FAQ

Can I do this in R?

The short answer is 'No'. There are Bayesian Network packages available in R, but none that provide support for any of the sophisticated algorithms and techniques that set Inatas apart (see here).

Of course, this might change - but there is limited scope for a R implementation to compete. Our software is written in highly optimized C++. It would take a pure R implementation over a year to process the information which Inatas System Modeler can process in under four hours. While existing packages call non-R procedures, none would be able to handle the complexity required to implement the alogrithms we currently use.

We hope, though, that it makes no real sense to try and 'do it in R'. We want to be able to provide a platform that, as well as being much more advanced than completing offerings, is cheaper and easier for the user to work with than attempting to programatically produce similar models themselves - even using a high level, low performance language like R. If you think that this is not the case, please contact us and we will work with you to make sure it is.

Why is network learning slower when involving missing data?

When learning a network from a dataset which involves missing data there is considerable additional overhead. Firstly, before each step in the learning algorithm, the missing data algorithm must be run in order to recalculate the estimates of the missing data items given the current topology. Secondly and consequently, since the data differs at each step, results from previous steps cannot be stored.