3 ms·
How does this scale for high-dimensional data? The example uses 2 dimensions - connections and latency - but real world (performance) data will often have many
by huac 11y ago
How does this scale for high-dimensional data? The example uses 2 dimensions - connections and latency - but real world (performance) data will often have many more associated variables. I don't know a lot about Bayesian methods but it seems like calculating joint posteriors would become slow.
But definitely a very interesting approach, and the Stan code is surprisingly readable!
- ppcsf 11y agoStan uses Markov chain Monte Carlo for inference. MCMC algorithms can do very well at approximating the high-dimensional integrals required to calculate the posterior; estimating 1000-dimensional distributions with MCMC methods is not unusual.
- glial 11y agoNot unusual, but the sampling can take a long time and scales poorly with the amount of data you have.