/neural-approx-ss-lfi

Codes for ICLR 21 paper: Neural Approximate Sufficient Statistics for Implicit Models

Primary LanguageJupyter Notebook

Neural Approximate Sufficient Statistics

Official codes for paper "Neural Approximate Sufficient Statistics for Implict Models", ICLR 21 spotlight, 2010.10079

Package Dependencies

  • Python 3
  • Pytorch
  • Matplotlib

Summary statistics

at /nn

  • Mean as statistics (MSN.py)
  • Infomax statistics (ISN.py)

Neural density estimators

at /nde

  • Mixture Density Network (MDN.py)
  • Maksed Autoregressive Flow (MAF.py)

Likelihood-free algorithms

at /algorithms

  • Sequential Monte Carlo ABC (SMC_ABC.py)
  • Sequential Monte Carlo ABC with s.s (SMC2_ABC.py)
  • Sequential Neural Likelihood (SNL_ABC.py)
  • Sequential Neural Likelihood with s.s (SNL2_ABC.py)

Inference problems

at /problems

  • Ising Model
  • Gaussian copula Model
  • Ornstein-Uhlenbeck process