Visualization
=================

In this section, we aim to provide functionalities to have visual representation of embeddings.
In this first version, we provide a wrapper class for the :class:`sklearn.manifold.TSNE`, reducing, by default, the dimensionality of embeddings to two dimensions.


To use this model, we assume we already have some embedding vectors and some labels for each of them.
For this example, let's create random vectors:


.. testcode::

   from numpy import array, random   
   
   embeddings = random.rand(100, 100)
   names = [f"name_{i}" for i in range(100)]
   classes = [1,2,3]
   labels = [random.choice(classes) for _ in range(100)]

   name_to_embedding = dict(zip(names, embeddings))
   name_to_label = dict(zip(names, labels))


.. note::

   Notice that the previous code is generating synthetic data. In real application of mOWL, the ``embeddings`` would come from a trained model and the ``labels`` can come from either the ontology or any other external source. Furthermore, the ``names`` would be ontology entities (classes, individuals, ...).
   
.. testcode::

   import mowl
   mowl.init_jvm("10g")

   from mowl.visualization import TSNE


   tsne = TSNE(name_to_embedding, name_to_label)
   tsne.generate_points(250, workers=4)

.. code:: python
	       
   tsne.show()


.. plot:: pyplots/tsne.py