.. DO NOT EDIT.
.. THIS FILE WAS AUTOMATICALLY GENERATED BY SPHINX-GALLERY.
.. TO MAKE CHANGES, EDIT THE SOURCE PYTHON FILE:
.. "auto_examples/detectors/gradnormkl.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_detectors_gradnormkl.py>`
        to download the full example code.

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

.. _sphx_glr_auto_examples_detectors_gradnormkl.py:


GradNormKL
==============================

Running :class:`GradNormKL <pytorch_ood.detector.GradNormKL>` on CIFAR 10.

This method computes the :math:`\ell_1`-norm of the gradient of the KL divergence between the
softmax output and a uniform distribution, with respect to the weights of the final classification
head. In-distribution inputs tend to produce more peaked predictions (larger divergence from
uniform), yielding higher gradient norms. The score is negated so that higher values indicate OOD.

Unlike :class:`GradNorm <pytorch_ood.detector.GradNorm>`, no labeled OOD data or additional
classifier is required.

.. GENERATED FROM PYTHON SOURCE LINES 16-33

.. code-block:: Python
   :lineno-start: 17


    import logging

    from torch.utils.data import DataLoader
    from torchvision.datasets import CIFAR10

    from pytorch_ood.dataset.img import Textures
    from pytorch_ood.detector import GradNormKL
    from pytorch_ood.model import WideResNet
    from pytorch_ood.utils import OODMetrics, ToUnknown, fix_random_seed

    logging.basicConfig(level=logging.INFO)

    fix_random_seed(123)

    device = "cuda"


.. GENERATED FROM PYTHON SOURCE LINES 34-35

Setup preprocessing and data

.. GENERATED FROM PYTHON SOURCE LINES 35-44

.. code-block:: Python
   :lineno-start: 35

    trans = WideResNet.transform_for("cifar10-pt")

    dataset_in_test = CIFAR10(root="data", train=False, download=True, transform=trans)
    dataset_out_test = Textures(
        root="data", download=True, transform=trans, target_transform=ToUnknown()
    )

    test_loader = DataLoader(dataset_in_test + dataset_out_test, batch_size=64, num_workers=4)


.. GENERATED FROM PYTHON SOURCE LINES 45-47

Stage 1: Load pre-trained WideResNet for CIFAR-10.
Disable gradients for the backbone; only the final FC layer needs them.

.. GENERATED FROM PYTHON SOURCE LINES 47-52

.. code-block:: Python
   :lineno-start: 47

    model = WideResNet(num_classes=10, pretrained="cifar10-pt").to(device).eval()

    model.requires_grad_(False)
    model.fc.requires_grad_(True)


.. GENERATED FROM PYTHON SOURCE LINES 53-54

Stage 2: Create detector — no fitting required.

.. GENERATED FROM PYTHON SOURCE LINES 54-56

.. code-block:: Python
   :lineno-start: 54

    detector = GradNormKL(model, param_filter=lambda name: name.startswith("fc"))


.. GENERATED FROM PYTHON SOURCE LINES 57-58

Stage 3: Evaluate

.. GENERATED FROM PYTHON SOURCE LINES 58-65

.. code-block:: Python
   :lineno-start: 58

    print("Testing...")

    metrics = OODMetrics()
    for x, y in test_loader:
        metrics.update(detector(x.to(device)), y)

    print(metrics.compute())


.. _sphx_glr_download_auto_examples_detectors_gradnormkl.py:

.. only:: html

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

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

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

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

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

    .. container:: sphx-glr-download sphx-glr-download-zip

      :download:`Download zipped: gradnormkl.zip <gradnormkl.zip>`


.. only:: html

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

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