.. _MNIST_label:

=============================
Experiments on MNIST Database
=============================

The MNIST_ database is a large database of handwritten digits
commonly used to test image processing systems.

Preparation
^^^^^^^^^^^^

Get the raw data
################
To download the raw MNIST data set with the default parameters, run:

.. code-block:: console

   python -m datasets.mnist.get_raw -m sacred

In this case, the process is very simple. We use Keras_ to download the MNIST data to :file:`~/.keras/datasets/mnist.npz.`


.. _MNIST: https://en.wikipedia.org/wiki/MNIST_database
.. _Keras: https://keras.io/

Serialize the raw data
######################

To serialize the raw data use:

.. code-block:: console

   python -m datasets.mnist.serialize -m sacred


Unconditioned MNIST GAN
^^^^^^^^^^^^^^^^^^^^^^^

The following experiment shows how to generate random numbers between 0 and 9 with
the GanEstimator_ module. For implementation details see :ref:`components_label` or :ref:`API_label`

To run the code with the default parameters defined in the config function
just type from the root directory:

.. code-block:: console

   python -m exps.mnist.train

Or use the command line to set different parameters:

.. code-block:: console

   python -m exps.mnist.train with mnist_feeder.batch_size=16 gan_type="UNCOND" -m sacred

.. _GanEstimator: https://www.tensorflow.org/api_docs/python/tf/contrib/gan/estimator/GANEstimator

Example output
###############
With the default parameters we will obtain the following result for training steps 100, 400 and 2000:

.. raw:: html

   <iframe src="../_static/slider_mnist_uncond.html" height="420px" width="410px" frameborder=0></iframe>

Conditioned MNIST GAN
^^^^^^^^^^^^^^^^^^^^^^
In this experiment, we wish to condition on both the generator and discriminator.
We generate MNIST digits conditioned on class labels.

To run the conditioned MNIST GAN use:

.. code-block:: console

   python -m exps.mnist.gan with mnist_feeder.batch_size=16 gan_type="COND" -m sacred

Example output
###############
With the default parameters we will obtain the following result for training steps 100, 400 and 2000:

.. raw:: html

   <iframe src="../_static/slider_mnist_cond.html" height="420px" width="410px" frameborder=0></iframe>