Approximating the response of a Compton telescope with a neural network

This toolset aims to approximate the all-sky imaging response of a Compton telescope with a neural network.

For a description of how the imaging response of a Compton telescope looks like, see here: Zoglauer et a., COSI: From Data to Images

Setup

Prerequisites

The only requirements are access to the command line, git, and Python 3 (3.6 or higher). The following assumes you are using bash either on Windows, Linux, or Mac.

Get the source code

The current version of the source code is hosted on GitHub. Use these commands to clone the repository:

git clone https://github.com/zoglauer/nn-response-approximation ResponseApproximation

Then switch to the newly created folder:

cd ResponseApproximation

Creating the Python environment

In order to have a clean Python environment into which we can install all the packages we need for training and analysis, we create a virtual python environment. Do this via:

python3 -m venv python-env
. python-env/bin/activate
pip3 install -r Requirements.txt

This creates the environment, activates it, and installed all required packages.

Using it

Remember to activate your python environment whenever you are switching to a new bash shell:

. python-env/bin/activate

Now you are ready to run the proof-of-concept approach.

Run it

The proof-of-concept approach can be started simply via:

python3 run.py

This will show two windows, one with the reference cone sections and one with the learned cone sections. Good results are reached after ~10,000 iterations. Press Ctrl-C 3 times to stop the training process.