OrcaNet: A training organizer for Deep Learning in KM3NeT

https://badge.fury.io/py/orcanet.svg https://git.km3net.de/ml/OrcaNet/badges/master/pipeline.svg https://examples.pages.km3net.de/km3badges/docs-latest-brightgreen.svg https://git.km3net.de/ml/OrcaNet/badges/master/coverage.svg Codacy Badge

OrcaNet is a deep learning framework based on tensorflow in order to simplify the training process of neural networks for astroparticle physics, in particular for Orca, Arca and Antares. It incorporates automated logging, plotting and validating during the training, as well as saving and continuing the training process. Additionally, it features easy management of multiple neural network inputs and the use of training data which is split over multiple files.

In this sense, it tackles many challenges that are usually found in astroparticle physics, like huge datasets.

Documentation is at https://ml.pages.km3net.de/OrcaNet/.

OrcaNet is a part of the Deep Learning efforts for the neutrino telescope KM3NeT. Find more information about KM3NeT on http://www.km3net.org .

OrcaNet is currently being developed at the official KM3NeT gitlab (https://git.km3net.de/ml/OrcaNet).

However, there’s also a github mirror that can be found at https://github.com/ViaFerrata/OrcaNet.


OrcaNet can be installed via pip by running:

pip install orcanet

In order to make use of tensorflow’s GPU acceleration, you need cuda and cudnn installed. You can see which of these each tensorflow version needs here https://www.tensorflow.org/install/source#gpu


The easiest way to run OrcaNet is with singularity. A Singularity image of the latest stable version of OrcaNet with tensorflow and cuda/cudnn for GPUs is automatically uploaded to our sftp server. Download it e.g. via:

wget http://pi1139.physik.uni-erlangen.de/singularity/orcanet_v0.13.4.sif

and run it e.g. via:

singularity shell orcanet.sif

Indices and tables