Tsfresh

- Breaking Change
- Fix the sorting of the parameters in the feature names (656)
The feature names consist of a sorted list of all parameters now.
That used to be true for all non-combiner features, and is now also true for combiner features.
If you relied on the actual feature name, this is a breaking change.
- Change the id after the rolling (668)
Now, the old id of your data is still kept. Additionally, we improved the way
dataframes without a time column are rolled and how the new sub-time series
are named.
Also, the documentation was improved a lot.
- Added Features
- Added variation coefficient (654)
- Added the datetimeindex explanation from the notebook to the docs (661)
- Optimize RelevantFeatureAugmenter to avoid re-extraction (669)
- Added a function `add_sub_time_series_index` (666)
- Added Dockerfile
- Speed optimizations and speed testing script (681)
- Bugfixes
- Increase the extracted `ar` coefficients to the full parameter range. (662)
- Documentation fixes (663, 664, 665)
- Rewrote the `sample_entropy` feature calculator (681)
It is now faster and (hopefully) more correct.
But your results will change!

Changelog and documentation fixes

- Added Features
- Add count_above and count_below feature (632)
- Add convenience bindings for dask dataframes and pyspark dataframes (651)
- Bugfixes
- Fix documentation build and feature table in sphinx (637, 631, 627)
- Add scripts to API documentation
- Skip dask test for older python versions (649)
- Add missing distributor keyword (648)
- Fix tuple input for cwt (645)

- Breaking Change
- Replace Benjamini-Hochberg implementation with statsmodels implementation (570)
- Refactoring and Documentation
- travis.yml (605)
- gitignore (608)
- Fix docstring of c3 (590)
- Feature/pep8 (607)
- Added Features
- Improve test coverage (609)
- Add "autolag" parameter to augmented_dickey_fuller() (612)
- Bugfixes
- Feature/pep8 (607)
- Fix filtering on warnings with multiprocessing on Windows (610)
- Remove outdated logging config (621)
- Replace Benjamini-Hochberg implementation with statsmodels implementation (570)
- Fix the kernel and the naming of a notebook (626)

- Drop python 2.7 support (568)
- Fixed bugs
- Fix cache in friedrich_coefficients and agg_linear_trend (593)
- Added a check for wrong column names and a test for this check (586)
- Make sure to not install the tests folder (599)
- Make sure there is at least a single column which we can use for data (589)
- Avoid division by zero in energy_ratio_by_chunks (588)
- Ensure that get_moment() uses float computations (584)
- Preserve index when column_value and column_kind not provided (576)
- Add set_property("input", "pd.Series") when needed (582)
- Fix off-by-one error in longest strike features (fixes 577) (578)
- Add `set_property` import (572)
- Fix typo (571)
- Fix indexing of melted normalized input (563)
- Fix travis (569)
- Remove warnings (583)
- Update to newest python version (594)
- Optimizations
- Early return from change_quantiles if ql >= qh (591)
- Optimize mean_second_derivative_central (587)
- Improve performance with Numpy's sum function (567)
- Optimize mean_change (fixes issue 542) and correct documentation (574)

- fixed bugs
- wrong calculation of friedrich coefficients
- feature selection selected too many features
- an ignored max_timeshift parameter in roll_time_series
- add deprecation warning for python 2
- added support for index based features
- new feature calculator
- linear_trend_timewise
- enable the RelevantFeatureAugmenter to be used in cross validated pipelines
- increased scipy dependency to 1.2.0