Requires miniconda (Python 3.7). Install as follows:
curl -OL https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
After installation, install the environment using environment.yml
included in this repository:
conda env create --file=environment.yml
For recurrent setup, also do the following or add the equivalent lines in a setup.sh
script:
conda init <shell>
Activate the environment before running:
conda activate tf_gpu
Example training command (mind that using inputs from eos can be very slow):
python -i train_ptmiss.py -i /eos/user/s/steggema/ptmiss/dy_pf_chunk56.h5 --embedding
In order to run over a large number of samples, mixing different processes, create a "virtual hdf5" index file, and then run it with the command above:
python create_virtual_hd5.py -i /tmp/${USER}/dy_pf_chunk60.h5,/tmp/${USER}/tt_pf_chunk240.h5,/tmp/${USER}/dy_pf_chunk59.h5,/tmp/${USER}/tt_pf_chunk249.h5 -o out_20190930_v1.h5 -d X,X_c_0,X_c_1,X_c_2,Y,Z