saved to files and loaded from files using `grid.save` and `fitgrid.load_grid`
- additional parameters now available in `epochs.lmer`: REML/ML, confidence interval types, etc.

Bug fixes

- exorbitant memory usage slightly mitigated by removing a number of attributes from lmer fit objects
- missing lmer correctness tests now added

Other

- `fitgrid` is now available from `conda`: run `conda install -c aportnoy fitgrid`
- Travis tests are now run using `conda`
- slight structure reorganization