To setup a conda environment named 'gum' to host the build process and install the python dependencies for the build bot, run the following command from the root (gum) folder:
conda env create --name gum --file environment.yml
conda activate gum
Step Two (Building the Cython dependencies) is optional and needs to be done only if you intend to re-generate the constituent parses using the neural parser with the -p flag.
Run the following to compile the cython packages used by the neural parser (assuming you're still at the root (gum) folder):
Navigate to the _build folder:
cd _build
(Windows machines only) Install the Visual C++ Build Tools, needed for compiling the Cython modules. This is a one-time installation activity on the Host OS.
Download and Install the latest Build Tools for Visual Studio 2019 from here
For other versions, please check here
Then run this to compile and install the cython packages.
python setup.py build_ext --inplace
If this works, you should see these files in the _build folder (indicating successful compilation). If no, consider upgrading cython.
const_decoder.c
hpsg_decoder.c
const_decoder.cpython-37m-x86_64-linux-gnu.so (on Ubuntu / Linux)
const_decoder.cp37-win_amd64.pyd (on Windows 64-bit)
const_decoder.cpython-37m-darwin.so (on MacOS)
hpsg_decoder.cpython-37m-x86_64-linux-gnu.so (on Ubuntu / Linux)
hspg_decoder.cp37-win_amd64.pyd (on Windows 64-bit)
hpsg_decoder.cpython-37m-darwin.so (on MacOS)
At this stage, the build bot process can be started as normal
Run the following to start the build bot. and generate fresh constituent parses
python build_gum.py -s src -t target -p -v
This will download the best PyTorch model for constituent parsing automatically and summon it for inferring the constituent parses, which are saved to the folder target/const