Alexey Radul
Alexey Radul
To keep track of them I would need to adjust the custom automatic differentiation system to propagate gradients from evaluation to constant nodes, read and them off the constant nodes...
The claim that Venture "has Hamiltonian Monte Carto" (HMC) is belied by the paucity of gradient annotations in the stochastic procedure library. The most common ones are there, but if...
Consider doing Hamiltonian Monte Carlo on a variable whose prior follows the gamma distribution. Gamma's output is always positive, but (for some parameter values) its density grows as the value...
Currently, the derivative information for the output's influence on the latent variable, and both of those influences on the inputs, is simply lost. Issues: - Is it reasonable to take...
distributed according to the current distribution. @vkmvkmvkmvkm envisions this as an interactivity feature. User story (reconstructed from memory): - I'm noodling around with Venture in the console or notebook, with...
Benefit: A presumably more scalable (than naive global rejection sampling) generic inference algorithm that also guarantees samples from the exact posterior. Reference: Mansinghka, Roy, Jonas, Tenenbaum, "Exact and Approximate Sampling...
against numerically approximating the gradients of their corresponding `logDensityOfData` methods. Possible paths: - Expand the evaluation faking that the randomized test suite performs (which should be similar to what the...
The following program fails with "Cannot make random choices downstream of a node that gets constrained during regen" (even though it does no explicit inference): ``` (assume x (flip)) (force...
(originally reported September 2014) https://app.asana.com/0/9277420529946/16064839336502 https://github.com/probcomp/Venturecxx/blob/ml-models/examples/ml-models/pmf.py
Consider the following program: ``` assume k = some gp covariance function assume gp1 = make_gp(k) assume gp2 = make_gp(k) .... use gp1 and gp2 ``` `gp1` and `gp2` represent...