Neural 3D Reconstruction for 21-cm Tomographic Data

Author: Nash Sabti
Paper: A Generative Modeling Approach to Reconstructing 21-cm Tomographic Data

How to Use the Code

1. Data Generation

   21-cm lightcones can be generated by running the following script:

   cd data_generation
   python run_21cmfast.py

   Each run will create 10 lightcones in a .npy file. We run this a total of 2500 times to create 25,000 boxes.

2. Data Processing

   python wedge_removal_and_augmentation.py

   This script augments the data to create 100,000 lightcones and removes modes in the wedge. Output is a 1.5TB hdf5 file with original and wedge-filtered lightcones.

3. Model Training

   python run.py

4. Sample Generation

   Set

   train_from_scratch=False
   write_samples=False

   to create a plot of the lightcone reconstruction, otherwise

   train_from_scratch=False
   write_samples=True
   N_samples=1000

   to generate samples and save them.

5. Plotting

   To recreate plots in the paper, run the files in the plotting folder:

   cd plotting

Contact

For any questions or issues, please contact: nash.sabti@gmail.com