/AstroVAE

Data reduction and inference problems using a combination of GP emulator and Variational autoencoder

Primary LanguageJupyter Notebook

AstroVAE

Data reduction, emulation and inference using a combination of GP processes and Variational autoencoder. Currently applied for CMB angular power spectra C_l and matter power spectra P(k). In principle, this can be extended to image emulations as well.

Parameter inference is done using MCMC for cosmological parameters, with public PLANCK/WMAP/SPT data.

Sync commands

From phoenix -> laptop (AstroVAE/Cl_data/Data)

scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Model/7500 mcs:/homes/nramachandra/DataP5/Model/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/norm7500 mcs:/homes/nramachandra/DataP5/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/mean7500 mcs:/homes/nramachandra/DataP5/ scp phoenix:/homes/nramachandra/AstroVAE/Cl_data/Data/encoded7500 mcs:/homes/nramachandra/DataP5/

From laptop -> phoenix

scp P25. mcs:/homes/nramachandra/DataP5/raw/

Future implementations

  1. Error propoagation using Bayesian neural networks
  2. t-SNE for reduction and visualization
  3. RNN for time analysis
  4. Adversarial AEs