OrcaSong: Preprocessing KM3NeT data for DL
OrcaSong is a project for preprocessing raw KM3NeT ORCA or ARCA event data for the use with deep neural networks, making use of km3nets data processing pipline km3pipe. Two different modes are available:
- For convolutional networks: produce n-dimensional 'images' (histograms)
- For graph networks: produce a list of nodes, each node representing infos about a hit in the detector
The input to Orcasong are offline/aanet root files, and the output are "DL" files in the hdf5 format, which can e.g. be used by the OrcaNet software. For more info, read the documentation here https://ml.pages.km3net.de/OrcaSong!
OrcaSong is a part of the Deep Learning efforts of the neutrino telescope KM3NeT. Find more information about KM3NeT on http://www.km3net.org.
OrcaSong can be installed via pip by running:
pip install orcasong
You can get a list of all the bash commands in orcasong by typing:
orcasong --help
Containerization
The easiest way to run OrcaSong is with singularity. A Singularity image of the latest stable version of OrcaSong is automatically uploaded to our sftp server. Download it e.g. via:
wget http://pi1139.physik.uni-erlangen.de/singularity/orcasong_v???.sif
where v??? is the version, e.g. orcasong_v4.3.2.sif. Run it e.g. via:
singularity shell orcasong_v???.sif