.. _`tasks`:

Supported Tasks
====================================================================

Each task has an associated `Dataset Type`, `Query Format` and `Prediction Format`. 

A task's `Dataset Type` specifies the format of the data sources ``train_dataset_uri`` and  ``test_dataset_uri`` point to. 
Datasets are prepared by `Application Developers` when they create `Train Jobs` 
and received by `Model Developers` when they define :meth:`rafiki.model.BaseModel.train` and :meth:`rafiki.model.BaseModel.evaluate`.

A task's `Query Format` specifies the format of queries when they are passed to models. 
Queries are generated by `Application Users` when they send queries to `Inference Jobs` 
and received by `Model Developers` when they define :meth:`rafiki.model.BaseModel.predict`.

A task's `Prediction Format` specifies the format of predictions made by models. 
Predictions are generated by `Model Developers` when they define :meth:`rafiki.model.BaseModel.predict`
and received by `App Users` as predictions to their queries sent to `Inference Jobs` 

.. seealso:: :ref:`making-predictions`


IMAGE_CLASSIFICATION
--------------------------------------------------------------------

Dataset Type
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

:ref:`dataset-type:IMAGE_FILES`

The train & test dataset's images should be have the same dimensions ``W x H``.

Query Format 
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

A ``W x H`` 2D list representing the grayscale version of the query image. 

Prediction Format 
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

A size-``k`` list of floats, representing the probabilities of each class from ``0`` to ``k - 1``.

POS_TAGGING
--------------------------------------------------------------------

Dataset Type
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

:ref:`dataset-type:CORPUS`, such that:

- Sentences are delimited by  ``\n`` tokens.
- There is only 1 tag column of ``tag`` corresponding to the POS tag of the token as an integer from ``0`` to ``k - 1``.


Query Format 
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

A list of strings representing a sentence as a list of tokens in that sentence.

Prediction Format 
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

A list of integers representing the list of predicted tag for each token, in sequence, for the sentence.