0.8.0 change log

[Darel13712]

New Features

• optimize method is refactored. Previously optimization could return suboptimal point because init arguments of a model were not tested. Now they are added as a starting trial if arguments are contained inside search space. param_drid parameter is renamed to param_borders and also there is a new new_study parameter that allows you to choose whether to continue adding trials to old study or to create a new study.
• KNN, PopRec and RandomRec now support arbitrary relevance values with a new use_relevance argument. This can be used to take weight of different types of actions into account.
• new module time provides functions to create and apply time-dependent weights to log, as described in Exponential time smoothing #39

Other Changes

• new tests are significantly faster because they don't call optuna each time
• new docs page about optimization
• comparison table on ml1m is added to choose a recommender page
• recommonmark is removed from dependencies as it is no longer maintened. myst-parser is used instead to support md files, sphinx-markdown-tables is also removed because myst does the job.
• bumped lower version of python from 3.6.2 to 3.7 because of conflicting dependencies in lightautoml and numpy. numpy > 1.20 requires python >= 3.7 so we now do too