BigML Python Bindings

BigML <https://bigml.com>_ makes machine learning easy by taking care of the details required to add data-driven decisions and predictive power to your company. Unlike other machine learning services, BigML creates beautiful predictive models <https://bigml.com/gallery/models>_ that can be easily understood and interacted with.

These BigML Python bindings allow you to interact with BigML.io <https://bigml.io/>, the API for BigML. You can use it to easily create, retrieve, list, update, and delete BigML resources (i.e., sources, datasets, models and, predictions). For additional information, see the full documentation for the Python bindings on Read the Docs <http://bigml.readthedocs.org>.

This module is licensed under the Apache License, Version 2.0 <http://www.apache.org/licenses/LICENSE-2.0.html>_.

Support

Please report problems and bugs to our BigML.io issue tracker <https://github.com/bigmlcom/io/issues>_.

Discussions about the different bindings take place in the general BigML mailing list <http://groups.google.com/group/bigml>. Or join us in our Campfire chatroom <https://bigmlinc.campfirenow.com/f20a0>.

Requirements

Only Python 3 versions are currently supported by these bindings. Support for Python 2.7.X ended in version 4.32.3.

The basic third-party dependencies are the requests <https://github.com/kennethreitz/requests>, unidecode <http://pypi.python.org/pypi/Unidecode/#downloads>, requests-toolbelt <https://pypi.python.org/pypi/requests-toolbelt>, bigml-chronos <https://pypi.org/project/bigml-chronos>, msgpack <https://pypi.org/project/msgpack>, numpy <http://www.numpy.org/> and scipy <http://www.scipy.org/>_ libraries. These libraries are automatically installed during the basic setup. Support for Google App Engine has been added as of version 3.0.0, using the urlfetch package instead of requests.

The bindings will also use simplejson if you happen to have it installed, but that is optional: we fall back to Python's built-in JSON libraries is simplejson is not found.

The bindings provide support to use the BigML platform to create, update, get and delete resources, but also to produce local predictions using the models created in BigML. Most of them will be actionable with the basic installation, but some additional dependencies are needed to use local Topic Models and Image Processing models. Please, refer to the Installation <#installation>_ section for details.

OS Requirements

The basic installation of the bindings is compatible and can be used
on Linux and Windows based Operating Systems.
However, the extra options that allow working with
image processing models (``[images]`` and ``[full]``) are only supported
and tested on Linux-based Operating Systems.
For image models, Windows OS is not recommended and cannot be supported out of
the box, because the specific compiler versions or dlls required are
unavailable in general.

Installation
------------

To install the basic latest stable release with
`pip <http://www.pip-installer.org/>`_, please use:

.. code-block:: bash

    $ pip install bigml

Support for local Topic Distributions (Topic Models' predictions)
and local predictions for datasets that include Images will only be
available as extras, because the libraries used for that are not
usually available in all Operative Systems. If you need to support those,
please check the `Installation Extras <#installation-extras>`_ section.

Installation Extras
-------------------

Local Topic Distributions support can be installed using:

.. code-block:: bash

    pip install bigml[topics]

Images local predictions support can be installed using:

.. code-block:: bash

    pip install bigml[images]

The full set of features can be installed using:

.. code-block:: bash

    pip install bigml[full]


WARNING: Mind that installing these extras can require some extra work, as
explained in the `Requirements <#requirements>`_ section.

You can also install the development version of the bindings directly
from the Git repository

.. code-block:: bash

    $ pip install -e git://github.com/bigmlcom/python.git#egg=bigml_python


Running the Tests
-----------------

The tests will be run using `pytest <https://docs.pytest.org/en/7.2.x/>`_.
You'll need to set up your authentication
via environment variables, as explained
in the authentication section. Also some of the tests need other environment
variables like ``BIGML_ORGANIZATION`` to test calls when used by Organization
members and ``BIGML_EXTERNAL_CONN_HOST``, ``BIGML_EXTERNAL_CONN_PORT``,
``BIGML_EXTERNAL_CONN_DB``, ``BIGML_EXTERNAL_CONN_USER``,
``BIGML_EXTERNAL_CONN_PWD`` and ``BIGML_EXTERNAL_CONN_SOURCE``
in order to test external data connectors.

With that in place, you can run the test suite simply by issuing

.. code-block:: bash

    $ pytest

Additionally, `Tox <http://tox.testrun.org/>`_ can be used to
automatically run the test suite in virtual environments for all
supported Python versions.  To install Tox:

.. code-block:: bash

    $ pip install tox

Then run the tests from the top-level project directory:

.. code-block:: bash

    $ tox

Importing the module
--------------------

To import the module:

.. code-block:: python

    import bigml.api

Alternatively you can just import the BigML class:

.. code-block:: python

    from bigml.api import BigML

Authentication
--------------

All the requests to BigML.io must be authenticated using your username
and `API key <https://bigml.com/account/apikey>`_ and are always
transmitted over HTTPS.

This module will look for your username and API key in the environment
variables ``BIGML_USERNAME`` and ``BIGML_API_KEY`` respectively.

Unix and MacOS
--------------

You can
add the following lines to your ``.bashrc`` or ``.bash_profile`` to set
those variables automatically when you log in:

.. code-block:: bash

    export BIGML_USERNAME=myusername
    export BIGML_API_KEY=ae579e7e53fb9abd646a6ff8aa99d4afe83ac291

refer to the next chapters to know how to do that in other operating systems.

With that environment set up, connecting to BigML is a breeze:

.. code-block:: python

    from bigml.api import BigML
    api = BigML()

Otherwise, you can initialize directly when instantiating the BigML
class as follows:

.. code-block:: python

    api = BigML('myusername', 'ae579e7e53fb9abd646a6ff8aa99d4afe83ac291')

These credentials will allow you to manage any resource in your user
environment.

In BigML a user can also work for an ``organization``.
In this case, the organization administrator should previously assign
permissions for the user to access one or several particular projects
in the organization.
Once permissions are granted, the user can work with resources in a project
according to his permission level by creating a special constructor for
each project. The connection constructor in this case
should include the ``project ID``:

.. code-block:: python

    api = BigML('myusername', 'ae579e7e53fb9abd646a6ff8aa99d4afe83ac291',
                project='project/53739b98d994972da7001d4a')

If the project used in a connection object
does not belong to an existing organization but is one of the
projects under the user's account, all the resources
created or updated with that connection will also be assigned to the
specified project.

When the resource to be managed is a ``project`` itself, the connection
needs to include the corresponding``organization ID``:

.. code-block:: python

    api = BigML('myusername', 'ae579e7e53fb9abd646a6ff8aa99d4afe83ac291',
                organization='organization/53739b98d994972da7025d4a')


Authentication on Windows
-------------------------

The credentials should be permanently stored in your system using

.. code-block:: bash

    setx BIGML_USERNAME myusername
    setx BIGML_API_KEY ae579e7e53fb9abd646a6ff8aa99d4afe83ac291

Note that ``setx`` will not change the environment variables of your actual
console, so you will need to open a new one to start using them.


Authentication on Jupyter Notebook
----------------------------------

You can set the environment variables using the ``%env`` command in your
cells:

.. code-block:: bash

    %env BIGML_USERNAME=myusername
    %env BIGML_API_KEY=ae579e7e53fb9abd646a6ff8aa99d4afe83ac291


Alternative domains
-------------------


The main public domain for the API service is ``bigml.io``, but there are some
alternative domains, either for Virtual Private Cloud setups or
the australian subdomain (``au.bigml.io``). You can change the remote
server domain
to the VPC particular one by either setting the ``BIGML_DOMAIN`` environment
variable to your VPC subdomain:

.. code-block:: bash

    export BIGML_DOMAIN=my_VPC.bigml.io

or setting it when instantiating your connection:

.. code-block:: python

    api = BigML(domain="my_VPC.bigml.io")

The corresponding SSL REST calls will be directed to your private domain
henceforth.

You can also set up your connection to use a particular PredictServer
only for predictions. In order to do so, you'll need to specify a ``Domain``
object, where you can set up the general domain name as well as the
particular prediction domain name.

.. code-block:: python

    from bigml.domain import Domain
    from bigml.api import BigML

    domain_info = Domain(prediction_domain="my_prediction_server.bigml.com",
                         prediction_protocol="http")

    api = BigML(domain=domain_info)

Finally, you can combine all the options and change both the general domain
server, and the prediction domain server.

.. code-block:: python

    from bigml.domain import Domain
    from bigml.api import BigML
    domain_info = Domain(domain="my_VPC.bigml.io",
                         prediction_domain="my_prediction_server.bigml.com",
                         prediction_protocol="https")

    api = BigML(domain=domain_info)

Some arguments for the Domain constructor are more unsual, but they can also
be used to set your special service endpoints:

- protocol (string) Protocol for the service
  (when different from HTTPS)
- verify (boolean) Sets on/off the SSL verification
- prediction_verify (boolean) Sets on/off the SSL verification
  for the prediction server (when different from the general
  SSL verification)

**Note** that the previously existing ``dev_mode`` flag:

.. code-block:: python

    api = BigML(dev_mode=True)

that caused the connection to work with the Sandbox ``Development Environment``
has been **deprecated** because this environment does not longer exist.
The existing resources that were previously
created in this environment have been moved
to a special project in the now unique ``Production Environment``, so this
flag is no longer needed to work with them.


Quick Start
-----------

Imagine that you want to use `this csv
file <https://static.bigml.com/csv/iris.csv>`_ containing the `Iris
flower dataset <http://en.wikipedia.org/wiki/Iris_flower_data_set>`_ to
predict the species of a flower whose ``petal length`` is ``2.45`` and
whose ``petal width`` is ``1.75``. A preview of the dataset is shown
below. It has 4 numeric fields: ``sepal length``, ``sepal width``,
``petal length``, ``petal width`` and a categorical field: ``species``.
By default, BigML considers the last field in the dataset as the
objective field (i.e., the field that you want to generate predictions
for).

::

    sepal length,sepal width,petal length,petal width,species
    5.1,3.5,1.4,0.2,Iris-setosa
    4.9,3.0,1.4,0.2,Iris-setosa
    4.7,3.2,1.3,0.2,Iris-setosa
    ...
    5.8,2.7,3.9,1.2,Iris-versicolor
    6.0,2.7,5.1,1.6,Iris-versicolor
    5.4,3.0,4.5,1.5,Iris-versicolor
    ...
    6.8,3.0,5.5,2.1,Iris-virginica
    5.7,2.5,5.0,2.0,Iris-virginica
    5.8,2.8,5.1,2.4,Iris-virginica

You can easily generate a prediction following these steps:

.. code-block:: python

    from bigml.api import BigML

    api = BigML()

    source = api.create_source('./data/iris.csv')
    dataset = api.create_dataset(source)
    model = api.create_model(dataset)
    prediction = api.create_prediction(model, \
        {"petal width": 1.75, "petal length": 2.45})

You can then print the prediction using the ``pprint`` method:

.. code-block:: python

    >>> api.pprint(prediction)
    species for {"petal width": 1.75, "petal length": 2.45} is Iris-setosa

Certainly, any of the resources created in BigML can be configured using
several arguments described in the `API documentation <https://bigml.com/api>`_.
Any of these configuration arguments can be added to the ``create`` method
as a dictionary in the last optional argument of the calls:

.. code-block:: python

    from bigml.api import BigML

    api = BigML()

    source_args = {"name": "my source",
         "source_parser": {"missing_tokens": ["NULL"]}}
    source = api.create_source('./data/iris.csv', source_args)
    dataset_args = {"name": "my dataset"}
    dataset = api.create_dataset(source, dataset_args)
    model_args = {"objective_field": "species"}
    model = api.create_model(dataset, model_args)
    prediction_args = {"name": "my prediction"}
    prediction = api.create_prediction(model, \
        {"petal width": 1.75, "petal length": 2.45},
        prediction_args)

The ``iris`` dataset has a small number of instances, and usually will be
instantly created, so the ``api.create_`` calls will probably return the
finished resources outright. As BigML's API is asynchronous,
in general you will need to ensure
that objects are finished before using them by using ``api.ok``.

.. code-block:: python

    from bigml.api import BigML

    api = BigML()

    source = api.create_source('./data/iris.csv')
    api.ok(source)
    dataset = api.create_dataset(source)
    api.ok(dataset)
    model = api.create_model(dataset)
    api.ok(model)
    prediction = api.create_prediction(model, \
        {"petal width": 1.75, "petal length": 2.45})

Note that the prediction
call is not followed by the ``api.ok`` method. Predictions are so quick to be
generated that, unlike the
rest of resouces, will be generated synchronously as a finished object.

The example assumes that your objective field (the one you want to predict)
is the last field in the dataset. If that's not he case, you can explicitly
set the name of this field in the creation call using the ``objective_field``
argument:


.. code-block:: python

    from bigml.api import BigML

    api = BigML()

    source = api.create_source('./data/iris.csv')
    api.ok(source)
    dataset = api.create_dataset(source)
    api.ok(dataset)
    model = api.create_model(dataset, {"objective_field": "species"})
    api.ok(model)
    prediction = api.create_prediction(model, \
        {'sepal length': 5, 'sepal width': 2.5})


You can also generate an evaluation for the model by using:

.. code-block:: python

    test_source = api.create_source('./data/test_iris.csv')
    api.ok(test_source)
    test_dataset = api.create_dataset(test_source)
    api.ok(test_dataset)
    evaluation = api.create_evaluation(model, test_dataset)
    api.ok(evaluation)

If you set the ``storage`` argument in the ``api`` instantiation:

.. code-block:: python

    api = BigML(storage='./storage')

all the generated, updated or retrieved resources will be automatically
saved to the chosen directory.

Alternatively, you can use the ``export`` method to explicitly
download the JSON information
that describes any of your resources in BigML to a particular file:

.. code-block:: python

    api.export('model/5acea49a08b07e14b9001068',
               filename="my_dir/my_model.json")

This example downloads the JSON for the model and stores it in
the ``my_dir/my_model.json`` file.

In the case of models that can be represented in a `PMML` syntax, the
export method can be used to produce the corresponding `PMML` file.

.. code-block:: python

    api.export('model/5acea49a08b07e14b9001068',
               filename="my_dir/my_model.pmml",
               pmml=True)

You can also retrieve the last resource with some previously given tag:

.. code-block:: python

     api.export_last("foo",
                     resource_type="ensemble",
                     filename="my_dir/my_ensemble.json")

which selects the last ensemble that has a ``foo`` tag. This mechanism can
be specially useful when retrieving retrained models that have been created
with a shared unique keyword as tag.

For a descriptive overview of the steps that you will usually need to
follow to model
your data