What is the status of BayesHive? Is it a finished product?
BayesHive is still a prototype, but we hope that even at this stage it is useful.
How long will it take to fit a model?
That depends n the model complexity more than on the amount of data you have. Models with many hidden parameters, such as unobserved timeseries, can take a long time. Simple statistical models such as regression and ANOVAs can be very quick, i.e. done in 1-2 seconds.
If fitting a model takes an unexpectedly long time, it may be a poor model.
How much data can BayesHive handle?
In practice, we are for the moment limited by data transfer over the web for displaying raw data and results rather than inference in large data sets. If your dataset is more than one megabyte, you may find it quite slow to click through the interface. We're working on improving this situation.
What did you use to build BayesHive?
What is Baysig?
Baysig is a new probabilistic programming language. It makes it easy to define complex statistical models and fit them to heterogeneous data.
Why should I learn Baysig? I already know R/Python/Matlab/etc. Why another programming language?
Working in Baysig is completely different from standard statistical programming languages or data analysis enviroments. In R/Python/Matlab the programmer principally transforms data or defines procedures that extract information from data. In contrast, in Baysig we build statistical models for the unprocessed data and then infer the parameters of these statistical models. These parameter estimates are then used to drive hypothesis tests, decisions, prediction and forecasts.
How is Baysig Different from WinBugs/JAGS/Stan?
Unlike WinBUGS, JAGS, and Stan, Baysig is a general-purpose programming language. This means that:
complex models can be built from components
new elementary probability distributions can be defined within the model
Post-inference calculations such as decisions can be expressed
There is support for plotting the model estimation results and/or the raw data
In addition, Baysig supports first-class ordinary and stochastic differential equations, which means you get inference in continuous dynamical systems.