How it Works
============

This package extracts models from `scikit-learn <https://scikit-learn.org>`_ and converts them into classes to be used with `msdss-models-api <https://rrwen.github.io/msdss-models-api>`_.

.. digraph:: methods

   compound=true;
   rankdir=TB;
   graph [nodesep="0.5", ranksep="0.65"];

   sklearn[label="scikit-learn" style=filled];
   modelsapi[label="msdss-models-api" style=filled];

   basemodel[label="msdss_models_api.models.Model" shape=rect];
   models[label="msdss_models_sklearn.core.models" shape=rect];

   getsklearnmodels[label="msdss_models_sklearn.core.get_sklearn_models" shape=rect style=rounded];
   createinputmethod[label="create_input_method" shape=rect style=rounded];
   createoutputmethod[label="create_output_method" shape=rect style=rounded];
   createupdatemethod[label="create_update_method" shape=rect style=rounded];

   subgraph cluster0 {
      label=< <B>msdss_models_sklearn.tools</B> >;
      createinputmethod;
      createoutputmethod;
      createupdatemethod;
   }

   sklearn -> getsklearnmodels -> basemodel;
   basemodel -> createoutputmethod[lhead=cluster0 ltail=cluster0];
   createoutputmethod -> models[lhead=cluster0 ltail=cluster0];
   models -> modelsapi;

Manually Selected Modules
-------------------------

First a list of modules with models are manually selected based on `scikit-learn documentation <https://scikit-learn.org/stable/modules/classes.html>`_:

.. jupyter-execute::
   :hide-code:

   from msdss_models_sklearn.defaults import DEFAULT_MODULES
   from pprint import pprint

   pprint(DEFAULT_MODULES)

Class Name Filters
------------------

For each module, classes are filtered to meet the following requirements:

* First letter of the name is capitalized and not an underscore ``_``
* Class is under the module and not imported from elsewhere

Model Conversion
----------------

The filtered classes are considered scikit-learn models and are converted to ``msdss-models-api`` compatible models by:

* Creating a dynamic class, extending from :class:`msdss_models_api:msdss_models_api.models.Model`
* Adding compatible methods to the dynamic class:
   * ``input`` using :func:`msdss_models_sklearn.tools.create_input_method`
   * ``output`` using :func:`msdss_models_sklearn.tools.create_output_method`
   * ``update`` using :func:`msdss_models_sklearn.tools.create_update_method`

Model Results
-------------

The result is a dictionary of ``msdss-models-api`` compatible models ready to be used with the API:

.. jupyter-execute::
   :hide-code:

   from msdss_models_sklearn import models as sklearn_models
   from pprint import pprint

   pprint(sklearn_models)

See :func:`msdss_models_sklearn.core.get_sklearn_models` for more details.