the problem of jSMILE parameter learning

[Ning]

Dear all,
i'm Ning, a master computer science student. now i'm doing my research project on traffic flow prediction based on Bayesian network. I worked on Genie to construct the road network which in DAG means that every node denotes sensor collection traffic information(speed, travel time, traffic flow.etc), and arcs denotes traffic flow direction. the construction of Bayesian network seen the diagram below:




I already have traffic dataset for one week. how can i learn the parameter from these data and my case is unlike other bayesian network, because they had different states associating to its node and the assigned probability.(e.g. disease symptoms, one symptom is caused by some external reason ), however how can i setup my state here?

secondly, because of considering the time factor, i plan to use previous time data for predicting the current time, in other words, in the diagram, both of X2 node is either cause node and effect node. how could i predict the performance in this case?


thank you very much in advance.
kind regards,
Ning

[mark]

I don't understand what you're trying to do but have you thought about using a Dynamic Bayesian Network to handle time?

[Ning]

I don't understand what you're trying to do but have you thought about using a Dynamic Bayesian Network to handle time?

Hi Mark,
okay, i give you more details.as seen the diagram, both node X1 and node X2 are cause nodes, considering on time factor(previous time t-j, j=1,2,3), actually the network structure has 6 logical nodes as cause nodes(x1(t-1),x1(t-2),x1(t-3),,x2(t-1),,x2(t-2),x2(t-3)), each of which affects traffic flow of X2 at t time(t is current time), the goal is that using x2 node its previous time traffic flow and its adjacent node x1 traffic flow values to predict x2 at t time traffic flow.

the historic traffic flow data contains sensors x1 and x2 data from 12/10/2011 to 18/10/2011, which recorded by sensors in every 10 minutes. so, assuming t is 9:30 t-1 means 9:20, x1(t-1) denotes traffic flow at 9:20,etc.
how does SMILE do prediction? could you give an example?

thank you and regards,
Ning

[mark]

Do you have any experience with Dynamic Bayesian Networks? It sounds to me they do exactly what you want.

[Ning]

Do you have any experience with Dynamic Bayesian Networks? It sounds to me they do exactly what you want.

no, i don't. could you show me any tutorial of genie_smile about DBN?

thanks!

[mark]

You can check the GeNIe documentation (http://genie.sis.pitt.edu/wiki/Main_Page) but I'm afraid it's lacking a bit at the moment.