home | library | feedback |
B-course is motivated by the problems in the current statistical practice. When applied by practitioners, in many cases the underlying assumptions and restrictions are not clear to the user, and the complicated nature of the software encourages the users to "black box" approach where default parameter values are used without any understanding of the actual modeling and analysis task. This has lead to the situation where frequently the conclusions derived from the analysis are far from the intended plausible reasoning. B-Course is an attempt to offer a multivariate modeling method that can also be understood by applied practitioners. It is a first step in this direction and consequently can definitely be improved upon. However, it makes a serious attempt to give an informal introduction to the approach. Finally, one should not be misguided by the tutorial nature of B-Course. With the restrictions stated in the support material, in its own class B-Course is a powerful analysis tool that can (and hopefully will) be expanded to address some of its current limitations.
We have picked the Bayesian modeling framework, since we find it easier to understand than the classical frequentist framework (and thus hope that it is more understandable to the users also). We also feel that it has benefits over the frequentist framework. This is not to say that Bayesian approaches come without problems, both theoretical and practical problems are widely discussed in the literature.
Our purpose is to explain the idea of modeling for (discrete) data so clearly that users are able to use their own judgment to critically evaluate the modeling and the inference results.
It is. We strongly believe that Bayesian dependency modeling is a valuable tool in practitioner's statistical toolbox. In this belief we are not alone. However, B-Course concentrates on being understandable, not being state-of-the-art. Almost all parts of B-Course can be improved upon (at least we know how to improve them). Unfortunately, there are not many tools around to do dependency modeling even at B-Course's current sophistication level. We thus would also like to encourage others like us (researchers in these modeling methodologies) to build better analysis tools, and thus eventually produce a new, better toolbox for those applying the analysis methods.
It is too. Despite its simplicity, Bayesian classification based on Naive Bayes model(s) has proved to be one of the best general classification schemes. Since classification accuracy is relatively non problematic measure, this can be stated by certain degree of objectivity. However, once again B-Course concentrates on being understandable, not being state-of-the-art. Almost all parts of B-Course classification trail can be improved upon (at least we know how to improve them).
» References to the Bayesian approach and
Bayesian networks in science
B-Course, version 2.0.0 |