annlia
commented
6 years ago Hi,
Thanks for your interest in our method for sampling DAGs.
I am not sure what you mean by outcome variable, since DAGs provide a representation of a joint multivariate probability distribution, and there is no such a thing as a unique outcome variable.
The partition MCMC software is a sampler, which means it provides a sample of DAGs from their posterior distribution given the data. The sample is provided as a list of DAGs (in the form of adjacency matrices) with associated scores. If you wish to find a Maximum a Posteriori DAG you may take the element of the returned list with maximum score. However in general this is not recommended, since depending on the number of nodes, this may not be very stable. Therefore for inferential purposes it is normally better to consider an ensemble of DAGs.
Finally I would suggest that you use our our more recent package biDAG available on CRAN
https://cran.r-project.org/web/packages/BiDAG/index.html https://cran.r-project.org/web/packages/BiDAG/index.html
Details about the method are provided in our preprint
https://arxiv.org/abs/1803.07859 https://arxiv.org/abs/1803.07859
Hope this helps and let me know if you have more questions.
Kind regards, Giusi
On 14 Jun 2018, at 01:23, Vincent Pisano notifications@github.com wrote:
Hi, so I read the paper and ran the BH example, but am a bit confused where to find the structure of the best fit DAG. If I'm not mistaken, the script should take a dataset of predictor variables and determine the best DAG for an outcome variable.
Where/how is the outcome variable set? How can I see the best scoring DAG structure for further analysis? I see the mcmc graph, but that tells me nothing about the DAG structure. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/annlia/partitionMCMC/issues/1, or mute the thread https://github.com/notifications/unsubscribe-auth/AEvvt6qQiHZB4lSSTClUGGsjpAPd0d0eks5t8Z7hgaJpZM4UnFhE.
Hi, so I read the paper and ran the BH example, but am a bit confused where to find the structure of the best fit DAG. If I'm not mistaken, the script should take a dataset of predictor variables and determine the best DAG for an outcome variable.