Merge branch 'main' of github.com:Red-Portal/MCMCTesting.jl

LICENSE

@@ -1,6 +1,6 @@
 MIT License
 
-Copyright (c) 2023 Ray Kim <[email protected]> and contributors
+Copyright (c) 2023 Kyurae Kim <[email protected]>
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

README.md

@@ -5,8 +5,7 @@
 [![Build Status](https://github.com/Red-Portal/MCMCTesting.jl/actions/workflows/CI.yml/badge.svg?branch=main)](https://github.com/Red-Portal/MCMCTesting.jl/actions/workflows/CI.yml?query=branch%3Amain)
 [![Coverage](https://codecov.io/gh/Red-Portal/MCMCTesting.jl/branch/main/graph/badge.svg)](https://codecov.io/gh/Red-Portal/MCMCTesting.jl)
 
-This package provides exactness tests for MCMC algorithms as described by Gandy and Scott (2020).
+This package provides exactness tests for MCMC algorithms proposed by Gandy and Scott[^GS2020].
 
-## References
-* Gandy, Axel, and James Scott. "Unit testing for MCMC and other Monte Carlo methods." arXiv preprint arXiv:2001.06465 (2020).
+[^GS2020]: Gandy, Axel, and James Scott. "Unit testing for MCMC and other Monte Carlo methods." arXiv preprint arXiv:2001.06465 (2020).
 

docs/src/exactranktest.md

@@ -5,7 +5,7 @@
 The exact rank hypothesis testing strategy is based on Algorithm 2 by Gandy & Scott (2021)[^GS2021].
 
 ## `ExactRankTest`
-The ranks are computed by simulating a single Markov chain backwards and forward.
+The ranks are computed by simulating a single Markov chain backward and forward.
 First, a random midpoint is simulated as
 ```math
 \begin{aligned}
@@ -25,7 +25,7 @@ forming the chain
 \theta_1,\; \ldots, \; \theta_{M},\; \ldots, \; \theta_{L}.
 ```
 The *rank* is the ranking of the statistics of $\theta_{M}$.
-If the sampler and the model are correct, the rank has an uniform distribution as long as the midpoint is independently sampled.
+If the sampler and the model are correct, the rank is uniformly distributed as long as the midpoint is independently sampled.
 
 ```@docs
 ExactRankTest

docs/src/index.md

@@ -9,8 +9,8 @@ Documentation for [MCMCTesting](https://github.com/Red-Portal/MCMCTesting.jl).
 `MCMCTesting` provides the MCMC testing algorithms developed by Gandy & Scott (2021)[^GS2021].
 These tests can be seen as an improvement of the hypothesis testing approach proposed by Geweke [^G2004].
 Unlike simulation-based calibration (SBC; [^TBSVG2018][^YD2023][^MMKBPBHFGV2022]), these tests are more appropriate for testing the exactness of the MCMC algorithm rather than the identifiability of the models.
-This is because the tests focus on maximizing the power for verify the validity of *individual Markov transitions* instead of a *set of samples*.
-Furthermore, unlike SBC, the approach of Gandy & Scott[^GS2021] is able to exactly satisfy the assumptions required for the theoretical guarantees.
+This is because the tests focus on maximizing the power of verifying the validity of *individual Markov transitions* instead of a *set of samples*.
+Furthermore, unlike SBC, the approach of Gandy & Scott[^GS2021] can exactly satisfy the assumptions required for the theoretical guarantees.
 
 
 ## References

src/exactranktest.jl

@@ -24,7 +24,20 @@ This test requires the following functions for `model` and `kernel` to be implem
 - `sample_joint`
 Furthermore, this test explicitly assumes the following
 - `kernel` is reversible.
-Applying this tests to an irreversible `kernel` will result in false negatives even if its stationary distribution is correct.
+
+For reference, the following kernels are reversible:
+- Any Metropolis-Hastings kernel
+- slice samplers
+- Hamiltonian Monte Carlo/No-U-Turn Samplers with resampling over the trajectory
+- random-scan/permutation-scan Gibbs samplers
+- Unadjusted Langevin
+However, the following kernels are not reversible:
+- Hamiltonian Monte Carlo with persistent momentum (a.k.a Horowitz's method, generalized Hamiltonian Monte Carlo)
+- Systematic-scan Gibbs samplers
+- Piecewise deterministic Markov processes
+
+!!! info
+    Applying this test to an irreversible `kernel` will result in false negatives even if its stationary distribution is correct.
 
 # Keyword Arguments for Tests
 When calling `mcmctest` or `seqmcmctest`, this tests has an additional keyword argument: