Dynamic Bayesian NetworksIn JMP / SAS lacks a whole part dedicated to both dynamic and static Bayesian networks. The importance of Bayesian networks and therefore of the world of machine learning is increasingly important and having the possibility of dedicated commands or packages would be very important. In R-project there is the bnlearn or bnstruct package that allow you to analyze data with repeated temporal dynamics, but in JMP and / or SAS nothing exists. In SAS there is a proc hpbnet procedure, but it generates networks with a target variable and not a network where you don't want to target, but free from constraints, as the two packages bnlearn and / or bnstruct do.
Automation and ScriptingContent OrganizationData AccessReliability AnalysisRe: Dynamic Bayesian NetworksIt can express complex phenomena. Can be expressed probabilistically. These expressions fit the human senses. I also strongly hope the Bayesian Network.
Re: Dynamic Bayesian NetworksA data science team at HP has started using BNA for predictor or factor screening for manufacturing problems. It's the first time someone has shown me a canned analysis available to engineers here where I had to say "JMP can't do that".
Re: Dynamic Bayesian NetworksA Bayesian Network analysis package created by Fenton and Neil has been available for years, it would be great if this type of analysis was possible in JMP. For example, could JMP implement a platform that allows an engineer to start with a directed acyclical graph (DAG) which can be continually matured into a proper mathematical analysis to calculate the joint and marginal probabilities given conditional probabilities. Two applications from the manufacturing space I would benefit from are A) extending the existing reliability block diagram platform to contain the conditional probabilities & perform the desired calculations and 2) build on the fishbone platform to allow DAG creation to conceptually frame out causal relationships of a complicated value stream or root cause problem and estimate probabilities throughout the network before investing in expensive DOEs just to get a an initial model that allows some manner of prediction (as is the basic approach established in the DMAIC framework).
