Alexey Radul

Also, for the reversibility proof to hold, the interval bounds have to be derived independently of the current state of the Markov chain, but that can be the user's problem....

Perhaps this could be viewed as a code style issue and thus linked to #429.

Tests that are broken in Puma because of this: - test_observe.py - test_inc_eval.py - test_make_sym_dir_mult.py - test_references.py

This is a variant of #581, except the latter omitted the possibility of using the quadrature CDF as an M-H proposal.

The project management decision is whether/when to do this. This also calls for design, and possibly math or planning.

Which, space usage aside, is only a problem while we are trying to traverse the keys in order to make multi-site Gibbs fail conservatively. When #462 is fixed, we can...

Should be implemented in Lite by #647.

Commit e339604ecbc75f0bf4d00d5d065895649371b8cd implements option 3. This makes dealing with the brittleness of the auto-bound algorithm much less pressing.

No, I think in the case of real derivatives this is simpler. The choice of which GP it was is discrete, so all partial derivatives of anything with respect to...

Could work for single-site. Will not generalize to block without a stack: consider detaching the state A = B = C = 1, D = 3, E = 5, for...