Bug#902390: ITP: keras-preprocessing -- data preprocessing module for the Keras deep learning framework

[Stephen Sinclair <radarsat1@gmail.com>]

Package: wnpp
Severity: wishlist
Owner: Stephen Sinclair <radarsat1@gmail.com>

* Package name    : keras-preprocessing
  Version         : 1.0.1
  Upstream Author : Francois Chollet <francois.chollet@gmail.com>
* URL             : http://keras.io/
* License         : Expat
  Programming Lang: Python
  Description     : data preprocessing module for the Keras deep learning framework

Keras is a Python library for machine learning based on deep (multi-
layered) artificial neural networks (DNN), which follows a minimalistic
and modular design with a focus on fast experimentation.

Features of DNNs like neural layers, cost functions, optimizers,
initialization schemes, activation functions and regularization schemes
are available in Keras a standalone modules which can be plugged together
as wanted to create sequence models or more complex architectures.
Keras supports convolutions neural networks (CNN, used for image
recognition resp. classification) and recurrent neural networks (RNN,
suitable for sequence analysis like in natural language processing).

It runs as an abstraction layer on the top of Theano (math expression
compiler) by default, which makes it possible to accelerate the computations
by using (GP)GPU devices. Alternatively, Keras could run on Google's
TensorFlow (not yet available in Debian).

Keras Preprocessing is the data preprocessing and data augmentation
module of the Keras deep learning library. It provides utilities for
working with image data, text data, and sequence data.

This new package is to be a dependency for an updated keras package,
following upstream's splitting of keras-preprocessing into its own
Python module.