Gismo

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

* Refresh dependencies, compatibilities, and such.

* Gismo is tested up to Python 3.10.

* Patch sklearn change of API
(`ngram_range` must be a tuple, `get_feature_names` has been renamed `get_feature_names_out`)

* Updates MixInIO logic: you now save with the `dump` method and load with the `load` **class** method.

* Package management now uses Github actions.

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

Minor patch

* Signature of the Gismo rank method changed to allow to enter directly a query vector instead of a string query
(useful if one wants to craft a custom query vector).
* Original source of the Reuters 50/50 dataset was discontinued; changed to an alternate source.
* Fix change in spacy API

------------------
Minor update.

* DBLP API modified to you can specify the set of fields you want to retrieve.
* Minor update in doctests.
* Python 3.9 compatibility added.

------------------
0.4 is a big update. Lot of things added, lot of things changed.

* New API for Gismo runtime parameters (see new parameters module for details). Short version:
* ``gismo = Gismo(corpus, embedding, alpha=0.85)``: create a gismo with damping factor set to 0.85 instead of default value.
* ``gismo.parameters.alpha = 0.85``: set the damping factor of the gismo to 0.85.
* ``gismo.rank(query, alpha=0.85)``: makes a query with damping factor temporarily set to 0.85.
* Landmarks! Half Corpus, half Gismo, the Landmarks class can simplify many analysis tasks.
* Landmarks are (small) corpus where each entry is augmented with the computation of an associated gismo query;
* Landmarks can be used to refine the analysis around a part of your data;
* They can be used as soft and fast classifiers.
* Landmarks' runtime parameters follow the same approach than for Gismo instances (cf above).
* See the dedicated tutorial to learn more!
* Documentation summer cleaning.
* ``query_distortion`` parameter (reshape subspace for clustering) is renamed ``distortion`` and is now a float instead of a bool (e.g. you can apply distortion in a non-binary way).
* Full refactoring of get_*** and post_*** methods and objects.
* The good news is that they are now more natural, self-describing, and unified.
* The bad news is that there is no backward-compatibility with previous Gismo versions. Hopefully this refactoring
will last for some time!
* Gismo logo added!

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

* New dataset: Reuters C50
* New module: sentencizer

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

* dblp module: url2source function added to directly load a small dblp source in memory instead of using a FileSource approach.
* Possibility to disable query distortion in gismo.
* XGismo class to cross analyze embeddings.
* Tutorials updated