mac version and other questions

The front end.
Post Reply
liid
Posts: 15
Joined: Tue Jun 19, 2018 5:19 am

mac version and other questions

Post by liid » Tue Jun 19, 2018 9:45 am

Hi,

I am using the GeNIe academic (bayesian networks) and have some questions:

1. Is there a mac version for that? I am currently working with the windows version on a VM and it is slow and very inconvenient.

2. When I choose the "validate" option, and test my network on a data file, how can I see the ids of the data records that were not classified correctly? all I can currently see is the amount of records which were not classified correctly, but I can not see what records are they.

3. When I choose the "Learn parameters" option, I see 3 types of "parameters initialization" options for the cpt: uniform, random, and "keep original" (in other words, priors). The problem is that those options can only be applied to the entire network. What I need however is to apply different "parameter initialization" options to different nodes. For example, for one node I have a cpt designed by an expert, so for this node I need to choose the "keep original" option, while another node is a hidden node so for this node I need to choose the randomize option. How can I apply different init options for different nodes for the parameter learning process?

4. Technical problem: when I code my own cpts to the "definition" tab in the "node properties", and then apply the "learn parameters" process to the network, the original cpts that I inserted disappear because they are updated according to the training data. This is a problem as I need to re-insert my cpts every time that I want to train the network on different data and since the cpts of those nodes are long, coding them every time all over again is a very time consuming and annoying task. Can I somehow save the original cpts that appeared on the "definition" tab of the "node properties" of a node before I run the training ("Learn parameters") on the network?

5. What is the "update beliefs" option? every time I run the parameter learning I see a small question mark on lower-right side of the nodes in the network and when I hover on that question mark it says "not updated" (though the cpt does seem as if it was updated) and offers the update beliefs option. I could not find documentation about the meaning of that question mark or the "update beliefs" and I am concerned that this may somehow affect the cpts.

Thanks!

marek [BayesFusion]
Site Admin
Posts: 271
Joined: Tue Dec 11, 2007 4:24 pm

Re: mac version and other questions

Post by marek [BayesFusion] » Wed Jun 20, 2018 8:26 am

Hi Lihi,

It would have been better if you asked your questions individually -- the subject of your query would be helpful to other users. Some of your questions are answered in the manual and some are answered on the Forum -- please make sure to browse the manual and the Forum first. Let me try to answer your questions below.
1. Is there a mac version for that? I am currently working with the windows version on a VM and it is slow and very inconvenient.
GeNIe comes only in Windows version. I am surprised you are experiencing problems with speed -- we have quite a number of users running GeNIe on Mac and Linux under Wine. SMILE is, of course, fully portable and available for Mac among many other platforms.
2. When I choose the "validate" option, and test my network on a data file, how can I see the ids of the data records that were not classified correctly? all I can currently see is the amount of records which were not classified correctly, but I can not see what records are they.
You should generate an output file (at the bottom of the Validation dialog). It will contain the test records along with the posterior probability distributions of the class nodes and the model's decisions. You can subsequently open the file in GeNIe or Excel to study individual records.
3. When I choose the "Learn parameters" option, I see 3 types of "parameters initialization" options for the cpt: uniform, random, and "keep original" (in other words, priors). The problem is that those options can only be applied to the entire network. What I need however is to apply different "parameter initialization" options to different nodes. For example, for one node I have a cpt designed by an expert, so for this node I need to choose the "keep original" option, while another node is a hidden node so for this node I need to choose the randomize option. How can I apply different init options for different nodes for the parameter learning process?
What you want to do will be a little laborious. You can learn individual CPTs by selecting them in the Match Network and Data dialog and then selecting for each of them what you want. You will have to do it for each of the nodes that you want to use different parameters for. If your model contains many nodes that have to be treated exceptionally, perhaps you can program this using SMILE?
4. Technical problem: when I code my own cpts to the "definition" tab in the "node properties", and then apply the "learn parameters" process to the network, the original cpts that I inserted disappear because they are updated according to the training data. This is a problem as I need to re-insert my cpts every time that I want to train the network on different data and since the cpts of those nodes are long, coding them every time all over again is a very time consuming and annoying task. Can I somehow save the original cpts that appeared on the "definition" tab of the "node properties" of a node before I run the training ("Learn parameters") on the network?
There are several ways of accomplishing this task. One way would be to save the original network before you start learning and always start from the original network. Another is copying your CPT and pasting it somewhere in the node description. GeNIe allows you to select, copy, and paste CPTs and their fragments between GeNIe and any other program. Whenever you want to start from scratch, you just copy the original contents of the CPT from the description and paste it into the CPT. Finally, doing this from inside your program using SMILE may be the best option for you if what you are doing is complicated.
5. What is the "update beliefs" option? every time I run the parameter learning I see a small question mark on lower-right side of the nodes in the network and when I hover on that question mark it says "not updated" (though the cpt does seem as if it was updated) and offers the update beliefs option. I could not find documentation about the meaning of that question mark or the "update beliefs" and I am concerned that this may somehow affect the cpts.
The question mark is described in detail in the manual (see "Node status icons" subsection in the "Components of GeNIe models" section, chapter "Building blocks of GeNIe"). A question mark means that the node is not updated, i.e., its marginal posterior probability distribution is not available. To update the network, please press the thunderbolt icon. Alternatively, please set the "Update immediately" flag in the "Network menu".
I hope this helps.

Marek

Post Reply