Transcript Document
New Developments in Bayesian Network Software (AgenaRisk) Fifth Annual Conference of the Australasian Bayesian Network Modelling Society (ABNMS2013), Hobart, Tasmania, 28 Nov 2013 Norman Fenton Web: www.AgenaRisk.com Email: [email protected] Key differentiating features Risk Table view (tailorable questionnaire) Multiple scenarios Simulation and dynamic discretization (leading to intelligent parameter and table learning) Sensitivity analysis and multivariate analysis Binary factorization Parameter Passing between models Ranked nodes Comprehensive models and tutorials A free version with full standard BN functionality Sensitivity analyser Risk explorer view (linked BNOs Simulation node tool Simulation node Multivariate analyser Ranked node Expanding a node monitor Statistics State values Changing graph defaults Defining the states of a numeric (simulation node) That’s it. No need to worry about discretization intervals Static v Dynamic Discretization Static v Dynamic Discretization Result has mean 25 Result has mean 30 Multiple scenarios Multiple scenarios in Risk Table view Sensitivity Analyser Sensitivity Analyser Sensitivity Analyser Results Statistical distributions Parameter learning: priors Parameter learning: 2 data points Parameter learning: 7 data points Parameter learning: inconsistent data Binary factorization Parameter Passing Parameter Passing Solves classic BN problem of how to access just the summary statistics for a node Ranked nodes example Whole NPT defined in seconds Whole NPT defined in seconds Priors Impact of some observations Add testing effort Now backwards inference Only want to spend minimal effort ..and staff have average experience Change the scale Instant rescaling AgenaRisk Versions AgenaRisk Free AgenaRisk Lite AgenaRisk Pro Open and run any model Risk map, risk table, and risk explorer views Fully configurable risk graphs Sensitivity analysis Multivariate analysis Import/export functionality Create new model Pre-supplied models, tutorials, User manual Save Model containing just Boolean and labelled nodes Save model containing ranked nodes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes max 5 max 10 Unlimited Save model containing simulation nodes Save model containing multiple BNOs max 5 max 2 max 10 max 5 Unlimited Unlimited Maintenance support Upgrades Cost None None Free None Unlimited None Unlimited Free to buyers of Subscription book Also API Version available Supporting Book CRC Press, ISBN: 9781439809105 , ISBN 10: 1439809100 www.bayesianrisk.com Supporting Book Chapters 1. There is more to assessing risk than statistics 2. The need for causal explanatory models in risk assessment 3. Measuring uncertainty: the inevitability of subjectivity 4. The Basics of Probability 5. Bayes Theorem and Conditional Probability 6. From Bayes Theorem to Bayesian Networks 7. Defining the Structure of Bayesian Networks 8. Building and Eliciting Probability Tables 9. Numeric Variables and Continuous Distribution Functions 10. Hypothesis Testing and Confidence Intervals 11. Modeling Operational Risk 12. Systems Reliability Modeling 13. Bayes and the Law Plus extensive resources and models at www.bayesianrisk.com Future Releases Version 6.1 (Dec 2013) New algorithm with enhanced DD accuracy and efficiency Many additional models Web services version BAYES-KNOWLEDGE add-ons www.eecs.qmul.ac.uk/~norman/projects/B_Knowledge.html