Category: web-programming

One thing I wanted to add to my new and shiny Pollen website was full-text search. I have about 200 old blog articles on here, and it would be convenient to be able to find things now and then. It seemed like an easy enough thing to do, especially since I only intended to do it at the level of software bricolage, using existing tools.

I've been doing some work on interactive features for the Radian plotting library recently, mostly just floating UI elements to allow you to switch between linear and log axes, to change the number of bins in histograms, and so on. There's an article on the BayesHive blog showing…

My Radian HTML plotting library is now open source! You can read about it on the BayesHive website here.

For the past few months, I've been spending most of my working hours on BayesHive, the brainchild of Tom Nielsen, CEO of OpenBrain. This is a web platform for doing Bayesian data analysis, intended to make these powerful methods more accessible for scientists and other data analysts.

We opened the system to some alpha users a while ago, but this week we've decided we're ready for wider beta testing, for people to have a look, kick the tyres and tell us what they think.

I've mostly been working on the web app part of the system, but in this post I want to talk a little bit about the general idea behind the system and the Bayesian approach to statistics. (As far as BayesHive goes, the system is made up of a reasonably complicated single page web app using the AngularJS framework, a Yesod server process to deal with the management of documents, data and models, and a model compiler and inference engine that does all the clever stuff.)

In the last three days, I've averaged 10 hours of billable time per day. That's ten hours per day actually standing at my desk coding, perhaps 10-20% more than that "on the job". I've been working on social and search aspects of the BayesHive web app: we have a pile of different…

UPDATE: Radian is now open source. Read about it on the BayesHive website here.

Courtesy of Hideharu Sakai, this article is also available in Japanese.

One of the things we needed to be able to do for the BayesHive software I'm working on with Tom Nielsen of OpenBrain is easily produce plots within our front-end web app. There are lots of JavaScript libraries for doing this: Tom started off using Flot, which is very quick and easy to use, but wasn't really flexible enough for what we wanted.

I'd just been working on another part of the software, a tool for constructing algebraic representations of dynamical systems (defined by systems of ordinary or stochastic differential equations), and I'd been using the AngularJS framework. It took a while to get used to, but it seemed as though it might, in conjunction with D3.js, provide a way to produce a declarative plotting API that would be really easy to use and could take advantage of the clever data binding features of Angular.

Thus was Radian born: a declarative extension to HTML for rendering data and functional plots as SVG graphics. You can go straight to a gallery of examples here. Details and some more examples below the fold. (Note that, since Radian uses SVG graphics, you'll need a fairly recent browser to see anything...)

So, I've not done any blogging since the ill-fated Midwinter Polyphasic Sleep Experiment. I have a backlog of articles I want to write, but most of those involve some significant investment of time and effort. So you get this instead...

Context

I've been doing a lot of JavaScript programming recently. One of the two contracts I'm working on at the moment is for a UK university spin-off called OpenBrain, who are trying to bring Bayesian statistics to the masses by making tools that allow scientists to build complex (or simple!) Bayesian models and have them analysed using Markov chain Monte Carlo methods. All the really clever stuff is in the server-side MCMC inference code, but the user-facing side of things is a web app with a Yesod backend. My mission is to prettify this and make it ready for prime-time.

It's a fairly complicated application that needs to manage literate Markdown documents describing statistical models, data sets for analysis and a bunch of other stuff. The first thing I looked at was a tool for building statistical models based on dynamical systems represented as systems of ordinary or stochastic differential equations. This tool allows you to give algebraic representations of ODEs or SDEs, which are then rendered as MathML (for prettiness and familiarity), you can do simulations of the systems and visualise the results, you can specify data sets that correspond to observations of system variables, you can set up prior distributions for system parameters, and you can then get some code (in a proprietary Haskell-like language specialised for statistical modelling) that can be used to drive MCMC inference for the parameter values.

All that stuff happens on the client side, except for the listing of the available data sets and the rendering of the model into code. This means that you need client side code for expression parsing, analysis of coupled systems of ODEs and SDEs (to turn them into a canonical form that you can simulate from), numerical integration of ODEs and SDEs, graphing, and a framework to tie it all together. That's a lot of client side code!