deep21 tutorial data

Sample data for training and testing the deep21 deep learning model for 21cm cosmology. The object of the experiment is to separate radio cosmological signal from foreground contaminants, with a Principal Component Analysis (PCA) preprocessing step. Data were originally generated in .fits file format via CRIME Simulation Package (see Alonso et al. 2014 for details). Included are binary numpy (.npy) files for 5 full-sky simluations of the cosmological signal, observed signal, and reference PCA-subtracted maps. Loading files with Numpy will yield arrays of shape (\(N_{\rm voxels}, N_x, N_y, N_\nu\)) = (950, 64, 64, 64), with 192 voxels per simulation. Designed for use with the browser-based deep21 tutorial on Google Colab (fuller explanation of experiment also available). Full-scale processing scripts are available on the deep21 GitHub repository.