Rémi Louf
Rémi Louf
We can port https://github.com/dfm/rmhmc with the author's blessing. It should be relatively straightforward, difficulties may come from the changes needed to support PyTrees instead of flat arrays. Since the integrator...
Nested Rhat allows to assess the convergence of many chains that have run for only a few steps. https://arxiv.org/abs/2110.13017
As discussed with the author of SGMCMCJax @jeremiecoullon we will be porting SGMCMCJax to blackjax. ## Gradient estimation - [x] Standard gradient estimator - [x] Control variate gradient estimator #299...
When #279 is merged we should be in a good position to add SMC-ABC algorithms. Some minor rethinking of the base SMC kernel may be required, but I expect it...
They are currently only tested as part of integration tests on full examples. They should be tested in isolation.
`kernel.py` is growing very large and the size is only ever going to increase not only as we add new algorithms but also as we add documentation. The fact that...
There are a few things that I find unsatisfactory with the SMC base kernel: 1. The use of `kernel_factory`, which prevents the users from passing different parameters at different iterations....
The (coupled) structure of window adaptation is a consequence of the use of Welford's algorithm to estimate the covariance, the structure of the pathfinder adaptation is a consequence of the...
The intent behind this function is unclear from a user perspective. `pathfinder` suffers from the same problem.