egamma

Project repository for NBI e/gamma hackathon 2018

Setup

Different parts of the project require h5py, ROOT, root_numpy, Jupyter notebook, Keras, TensorFlow, and Pytorch. These can all be installed as part of the conda environments specified in envs/. To set this up, first install e.g. Miniconda2

$ wget https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh
$ bash Miniconda2-latest-Linux-x86_64.sh
$ # Follow the screen prompts
$ # ...
$ rm Miniconda2-latest-Linux-x86_64.sh

then create and activate e.g. the CPU environment as

$ conda env create -f envs/egamma-cpu.yml
$ source activate egamma-cpu

Similarly, envs/egamma-gpu.yml provide an environment suitable for running on GPU(s).

Known issues

For some reason, there is a problem with linking to libstdc++ whe importing TensorFlow, manifesting as an error like:

ImportError: <conda/env/dir>/lib/./libstdc++.so.6:
version `CXXABI_1.3.9' not found (required by <conda/env/dir>/lib/python2.7/site-packages/google/protobuf/pyext/_message.so)

This can be fixed as follow:

$ source activate egamma
$ envdir="$(conda info --env | grep \* | sed 's/.* //g')"
$ linkpath="$envdir/lib/libstdc++.so.6"
$ latestlib="$(find $envdir/lib/ -name libstdc++.* ! -type l | grep -v .py | sort | tail -1)"
$ ln -s -f $latestlib $linkpath

savvy379/egamma

egamma

Setup

Known issues