Ranx

Added
- Added `make_comparable` to `run.py`. It makes a run comparable to a given qrels whether the run misses results for queries appearing in the qrels or have results for additional queries, which are removed.
- Added `make_comparable` parameter to `evaluate.py`.
- Added `make_comparable` parameter to `compare.py`.

Fixed
- Fixed a bug affecting `Tukey's HSD Test`: results from the test were not converted to proper dtypes from strings, causing the superscript reporting statistical significance differences in `report.py` to be wrong.

Changed
- Changed `tukey_hsd_test.py` to use `tukey_hsd` provided by `scipy`.
- `ranx` now requires `python>=3.8`.
- `ranx` now requires `scipy>=1.8`.

Removed
- Removed dependency from `statsmodels`.

Fixed
- Fixed a bug affecting `precision.py`, `recall.py`, and `f1.py`: `numba` does not raise ZeroDivisionError, added a control to make sure zero is returned when no retrieved results are provided for a specific query.
- Fixed a bug in `f1.py`: missing argument in function call.

[0.3.x] - 2022

Sorry, I have been lazy.

Fixed
- Fixed a bug in `posfuse.py`: `numba` does not raise out of bounds error in some specific cases, added a control to make sure ranking positions with no associated probability get 0 probability.
- Fixed a bug in `baysfuse.py`: as it uses log odds, which can be negative, `comb_sum` cannot be used. Added a `odds_sum` function to combine the log odds.

Fixed
- Fixed a bug in `data_structures/common.py:sort_dict_by_value` that was preventing result list sorting to be consistent for documents with the same score.
- Fixed a bug causing original runs to be modified by fusion methods.

Fixed
- Fixed a bug in `max_norm.py`, `min_max_norm.py`, and `sum_norm.py`: `min` and `max` functions called on empty lists do not raise error in `Numba` causing downstream miscalculations.