My WebPPL page

WebPPL is a Probabilistic Programming Language system and is available at http://webppl.org/ as an interactive system. The code is available via GitHub (https://github.com/probmods/webppl). However, it's a good idea to install it via node.js (npm install -g webppl) to be able to use it locally, especially since it is often faster (and IMHO neater) to run the models locally than via the web browser.
Unfortunately, WebPPL hasn't been updated for a while and to make it work properly requires a downgrade of the node.js version. Here's what I do when running WebPPL at command line: 
$ nvm use v10.22.0

Some of the packages that are available when running via a browser is not automatically available via the standalone version, e.g. the viz functions, the nice graph functions as well as the physics engine, etc. However, some of them can easily be installed via npm, e.g. the viz, see github.com/probmods/webppl-viz/blob/master/README.md for details.

Documentation and other resources

My WebPPL models

Many of these WebPPL model has been ports from (or via) my BLOG or PSI models. Inside the parenthesis, the "original" source are given. The models are also available at GitHub: github.com/hakank/hakank/tree/master/webppl (or will be soon)

Note that some most of the models use my WebPPL packages hakank_utils/distributions (in the directory node_modules/hakank_utils). These must thus be run with the --require parameter: 
  $ webppl--require hakank_utils model.wppl

Packages

The package includes the following utilities:

Models

Here are the models. As mentioned above, some model uses the hakank_util package (i.e. add --require hakank_utils when running from command line). Some models might also require the webppl-viz package (--require webppl-viz).

Distributions

Here are some "decompositions" of methods to generate different distributions (i.e. simple implementation often using the quantile/indidrect method). Many of these are collected in the package node_modules/hakank_utils/distributions.wppl (see a full list above). Note that these are in general not as stable/fast/good as the built-in distribution; for example some of these decompositions are not good for recovering parameters.
Back to my homepage
Created by Hakan Kjellerstrand (hakank@gmail.com)