.. _save_models:

Saving external models
======================

In some cases, there is a need to store an external model in WML project or WML deployment space.
Depending on model type, it has to be saved or archived with correct extension.

Following types and file extensions are supported:

- :ref:`tensorflow / keras (.tar.gz) <tf>`
- :ref:`pmml (.xml) <pmml>`
- :ref:`scikit-learn (.tar.gz) <sklearn>`
- :ref:`spss (.str) <spss>`
- :ref:`spark (.tar.gz) <spark>`
- :ref:`xgboost (.tar.gz) <xgboost>`

.. _tf:

Tensorflow / Keras
------------------

This section shows how to save and archive a tensorflow (keras) model.


.. code-block::

    import tarfile
    import os


    # Prepare your tf / keras model
    model = ... # (Sequential, Functional Model, or Model subclass)

    # Set names
    model_name = 'tf_model'
    folder_name = 'models'

    # Prepare directories
    file_name = model_name + '.h5'
    full_path = os.path.join(os.getcwd(), folder_name)
    os.makedirs(full_path, exist_ok=True)
    model_path = os.path.join(full_path, file_name)

    # Save model as '.h5' file
    model.save(model_path)

    # Archive model
    archive_name = model_name + '.tar.gz'
    archive_path = os.path.join(full_path, archive_name)
    with tarfile.open(archive_path, 'w:gz') as tar:
        tar.add(model_path, arcname=os.path.basename(model_path))

    # Path that can be used to store model in project or space
    path_to_store_model = archive_path

.. _pmml:

PMML
----


For ``PMML`` models, it is required to prepare a model as file in :code:`.xml` format,
that can be directly stored in project or space.


.. code-block::

    sample_model_path = "./models/pmml_model.xml"
    path_to_store_model = sample_model_path

.. _sklearn:

Scikit-learn
------------

This section shows how to save and archive a scikit-learn model.


.. code-block::

    import joblib
    import os
    import tarfile


    # Prepare your scikit-learn model
    model = ...

    # Set names
    model_name = 'sklearn_model'
    folder_name = 'models'

    # Prepare directories
    file_name = model_name + '.pkl'
    full_path = os.path.join(os.getcwd(), folder_name)
    os.makedirs(full_path, exist_ok=True)
    model_path = os.path.join(full_path, file_name)

    # Save model as '.pkl' file
    with open(model_path, 'wb') as file:
        joblib.dump(model, file)

    # Archive model
    archive_name = model_name + '.tar.gz'
    archive_path = os.path.join(full_path, archive_name)
    with tarfile.open(archive_path, 'w:gz') as tar:
        tar.add(model_path, arcname=os.path.basename(model_path))

    # Path that can be used to store model in project or space
    path_to_store_model = archive_path

.. _spss:

SPSS
----

For ``SPSS`` models, it is required to prepare a model as file in :code:`.str` format,
that can be directly stored in project or space.


.. code-block::

    sample_model_path = "./models/spss_model.str"
    path_to_store_model = sample_model_path

.. _spark:

Spark
-----

This section shows how to save and archive a spark model.

.. code-block::


    import os
    import tarfile


    # Prepare your spark model
    model = ...

    # Set names
    model_name = "spark_model"
    folder_name = "models"

    # Prepare directories
    full_path = os.path.join(os.getcwd(), folder_name)
    os.makedirs(full_path, exist_ok=True)
    model_path = os.path.join(full_path, model_name)

    # Save model
    model.save(model_path)

    # Archive model
    archive_name = model_name + ".tar.gz"
    archive_path = os.path.join(full_path, archive_name)
    with tarfile.open(archive_path, 'w:gz') as tar:
        tar.add(model_path, arcname=os.path.basename(model_path))

    # Path that can be used to store model in project or space
    path_to_store_model = archive_path

.. _xgboost:

XGBoost
-------

This section shows how to save and archive an XGBoost model.

.. code-block::


    import os
    import joblib
    import tarfile


    # Prepare your xgboost model
    model = ...

    # Set names
    model_name = "xgboost_model"
    folder_name = "models"

    # Prepare directories
    file_name = model_name + '.pkl'
    full_path = os.path.join(os.getcwd(), folder_name)
    os.makedirs(full_path, exist_ok=True)
    model_path = os.path.join(full_path, file_name)

    # Save model as '.pkl' file
    with open(model_path, 'wb') as file:
        joblib.dump(model, file)

    # Archive model
    archive_name = model_name + '.tar.gz'
    archive_path = os.path.join(full_path, archive_name)
    with tarfile.open(archive_path, 'w:gz') as tar:
        tar.add(model_path, arcname=os.path.basename(model_path))

    # Path that can be used to store model in project or space
    path_to_store_model = archive_path