darkCNN

A convolutional neural network to constrain dark matter

The objective of this algorithm is to take in simulated data, augment it and then train a convolutional neural network with multiple channels to estimate the model of dark matter it came from.

This has been run and tested in a virtual environment, which it is highly recommended carried out.

Pre-reduced data that has been extracted from simulations consist of 2 dimensional maps, each 2x2 Mpc, projected to 10 Mpc made up of three "channels":

1. Total projected mass maps
2. Xray lumonisity maps
3. Baryonic Matter maps

There are four dark matter models:

1. CDM
2. SIDM0.1 (SIDM with a cross-section of 0.1 cm2/g )
3. SIDM0.3
4. SIDM1

Each with four redshift slices: z = 0., 0.125, 0.250, 0.375

For this example, the data cube is binned in to 40 kpc resolution pixels, resulting in maps of 50 x 50 pixels. Each map comes with auxillary data or "attributes", consisting of

1. Redshift
2. Cluster mass
3. Xray concentration ( equal to S(<100 kpc) / S(<400 kpc) )

Dark matter self-interactions are rotationally symmetric so augmentations rotating and flipping can artificially increase the sample. This is carried out with 10 rotations and a random flipping of the image.

Note that the example database that has been packaged with this is a subset of al the data.

To Install - Tested on Fedora 33, Mac OS X Big Sur with python 3.7

setup a virtual environment

virtualenv darkCNN -p /usr/local/bin/python3.7

cd darkCNN

source bin/activate

git clone https://github.com/davidharvey1986/darkCNN.git

cd darkCNN

python setup.py install

To run the unit tests run

python setup.py test

To Run

darkCNN -h #to bring up the help

Example (included in package)

Make sure that jupyter-lab is installed on the venv via

pip install jupyterlab

To run the example, enter the "example" directory and simply type

darkCNN

Then to run the jupyter-lab notebook ensure to add the virtual environment kernel via ipython kernel install --name "darkCNN" --user

Python Modules

augmentData.py : augment input data to increase sample size getSIDMdata.py : get training and tests with labels globalVariables.py : standarised modules required inceptionModules.py : the inception layers for the main model main.py : the main modules that trains and saves the models mainModel.py : contains two models : mainModel - from the Merten paper, and a simple model tools.py : a distribution of tools used in the suite of code

Requirements

Python == 3.7 tensorFlow astropy==4.0 matplotlib==3.3.3 OyQt5 keras scipy