Paper reproduction walkthrough

Reproducing the models

To reproduce the stats on the paper you need first to run all the models

python run_paper.py

With a GPU and a --batch_size = 128 (default) this takes around two weeks. If you increase batch_size it may be reduced to a couple of days but performance can be slightly reduced.

Reproducing the stats and plots

Summary statistics for all trained models and a printout with the stats and plots used in the paper are produced by:

python run.py --performance

Summary statistics will be found in snndump/stats/summary_stats.csv. Statistics used in the paper are printed out and latex tables created in snndump/latex/. Plots and figures are found in snndump/figures/ and snndump/lightcurves/.

To obtain summary statistics only, comment in the two lines after # Stats and plots in paper in run.py.

To obtain stats only, comment the plotting function at SuperNNova/supernnova/paper/paper_thread.py by changing SuperNNova_stats_and_plots_thread(df_stats, settings, plots=False).