DL4DI: Deep Learning Assisted Data Inspection for Radio Astronomy

A repository containing the implementation of the paper entitled "Deep Learning Assisted Data Inspection for Radio Astronomy"

Installation

Install conda environment by:

    conda create --name dl4di_test python=3.7

Run conda environment by:

    conda activate dl4di_test

Install dependancies by running:

    pip install -r requirements

Usage

You need to create the training set using either generate_hera_data.py or generate_lofar_data.py (given you have access to the preprocessed lofar .hdf5 files).

Data set creation

For HERA data creation run the following from inside the data_generation directory

    python3 generate_hera_data.py

For LOFAR dataset creation run the following from inside the data_generation directory given that the 'path' field is specified correctly in config.py and you have the correctly preprocessed .h5 LOFAR spectrograms available. The downsampled dataset may be found at: https://doi.org/10.5281/zenodo.3702430.

Note that in order to use this dataset, each of the .zip files need to be extracted to the directory specified in 'path'

    python3 generate_lofar_data.py

Training

Run the following given the correctly generated training files

    python3 train.py <training_file> <archtitecutre> -p <wandb_project> -l <latent_dim>

Reference this work

@article{10.1093/mnras/staa1412,
    author = {Mesarcik, Michael and Boonstra, Albert-Jan and Meijer, Christiaan and Jansen, Walter and Ranguelova, Elena and van Nieuwpoort, Rob V},
    title = "{Deep Learning Assisted Data Inspection for Radio Astronomy}",
    journal = {Monthly Notices of the Royal Astronomical Society},
    year = {2020},
    month = {05},
    issn = {0035-8711},
    doi = {10.1093/mnras/staa1412},
    url = {https://doi.org/10.1093/mnras/staa1412},
    note = {staa1412},
    eprint = {https://academic.oup.com/mnras/advance-article-pdf/doi/10.1093/mnras/staa1412/33319604/staa1412.pdf},
}

Notes

• The logging and visualisation data is dependant on wandb (https://www.wandb.com/)

Licensing

Source code of DL4DI are licensed under the Apache License, version 2.0.