Table of Contents generated with DocToc

• PyDeepCL
• How to use
• Notes on how the wrapper works
• To install from pip
• To build directly
  • Pre-requisites:
    • Compilers
    • Python packages
  • To build:
• Considerations for Python wrapper developers

PyDeepCL

Python wrapper for DeepCL

How to use

See test_clconvolve.py for an example of:

• creating a network, with several layers
• loading mnist data
• training the network using a higher-level interface (NetLearner)

For examples of using lower-level entrypoints, see test_lowlevel.py:

• creating layers directly
• running epochs and forward/backprop directly

Notes on how the wrapper works

• cDeepCL.pxd contains the definitions of the underlying DeepCL c++ libraries classes
• PyDeepCL.pyx contains Cython wrapper classes around the underlying c++ classes
• setup.py is a setup file for compiling the PyDeepCL.pyx Cython file

To install from pip

pip install DeepCL 

• related pypi page: https://pypi.python.org/pypi/DeepCL

To build directly

Pre-requisites:

Compilers

• on Windows:
  • Python 2.7 build: need Visual Studio 2008 for Python 2.7 from Microsoft
  • Python 3.4 build: need Visual Studio 2010, eg Visual C++ 2010 Express

Python packages

• Need the following python packages installed, eg via pip install:
  • cython

To build:

cd python
python setup.py build_ext -i

Considerations for Python wrapper developers

• By default, cython disables ctrl-c, so need to handle this ourselves somehow, eg see NetLearner.learn for an example
• To handle ctrl-c, we need to use nogil, which means we cant use the except + syntax, I think, hence need to handle this ourselves too :-) see again `Netlearner.learn for an example