Instructions on how to run:
Required Dependencies: keras, tensorflow, matplotlib, numpy, h5py, pycbc, scipy, sympy
If you want to use just one GPU for a run then you must set the following environment variable to your desired GPU number
export CUDA_VISIBLE_DEVICES="1"
If you want to produce features only for injections rather than for both injections and background triggers, then set the --just-inj argument to "True" (MUST have a capital letter in front).
1.) Login to lho cluster.
gsissh albert.einstein@ldas-pcdev2.ligo-wa.caltech.edu
2.) Login to dgx1 machine. (you must ask Stuart for an account fist ... anderson@ligo.caltech.edu)
ssh dgx-1
3.) In feature_gen_scripts, run get_sngl_stats python script to get features for testing and training.
e.g. python get_sngl_stats --ifo H1 --single-trigger-files H1-HDF_TRIGGER_MERGE_FULL_DATA-1128299417-1083600.hdf --veto-file H1L1-CUMULATIVE_CAT_12H_VETO_SEGMENTS.xml --veto-segment-name CUMULATIVE_CAT_12H --found-injection-file H1L1-HDFINJFIND_BBH01_INJ_INJ_INJ-1128299417-1083600.hdf --window 1 --output-file BBH01_test.hdf --temp-bank H1L1-BANK2HDF-1128299417-1083600.hdf --inj-file H1-HDF_TRIGGER_MERGE_BBH01_INJ-1128299417-1083600.hdf --inj-coinc-file H1L1-HDFINJFIND_BBH01_INJ_INJ_INJ-1128299417-1083600.hdf --ifar-thresh 0.1 --verbose
4.) In main directory, run pycbc_neural_network.py script to train/test on feature set generated in the previous step.
e.g. python pycbc_neural_network.py -d path/to/data/chunk*/*.hdf -b path/to/one/result/file/from/each/chunk*/BBH01.hdf -o /path/to/output/directory -t 0.7 -e 10 -bs 1000 -u usertag >/dev/null 2>err.txt &