greedythickthinning and k2 priors

[cs]

i have some questions about the k2 priors and your implementation of greedythickthinning:

1. greedythickthinning: in an older post i read that greedythickthinning is based on heckerman's tutorial and i guess it is based on the algorithm described in section 11 of this paper. but if so, there are still some unclear topics. first, what is the scoring criterion you use? BD? second, how do you determine the network structure for the start of the greedy search? the name of the algorithm suggests that a fully connected graph is used as the initial network.

2. k2 priors: is there any literature about this method?

thanks in advance

[mark]

Thanks for using the search first. The scoring criterion used is the marginal likelihood given in equation 35 of the Heckerman tutorial. GreedyThickThinning starts with an empty network, then first greedily thickens it and finally thins it.

I don't have a literature reference on K2, but I think that it was first used in software that carried the same name. So you can do a search on that, but be careful not to confuse it with the K2 learning algorithm.

[edimilsonbatista]

Hi!

Is there a function that returns me the value of this score calculated to determine the best structure?

Thanks!

[mark]

No, not at the moment.