Dear All,
I am using GeNIe 2.5 academic version for bayesian network analysis.
Please, I need to count the number of relationships between a specific variable and the main parent in the Bayesian Network resulting after the learning. What is the quickest way to do this? One way is counting manually the number of arrows, but we need a more automated way to do this. Can we access to this specific information (number of relationships from each variable to the main parent) in another format more easy to handle it (in an Excel file, for instance)? Many thanks in advance!
Regards
Ashwani
Counting of relationship between a specific variable and the main parent
-
- Posts: 3
- Joined: Wed Oct 07, 2020 4:36 pm
-
- Site Admin
- Posts: 432
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Counting of relationship between a specific variable and the main parent
Hi Ashwani,
First of all, I suggest that you upgrade to GeNIe 3.0 -- you will like it more!
I believe you are asking for limiting the number of parents of a node in the process of learning the model structure. If so, the quickest way to achieve this is to set the "Max Parent Count" parameter (in the PC algorithm, it is called "Max Adjacency Size," as arcs can be undirected, so technically nodes are not always parents). The structure learning algorithm will not create more parents (neighbors) than this parameter.
You may also mean the number of parents of an existing network (whether it has been just learned or created manually). GeNIe shows the average and maximum in-degrees (to see these, please look at Network Properties, Summary tab). In-degree is essentially the number of incoming arcs, hence the number of parents. To see how many parents each node has, you will need to count them individually or write a simple SMILE program that will output a text or a CSV file for opening in Excel.
I hope I have understood your questions correctly. If not, please let me know and I will try again.
Cheers,
Marek
First of all, I suggest that you upgrade to GeNIe 3.0 -- you will like it more!
I believe you are asking for limiting the number of parents of a node in the process of learning the model structure. If so, the quickest way to achieve this is to set the "Max Parent Count" parameter (in the PC algorithm, it is called "Max Adjacency Size," as arcs can be undirected, so technically nodes are not always parents). The structure learning algorithm will not create more parents (neighbors) than this parameter.
You may also mean the number of parents of an existing network (whether it has been just learned or created manually). GeNIe shows the average and maximum in-degrees (to see these, please look at Network Properties, Summary tab). In-degree is essentially the number of incoming arcs, hence the number of parents. To see how many parents each node has, you will need to count them individually or write a simple SMILE program that will output a text or a CSV file for opening in Excel.
I hope I have understood your questions correctly. If not, please let me know and I will try again.
Cheers,
Marek
-
- Posts: 3
- Joined: Wed Oct 07, 2020 4:36 pm
Re: Counting of relationship between a specific variable and the main parent
Dear Marek,
Thank you very much for your explanation. Appreciated.
I have one more doubt about the algorithms bering used for BN learning.
According to the methodology of my analysis, I must use K2 algorithm introduces by Cooper & Herkovitz, 1992. (Please see attachment)
According to the GeNIe User's guide, I think "Bayesian Search" is based on K2. Am I right?
Can you please explain me in more detail way?
Thanks
Ashwani
Thank you very much for your explanation. Appreciated.
I have one more doubt about the algorithms bering used for BN learning.
According to the methodology of my analysis, I must use K2 algorithm introduces by Cooper & Herkovitz, 1992. (Please see attachment)
According to the GeNIe User's guide, I think "Bayesian Search" is based on K2. Am I right?
Can you please explain me in more detail way?
Thanks
Ashwani
- Attachments
-
- K2 algorithm model.
- K2 Algorithm.jpg (89.89 KiB) Viewed 3559 times
-
- Site Admin
- Posts: 432
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Counting of relationship between a specific variable and the main parent
Hi Ashwani,
You are right in that the Bayesian Search implementation in GeNIe/SMILE is based on the algorithm proposed by Cooper and Herskovitz. You may also want to look at the paper by David Heckerman "A Tutorial on Learning With Bayesian Networks". It contains a couple of new ideas that enhance Cooper and Herskovitz algorithm. Here is one of the places where you can find it:
https://arxiv.org/abs/2002.00269
I hope this helps,
Marek
You are right in that the Bayesian Search implementation in GeNIe/SMILE is based on the algorithm proposed by Cooper and Herskovitz. You may also want to look at the paper by David Heckerman "A Tutorial on Learning With Bayesian Networks". It contains a couple of new ideas that enhance Cooper and Herskovitz algorithm. Here is one of the places where you can find it:
https://arxiv.org/abs/2002.00269
I hope this helps,
Marek
-
- Posts: 3
- Joined: Wed Oct 07, 2020 4:36 pm
Re: Counting of relationship between a specific variable and the main parent
Dear Marek,
Is it possible to download DAG list into excel after learning the network?
Thanks
Ashwani
Is it possible to download DAG list into excel after learning the network?
Thanks
Ashwani
-
- Site Admin
- Posts: 432
- Joined: Tue Dec 11, 2007 4:24 pm
Re: Counting of relationship between a specific variable and the main parent
Actually, there is but it is a trick that has not really been designed for this purpose. It will work for discrete Bayesian networks. When the network is open, please invoke the "Strength of Influence" mode (either through a choice in the Network menu or by pressing a corresponding button in the toolbar). The small narrow "Influence" dialog that pops-up has a button that looks like a spreadsheet. When you press it, you will see a list of all arcs (actually parent-child pairs) with their strengths. You can copy this list and paste it later in Excel or an editor. This will give you all parent-child pairs, hence, all arcs.
I hope this helps,
Marek
I hope this helps,
Marek