DeepRoulette

droulette is a deep learning framework for estimating local gravitational lensing coefficients (Roulette parameters) directly from simulated image data. The goal of this project is to provide a standardised experimental protocol, using simulated images from CosmoSim.

The current version is closed source. We plan to release it open source when we first publish results based on the work.

This version is based on nicolopinci/droulette. The original version history has been squashed to get rid of BLOBs. It is still work in progress.

For user documentation, see the CosmoAI web page, particularly the Pipeline for roulette parameter recovery.

If you want to use the tool, please get in touch with me to discuss colaboration.

• Initial Prototype by Nicolò Pinciroli
• Current developer Hans Georg Schaathun

Installation

The library is not yet published on PyPI and has to be installed from the working directory.

The venv.sh script creates a virtual environment under /tmp and installs the library. If you want to use this, run

. venv.sh

For subsequent use, without reinstalling, use

. /tmp/venv/bin/activate

If you want to install in an existing environment, instead of using venv.sh, use this:

pip install -e .