Gdsctools

----------------------------
* BUG Fixes

- Fix pending issue related to 153
ic_input.csv", "test_gf_input.csv" for the case where some media factor
are missing. This is now handle properly.
- Fix 151 : large integer are not cast properly with consequence that
indices are strings, not integers leading to further issues in the
HTML pages.

- NEWS:

- anova_one_drug_one_feature_custom allows to perform any regression using
formula like in R. This is not for production but should be useful to
perform custom analysis
- Include the MEDIA factor boxplot in the library and reports

-------------------------------------

* BUG Fixes

- Fix issue 156 (GDSC failures in some cases). This was due to special
MOBEM input files for which no tests are performed. In such cases, some
codes were failing in ANOVAReport and ANOVAResults, which have been fixed
in this release.
- Fix buggy volcano plot (no plot if no data); if FDR threshold below
minimum value, set fdr_threshold to minimum value so that it scaled the
plots properly. Now, users can add as many lines as desired using
settings.additional_fdr. pvalues found to be NAN are set to 0 to prevent
plotting issues.
- anova module: regression in the case Y ~ C(TISSUE) + C(MSI) + feature,
the tissue sum of squares was using N-1 tissues (one missing).
- anova module: regression in the case Y ~ C(TISSUE) + C(MEDIA) + C(MSI) + feature,
the media sum of squares was not normalised properly.

* CHANGES:

- elastic_net module renamed into regression
- ANOVASetting prints keys in alphabetical order (instead of randomly)
- ANOVASetting: regression_formula is added; other regression_XX settings
are not removed but not used anymore.
- GenomicFeature reader: if a tissue is empty, it is replaced by UNDEFINED.
- standalone: the drug option must be an integer. THis is now caught are the
option level, not later in the code.
- anova module: remove code related to elastic net, ridge, and lasso. This
won't be used in production with ANOVA. EN, Ridge and Lasso are used
in the regression module and will be part of an independent type of
analysis. See NEWS

* NEWS:

- Add Ridge + Lasso + LassoLars classes in addition to ElasticNet regression
method into the regression module.
- Add more features in regression module (boxplot, dendogram)
- New regression notebook in the notebooks directory
- anova module: We can now use any combo of regression formula using
statsmodels. This is slower but one can do use any formula accepted
by statsmodels. The previous faster code is still used for the standard
analysis.

---------------------------------

* NEWS:

- ElasticNet: new method elastic_all()
- plot_elastic_weight in the gallery

* CHANGES:

- ElasticNet plot_weights is now split into plot_weights
and plot_importance.

* BUGS:

- Fixes missing files in the pypi distributino (MANIFEST changed)

* CHANGES:

- **DrugDecode**: In brief, the DRUG ID in the IC50 input file and the
DrugDecode files should be integers. Some old data sets use the
following convention to refer to a drug Drug_<ID>_IC50 and so DrugDecode
was using the same convention. However, we now convert this type of
identifier into integers. This is done internally for the IC50 file,
however, was not done inside the DrugDecoder file. This is now effective.
- HTML reports when using the GDSC class:
- Company names now appear systematically in the top of the company data
packages.
- Drug Names were missing and do now appear in top of the relevant HTML
pages.
- Boxplots: If a DrugDecode file is provided Boxplots show the DRUG ID
and the real drug name in the matplotlib and JS boxplots

-------------------------------

* CHANGES:

- Reader class simplification and improvments: files can now be compressed
using gzip but also xz, zip and bz2 formats. The NA can be encoded as NA
or NaN strings. Spaces are interpreted as NA.
- Sort DrugDecode's dataframe columns
- Updated all documentation

* BUG:

- Fix scaling of the data with newest version of scikit-learn
- fix typo in the setup.py file. Passed travis + all tests before main
release.