.. DO NOT EDIT.
.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY.
.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE:
.. "getting_started/auto_examples/plot_VHCT_Landmine.py"
.. LINE NUMBERS ARE GIVEN BELOW.

.. only:: html

    .. note::
        :class: sphx-glr-download-link-note

        :ref:`Go to the end <sphx_glr_download_getting_started_auto_examples_plot_VHCT_Landmine.py>`
        to download the full example code

.. rst-class:: sphx-glr-example-title

.. _sphx_glr_getting_started_auto_examples_plot_VHCT_Landmine.py:


Real-life Data
============================
In this example, we run the VHCT algorithm to tune hyperparameters for Support Vector Machine (SVM)

.. GENERATED FROM PYTHON SOURCE LINES 7-19

.. code-block:: default


    from PyXAB.algos.VHCT import VHCT
    from PyXAB.partition.BinaryPartition import BinaryPartition
    from PyXAB.utils.plot import plot_regret

    # Useful functions and classes for the learning process
    from sklearn import svm
    from sklearn.metrics import roc_auc_score
    import numpy as np
    import pickle









.. GENERATED FROM PYTHON SOURCE LINES 20-22

We first define the objective as maximizing the ROC_AUC score on the testing dataset after training the SVM using
the hyperparameters with the training dataset.

.. GENERATED FROM PYTHON SOURCE LINES 22-44

.. code-block:: default


    class obj_func_landmine():

        def __init__(self, X_train, Y_train, X_test, Y_test):

            self.X_train = X_train              # Training X
            self.Y_train = Y_train              # Training Y
            self.X_test = X_test                # Testing X
            self.Y_test = Y_test                # Testing Y
            self.fmax = 1

        def f(self, point):
            C = point[0]                        # First parameter
            gam = point[1]                      # Second parameter

            clf = svm.SVC(kernel="rbf", C=C, gamma=gam, probability=True)       # The machine learning model is SVM
            clf.fit(self.X_train, self.Y_train)                                 # Fit the model using training data
            pred = clf.predict_proba(self.X_test)                               # Make prediction on the testing
            score = roc_auc_score(self.Y_test, pred[:, 1])                      # The reward is the ROC_AUC score

            return score








.. GENERATED FROM PYTHON SOURCE LINES 45-47

Input the data and then split the data into the training dataset and the testing dataset. Then define the objective
function using the datasets.

.. GENERATED FROM PYTHON SOURCE LINES 47-59

.. code-block:: default


    landmine_data = pickle.load(open("../../PyXAB/landmine/landmine_formated_data.pkl", "rb"))
    all_X_train, all_Y_train, all_X_test, all_Y_test = landmine_data["all_X_train"], landmine_data["all_Y_train"], \
                                                           landmine_data["all_X_test"], landmine_data["all_Y_test"]

    X_train = all_X_train[0]
    Y_train = np.squeeze(all_Y_train[0])
    X_test = all_X_test[0]
    Y_test = np.squeeze(all_Y_test[0])

    target = obj_func_landmine(X_train=X_train, X_test=X_test, Y_train=Y_train, Y_test=Y_test)








.. GENERATED FROM PYTHON SOURCE LINES 60-61

Define the number of rounds, the domain, the partition, and the algorithm for the learning process

.. GENERATED FROM PYTHON SOURCE LINES 61-72

.. code-block:: default

    T = 500
    domain = [[1e-4, 10.0], [1e-2, 10.0]]
    partition = BinaryPartition
    algo = VHCT(domain=domain, rho=0.5, partition=partition)

    # To plot the regret, we can initialize the cumulative regret and the cumulative regret list

    cumulative_regret = 0
    cumulative_regret_list = []









.. GENERATED FROM PYTHON SOURCE LINES 73-75

In each iteration of the learning process, the algorithm calls the ``pull(t)`` function to obtain a point, and then
the reward for the point is returned to the algorithm by calling ``receive_reward(t, reward)``.

.. GENERATED FROM PYTHON SOURCE LINES 75-88

.. code-block:: default


    for t in range(1, T+1):

        point = algo.pull(t)
        reward = target.f(point)
        algo.receive_reward(t, reward)
        inst_regret = target.fmax - target.f(point)
        cumulative_regret += inst_regret
        cumulative_regret_list.append(cumulative_regret)


    # plot the regret
    plot_regret(np.array(cumulative_regret_list), name='VHCT')



.. image-sg:: /getting_started/auto_examples/images/sphx_glr_plot_VHCT_Landmine_001.png
   :alt: plot VHCT Landmine
   :srcset: /getting_started/auto_examples/images/sphx_glr_plot_VHCT_Landmine_001.png
   :class: sphx-glr-single-img






.. rst-class:: sphx-glr-timing

   **Total running time of the script:** (0 minutes 13.222 seconds)


.. _sphx_glr_download_getting_started_auto_examples_plot_VHCT_Landmine.py:

.. only:: html

  .. container:: sphx-glr-footer sphx-glr-footer-example




    .. container:: sphx-glr-download sphx-glr-download-python

      :download:`Download Python source code: plot_VHCT_Landmine.py <plot_VHCT_Landmine.py>`

    .. container:: sphx-glr-download sphx-glr-download-jupyter

      :download:`Download Jupyter notebook: plot_VHCT_Landmine.ipynb <plot_VHCT_Landmine.ipynb>`


.. only:: html

 .. rst-class:: sphx-glr-signature

    `Gallery generated by Sphinx-Gallery <https://sphinx-gallery.github.io>`_