Alexey Radul
Alexey Radul
on the grounds that the derivation is long enough that it should be tested. For instance, one could test that gradient ascent recovers the correct parameter values on an example...
Venture Lite fails to rejection sample an annoyingly large class of models that should be amenable to rejection sampling. For instance, consider the collapsed version of the Dirichlet process clustering...
Slice sampling suffers from a milder version of #384. Consider doing slice sampling on a variable with a beta prior. While looking for the region to slice over, the slice...
The basic idea called "Griddy Gibbs" is that one can (approximately) Gibbs-sample from any unnormalized 1-dimensional continuous posterior density by just doing the quadrature. Somehow this idea has gotten metamorphosed...
I have [recently thought](http://alexey.radul.name/ideas/2016/on-testing-probabilistic-programs/) about this in the abstract, and I mean to organize my thoughts into a concrete proposal for improving the usefulness of Venture's test suite. The resulting...
Venture has only one nesting level of "particles", which can be used either for running N independent Markov chains (by `(resample N)` once at the beginning and not otherwise), or...
Can we do better than 0? The definition is the maximum over all possible initial vectors and transmission and emission matrices of the marginal probability of the observations, integrating out...
Custom multinode blocks introduce a correctness problem that does not occur with single-site proposals. To wit, what happens if a block proposal changes the membership of the block being proposed...
Consider the following innocent-looking program: ``` [assume coin (flip)] [assume coin2 (if coin (flip) (flip))] [infer (gibbs default all 1)] ``` One would think that this would lead to the...
For example, `reportKnownGaussian` could use the K-S test to check the shape of the distribution, and also the Z test to check its mean. This might be desirable due to...