davecwright3/holodeck

Massive Black-Hole Binary Population Synthesis for Gravitational Wave Calculations ≋●≋●≋

holodeck's Issues

Add functions for SBI https://www.mackelab.org/sbi/

We need functions to

• Load the KDE data for the SBI pipeline
• Extract features from the 2D Nfreq x 1000 "stacked" KDEs to a 1D vector (both free spec KDEs and Holodeck, but it will be the same function)
• Train the normalizing flow model using the SBI package
• Sample from the posteriors using the trained model
• Compare inferred parameters to injected using a test set (phenom library has one)

Figure out how to compare free spectra to collection of holodeck simulations

The free spectra are KDEs of the PSD in a given PTA dataset without any imposed spectral shape. We need to compare a discrete distribution of gwb spectra to these. I see two methods for this

1. Compute KDEs for holodeck realizations (slow) NOT SLOW. Going with this.
2. Bin the free spectra and holodeck realizations similarly so that we can extract features with some kind of embedding network

Write method to find nearest parameter tuple from a holodeck library using Euclidean distance