1. Notre Dame NLP
  2. Notre Dame NLP
  3. penne

Overview

HTTPS SSH

Python Easy Neural Network Extruder

This is a library that tries to make creating neural networks as easy as possible by being as similar to Python/NumPy as possible. It borrows heavily from Theano and CNN. If you find it useful, or want to help improve it, please let me (David Chiang) know!

Installing

You need to have NumPy (>= 1.9) and six (>= 1.10). Having SciPy (>= 0.14) enables n-dimensional convolutions and a few other things.

pip install numpy scipy six
git clone https://bitbucket.org/ndnlp/penne.git
cd penne
python setup.py install

Building and installing libgpuarray

There is also experimental GPU support, which requires the development version of libgpuarray. See libgpuarray's documentation for detailed instructions, but here is the short version:

Install the CUDA toolkit and make sure that your PATH includes nvcc. OpenCL should work as well. Note that if the build process doesn't find either one, it will issue a warning, build successfully, and nothing will work!

This is what I used (replacing /usr/local with wherever you want the library to be installed):

pip install numpy cython mako nose
git clone https://github.com/Theano/libgpuarray
cd libgpuarray
cmake -DCMAKE_INSTALL_PREFIX=/usr/local
make install
CFLAGS=-Isrc LDFLAGS=-L/usr/local/lib python setup.py install

Then be sure to set your LD_LIBRARY_PATH to include both the CUDA and libgpuarray libraries.

Contributors

  • David Chiang (University of Notre Dame)
  • Ke Tran (University of Amsterdam)

License

Penne is open-source software under the MIT License. See the file LICENSE for more information.