.. _user_guide.machine_learning.introduction:

=================================
Introduction to Machine Learning
=================================

One of the last stages of the data science life cycle is the Data Modeling. Machine learning algorithms are a set of statistical techniques that build mathematical models from training data. These algorithms come in two types:
 - **Supervised:** these algorithms are used when we want to predict a response column.
 - **Unsupervised:** these algorithms are used when we want to detect anomalies or when we want to segment the data. No response column is needed.

Supervised Learning
--------------------

Supervised Learning techniques map an input to an output based on some example dataset. This type of learning consists of two main types:
 - **Regression:** The Response is numerical (``Linear Regression``, ``SVM Regression``, ``RF Regression``...).
 - **Classification:** The Response is categorical (``Gradient Boosting``, ``Naive Bayes``, ``Logistic Regression``...).

For example, predicting the total charges of a Telco customer using their tenure would be a type of regression. The following code is drawing a linear regression using the ``TotalCharges`` as a function of the ``tenure`` in the `telco churn dataset <https://github.com/vast-data/VAST-Orbit/tree/master/docs/source/notebooks/data_exploration/correlations/data>`__.

.. code-block:: python

    import vastorbit as vo

    churn = vo.read_csv("churn.csv")

    from vastorbit.machine_learning.vast import LinearRegression

    model = LinearRegression()
    model.fit(churn, ["tenure"], "TotalCharges")
    model.plot()

.. ipython:: python
    :suppress:
    :okwarning:

    import vastorbit as vo
    vo.set_option("plotting_lib", "plotly")

    churn = vo.read_csv("/Users/badr.ouali/Documents/VastOrbit-master/docs/source/_static/website/user_guides/data_exploration/churn.csv")

    from vastorbit.machine_learning.vast import LinearRegression

    model = LinearRegression()
    churn = churn.dropna(columns = ["tenure", "TotalCharges"])
    model.fit(churn, ["tenure"], "TotalCharges")
    fig = model.plot()
    fig.write_html("/Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_1.html")

.. raw:: html
    :file: /Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_1.html

In contrast, when we have to predict a categorical column, we're dealing with classification.

In the following example, we use a ``Linear Support Vector Classification`` (SVC) to predict the species of a flower based on its petal and sepal lengths.

.. code-block::

    from vastorbit.datasets import load_iris

    iris = load_iris()
    iris.one_hot_encode()

.. ipython:: python
    :suppress:
    :okwarning:

    from vastorbit.datasets import load_iris

    iris = load_iris()
    res = iris.one_hot_encode()
    html_file = open("/Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_table_introduction_1.html", "w")
    html_file.write(res._repr_html_())
    html_file.close()

.. raw:: html
    :file: /Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_table_introduction_1.html

.. code-block:: python

    from vastorbit.machine_learning.vast import LinearSVC

    model = LinearSVC(max_iter = 1000)
    model.fit(iris, ["PetalLengthCm", "SepalLengthCm"], "Species_Iris-setosa")
    model.plot()

.. ipython:: python
    :suppress:
    :okwarning:

    from vastorbit.machine_learning.vast import LinearSVC

    model = LinearSVC(max_iter = 1000)
    model.fit(iris, ["PetalLengthCm", "SepalLengthCm"], "Species_Iris-setosa")
    fig = model.plot()
    fig.write_html("/Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_2.html")

.. raw:: html
    :file: /Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_2.html

When we have more than two categories, we use the expression ``Multiclass Classification`` instead of ``Classification``.

Unsupervised Learning
----------------------

These algorithms are to used to segment the data (:py:mod:`~vastorbit.machine_learning.vast.cluster.KMeans`, :py:mod:`~vastorbit.machine_learning.vast.cluster.DBSCAN`, etc.) or to detect anomalies (:py:mod:`~vastorbit.machine_learning.vast.LocalOutlierFactor`, ``Z-Score`` Techniques...). In particular, they're useful for finding patterns in data without labels. For example, let's use a :py:mod:`~vastorbit.machine_learning.vast.cluster.KMeans` algorithm to create different clusters on the Iris dataset. Each cluster will represent a flower's species.

.. code-block:: python

    from vastorbit.machine_learning.vast import KMeans

    model = KMeans(n_clusters = 3)
    model.fit(iris, ["PetalLengthCm", "SepalLengthCm"])
    model.plot()

.. ipython:: python
    :suppress:
    :okwarning:

    from vastorbit.machine_learning.vast import KMeans

    model = KMeans(n_clusters = 3)
    model.fit(iris, ["PetalLengthCm", "SepalLengthCm"])
    fig = model.plot()
    fig.write_html("/Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_3.html")

.. raw:: html
    :file: /Users/badr.ouali/Documents/VastOrbit-master/docs/figures/ug_ml_plot_introduction_3.html

In this section, we went over a few of the many ML algorithms available in vastorbit.

In the next lesson, we'll go over :ref:`user_guide.machine_learning.time_series`

.. ipython:: python
   :suppress:

   from vastorbit._utils._sql._sys import purge_memory
   purge_memory()