David Widmann

Is there actually anything to implement for AdvancedMH here? I skimmed through the paper and it seems that the user just has to provide an RNG that returns a QMC...

Sure, this example is also mentioned in the paper. As far as I understood, therefore the authors propose to use so-called CUD sequences. In general, it seems difficult to avoid...

As I said, I am aware of the problems. My main point is that it seems difficult to ensure that these issues are avoided in the abstract interface in AdvancedMH...

AdvancedMH let's you provide an arbitrary RNG. And in particular you could provide a custom RNG that returns quasi-random numbers in the required order and with the required properties.

> For instance, does AdvancedMH require arrays of random numbers, or just one random number at a time? If so, how do I know which random sequence will be used...

The test failures seem to be real, the sampled values do not match the expected ones. As a side remark, probably it would be helpful for debugging to use `sample(...;...

> Where do you see this? As far as I can tell, the three failed runs show the following results: I looked at the raw logs, e.g., https://pipelines.actions.githubusercontent.com/1iyxZmDwGspc6UbQKDLTZ8y3DHazaAGCEXSw02G4avGy4Eell5/_apis/pipelines/1/runs/7302/signedlogcontent/15?urlExpires=2020-11-23T10%3A22%3A22.7581222Z&urlSigningMethod=HMACV1&urlSignature=%2FUDiYzMH2mBCIhIroN0uueS%2FjYsHAv7zGwfFj%2FCie90%3D. The nicely...

Is it possible to just add a new proposal and extend the existing sampler, if needed, but not add a new sampler? I think this would simplify the code and...

> Furthermore the accept/reject decision is taken in step, so the proposed samples might not end up in the chain - the AM algorithm only uses samples in the chain...

I'm not strictly against adding such a function but there is one major problem: transitions contain all parameters but only for a subset of them (e.g. just one parameter) one...