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

.. only:: html

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

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

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

.. _sphx_glr_auto_examples_plot_linear_quickstart.py:


Linear Model for Multi-label Classification.
============================================

This guide will take you through how LibMultiLabel can
be used to train a linear classifier in python scripts.

For this guide, we will only need the linear module:

.. GENERATED FROM PYTHON SOURCE LINES 10-13

.. code-block:: Python


    import libmultilabel.linear as linear


.. GENERATED FROM PYTHON SOURCE LINES 14-15

To start, we need to read and preprocess the input data:

.. GENERATED FROM PYTHON SOURCE LINES 15-25

.. code-block:: Python


    datasets = linear.load_dataset("txt", "data/rcv1/train.txt", "data/rcv1/test.txt")
    preprocessor = linear.Preprocessor()
    preprocessor.fit(datasets)
    datasets = preprocessor.transform(datasets)

    # Alternatively, you can fit and transform the dataset at the same time:
    # datasets = preprocessor.fit_transform(datasets)



.. GENERATED FROM PYTHON SOURCE LINES 26-48

.. note::
  The ``fit`` method calculates the parameters (such as label mapping) that will be used for later transformation
  from the given dataset.

  The ``transform`` method applies the learned parameters to the given dataset and returns a new transformed dataset.

The preprocessor handles many issues such as: mapping
the labels into indices and transforming textual data to
numerical data. The loaded dataset has the structure::

    {
        'train': {
            'x': # training features
            'y': # training labels
        },
        'test': {
            'x': # test features
            'y': # test labels
        },
    }

Next we train the model:

.. GENERATED FROM PYTHON SOURCE LINES 48-51

.. code-block:: Python


    model = linear.train_1vsrest(datasets["train"]["y"], datasets["train"]["x"], "")


.. GENERATED FROM PYTHON SOURCE LINES 52-57

The third argument is the options string for
`LibLinear <https://www.csie.ntu.edu.tw/~cjlin/liblinear/>`__.
We may leave it as the default for now.

Once we have the model, we may predict with it:

.. GENERATED FROM PYTHON SOURCE LINES 57-60

.. code-block:: Python


    preds = linear.predict_values(model, datasets["test"]["x"])


.. GENERATED FROM PYTHON SOURCE LINES 61-65

``preds`` holds the decision values, i.e. the raw values
outputted by the model. To transform it into predictions,
you can apply the API ``get_positive_labels`` to get predicted labels and their corresponding scores
by using ``label_mapping`` in ``preprocessor`` and ``preds`` from the last step.

.. GENERATED FROM PYTHON SOURCE LINES 65-68

.. code-block:: Python


    pred_labels, pred_scores = linear.get_positive_labels(preds, preprocessor.label_mapping)


.. GENERATED FROM PYTHON SOURCE LINES 69-71

We now have the labels (``pred_labels``) and scores (``pred_scores``).
You can use the following code to save the prediction to a list.

.. GENERATED FROM PYTHON SOURCE LINES 71-76

.. code-block:: Python


    prediction = []
    for label, score in zip(pred_labels, pred_scores):
        prediction.append([f"{i}:{s:.4}" for i, s in zip(label, score)])


.. GENERATED FROM PYTHON SOURCE LINES 77-87

The first instance looks like:

  >>> print(prediction[0])
  ...
      ['GCAT:1.345', 'GSPO:1.519']

To see how well we performed, we may want to check various
metrics with the test set.
Since the dataset we loaded are stored as ``scipy.sparse.csr_matrix``,
we will first transform the dataset to ``np.array``.

.. GENERATED FROM PYTHON SOURCE LINES 87-90

.. code-block:: Python


    target = datasets["test"]["y"].toarray()


.. GENERATED FROM PYTHON SOURCE LINES 91-92

Then we will compute the metrics with ``compute_metrics``.

.. GENERATED FROM PYTHON SOURCE LINES 92-101

.. code-block:: Python


    metrics = linear.compute_metrics(
        preds,
        target,
        monitor_metrics=["Macro-F1", "Micro-F1", "P@1", "P@3", "P@5"],
    )

    print(metrics)


.. GENERATED FROM PYTHON SOURCE LINES 102-105

The results will look similar to::

    {'Macro-F1': 0.5171960144875225, 'Micro-F1': 0.8008124243391698, 'P@1': 0.9573153795447128, 'P@3': 0.799074151109632, 'P@5': 0.5579924865442584}


.. _sphx_glr_download_auto_examples_plot_linear_quickstart.py:

.. only:: html

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

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

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

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

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


.. only:: html

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

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