Martin Trapp

icon indicating a website link https://trappmartin.github.io/ icon indicating an email [email protected]

icon company Aalto University icon location Europe icon location Postdoctoral researcher @ Aalto University | Academy of Finland fellow Research interests: probabilistic models, tractability, nonparametrics, Bayesian stats

Results 65 comments of Martin Trapp

Poor inference for CRP DP GMM via SMC, PG, IS

Yes.

Benchmarks for other models

One interesting additional model could be the warped mixture model, as it requires the combination of a DP with a GP. paper: https://arxiv.org/abs/1206.1846

Error in gradient computation during StatsFuns benchmarks

I'll have a look at it.

1D gaussian input node implementation (WIP)

> 'sample_state' function currently does not use the 'threshold' parameter, uses randn(). What problems might this pose in terms of reproducibility? Is the `threshold` parameter the seed? It seems that...

Doc: more informative visualisation about MCMC

Sure, but let me tweak it a little so that it is somewhat modular. The current version is pretty much quick and dirty. 😸