probably-baysig

probably-baysig copied to clipboard

there is code for detercting burn-in and effective sample size (ESS). Code for ESS needs to be checked by someone more capable than me.

glutamate

glutamate

It would be cool if all the function that require a gradient-calculation function on their input, e.g., the BFGS routine, had variants that took just the function whole gradient is...

barak

In particular, a Summary of the AD work can be found in ad-experiments/README.md and ad-experiments/Example.hs.

jtobin

discrete parameters

1

if we modify the type of runChain such that the posterior function has type Vector Double -> Vector Int -> Double (but the gradient has the same type, or maybe...

glutamate

write a test program to MCMC from a simple posterior, plot samples

1

using whatever plot library you like, ad for graident if necessary.

glutamate

we need a sampler and PDF of the Inverse-Wishart distribution to have a good prior for covariance matrices

glutamate

``` interleave :: Strategy a -> Strategy b -> Strategy (a,b) interleave (GStrategy a b) (VStrategy c d) = GStrategy ? ? interleave (VStrategy a b) (VStrategy c d) =...

glutamate

MALA strategy

1

port MALA code from probably/Math.Probably.MALA (or make it up from scratch) to probably-baysig. Need to check this is a correct MALA. I had to guess my way a bit when...

glutamate

better initial values finder in runChain

1

Right now the initial value is a single sample from the prior. We need to check that posterior values are not NaN and do a greedy optimisation.

glutamate