Sampler roadmap / overview

[MichaelClerx]

Can't find an existing ticket for this, though sure we had one at some point.

Classed as black, blue, and red in Ben's diagram

Likelihood-free

• Hamiltonian ABC
• Markov chain monte-carlo ABC --> TICKET OR PR NEEDED
• Regression-adjustment ABC
• Rejection ABC --> PR ABC Rejection method #881 ABC rejection - rebased #925 ABC SMC/PMC #1442
• Sequential monte-carlo ABC PR ABC-SMC #1055 ABC SMC/PMC #1442

Derivative-free

• AM global-adaptation MCMC --> PR Adaptive MCMC: New class structure and multiple variants #491
• AM local-componentwise MCMC --> PR Adaptive MCMC: New class structure and multiple variants #491
• Delayed rejection adaptive metropolis (DRAM) --> PR Adaptive MCMC: New class structure and multiple variants #491
• Differential evolution --> DifferentialEvolutionMCMC
• Differential evolution adaptive metropolis (DREAM) --> DreamMCMC
• Ellipsoidal nested sampling --> NestedEllipsoidSampler
• Emcee hammer --> EmceeHammerMCMC
• Generalised elliptical slice sampling --> PR Integrate Slice Sampling: Generalised Elliptical. #900
• Gradient-free HMC Add gradient-free HMC with efficient kernel exponential families #101
• Localised adaptive covariance --> PR Adaptive MCMC: New class structure and multiple variants #491
• Mode jumping adaptive MCMC Mode jumping adaptive MCMC #492
• Population MCMC --> PopulationMCMC
• Rao-blackwell AM MCMC --> PR Adaptive MCMC: New class structure and multiple variants #491
• Random walk metropolis --> MetropolisRandomWalkMCMC
• Remi AM MCMC --> AdaptiveCovarianceMCMC
• Rejection nested sampling --> NestedRejectionSampler
• Sequential Monte Carlo --> PR Add SMC sampler #749
• Slice sampling with doubling --> SliceDoublingMCMC
• Slice sampling with hyperrectangles --> PR Integrate Slice Sampling: Hyperrectangles-based Methods. #895
• Slice sampling with overrelaxation --> Temp PR Slice sampling integration: 1. Stepout, 2. Doubling, 3. Overrelaxation, 4. Hyperrectangles, 5. Adaptive hyperrectangles, 6. Rank Shrink, 7. Covariance Match, 8. Generalised Elliptical #863
• Slice sampling with Peano curves
• Slice sampling with stepout --> SliceStepoutMCMC
• Transport map accelerated MCMC
• Vanilla AM MCMC --> PR Adaptive MCMC: New class structure and multiple variants #491
• Vina copula adaptive MCMC Ticket Adaptive MCMC using vine copulas #495

1st order sensitivities

• Adaptive HMC
• Adaptive MALA #adapti
• Bouncy particle HMC
• Covariance matching SS --> PR Integrate Slice Sampling: Covariance Matching MCMC #896
• Diffusive nested sampling Add diffusive nested sampling #152
• Dynamic nested sampling
• dyPolychord
• Galilean monte carlo Investigate Galilean Monte Carlo (a type of MCMC nested sampler) #341
• Hamiltonian Monte Carlo (HMC) --> HamiltonianMCMC
• Metropolis-adjusted langevin algorithm (MALA) --> MALAMCMC
• Monomial gamma HMC --> MonomialGammaHamiltonianMCMC
• Multinest Multinest #1034
• No-U-Turn sampler (NUTS) I293 no-u-turn sampler #1031
• Polychord Implement polychord nested sampling #305
• Rank shrinking SS --> Integrate Slice Sampling: Rank Shrinking MCMC (Correct) #899
• Shadow HMC Investigate Shadow Hamiltonian Monte Carlo #521

2nd order sensitivities

• Riemannian manifold HMC
• Riemannian manifold MALA

Framework for all sampling (temper distribution, sample from it, reweight via importance sampling):

• SIS MCMC --> PR SIS MCMC redux #760

[MichaelClerx]

@ben could you check if the bottom 6 are e.g. different names for ones above? Found them on github but not in the pyramid

[MichaelClerx]

Aha, here it is: https://github.com/orgs/pints-team/projects/1#card-15881685

[ben18785]

Hi Michael,

Thanks! I've edited the comment to remove those methods I think are duplicates. Not sure what to call SIS MCMC -- it's an approach that can be applied to all samplers (and likelihood-based optimisers) essentially. Just involves heating the distribution, sampling from it, then reweighting based on importance weights. Can leave it as it stands for now though!

[MichaelClerx]

Thanks!