validation,inference,learn

The front end.
Post Reply
safaa
Posts: 31
Joined: Sat May 24, 2014 7:00 am

validation,inference,learn

Post by safaa »

hello!
1. if we want to make parameter learning again. how to delete the probabilities for all nodes.
2. if we make inference in which the target node is predicted then this is validation? is that right?
3. cross validation is for managing the validation process. inference algorithm is used to predict? why using EM algorithm would you pleas give me reference to understand?
4. when we make network learning and parameter learning we are make the prior probabilities, when making inference we are updating the network then the is the posterior? if that so does the network keep the inference analysis to update itself. (when we choose update network)
5. copy cases is for inference (in older posts you say we can us jsmile to loop on all cases),if so when will we use it?we are using validation for predictions? apologize me if my questions seems not necessary. but I want to understand
6.best score in iteration 9 is = -40834.7 is it good network ?
7. when making network learning . what is the probabilities that exist in definition tab in node properties dialog.(keep in mind that we haven't made parameter learning yet!)

thank you for your cooperation.
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: validation,inference,learn

Post by shooltz[BayesFusion] »

safaa wrote:1. if we want to make parameter learning again. how to delete the probabilities for all nodes.
You don't need to "delete" the parameters. In the EM settings dialog you have the options to randomize or uniformize the parameters before learning begins - this effectively wipes out the existing parameters. You can also set equivalent sample size to zero, which assigns zero weight to existing parameters.
2. if we make inference in which the target node is predicted then this is validation? is that right?
To validate, you'll have to compare the network prediction to some data.

3. cross validation is for managing the validation process. inference algorithm is used to predict? why using EM algorithm would you pleas give me reference to understand?
http://en.wikipedia.org/wiki/Cross-vali ... tatistics)
4. when we make network learning and parameter learning we are make the prior probabilities, when making inference we are updating the network then the is the posterior? if that so does the network keep the inference analysis to update itself. (when we choose update network)
When you invoke the "Update" command in GeNIe you're calculating the posteriors. These depend on the (learned or manually created) network structure and parameters and current evidence set. The "Update" command does not change the network parameters.

5. copy cases is for inference (in older posts you say we can us jsmile to loop on all cases),if so when will we use it?we are using validation for predictions? apologize me if my questions seems not necessary. but I want to understand
The "Copy Cases" command in the "Data" menu is indirectly related to inference. It allows you to copy the data from external dataset (csv or gdat file, for example) into the .xdsl file. You can use the Case Manager pane later to perform inference by instantiating a case (setting the evidence defined in the case). You can also learn parameters using case-based EM. Other than source of input data, there's no difference between case-based and external file-based EM.

6.best score in iteration 9 is = -40834.7 is it good network ?
I assume the "best score" you're referring comes from Bayesian Search. This measure can be used to compare Bayesian Search runs over the same data set. Its value alone is not a indicator of the quality of learned network structure.

7. when making network learning . what is the probabilities that exist in definition tab in node properties dialog.(keep in mind that we haven't made parameter learning yet!)
Bayesian Search and augmented Naive Bayes variants automatically call EM when network structure is learned. For PC learning you have an option to follow with EM or use the uniform distributions.
safaa
Posts: 31
Joined: Sat May 24, 2014 7:00 am

Re: validation,inference,learn

Post by safaa »

how validation is done using EM algorithm? it it type of inference?
shooltz[BayesFusion]
Site Admin
Posts: 1417
Joined: Mon Nov 26, 2007 5:51 pm

Re: validation,inference,learn

Post by shooltz[BayesFusion] »

safaa wrote:how validation is done using EM algorithm? it it type of inference?
EM will be used if you select K-fold or Leave one out crossvalidation. Did you read the Wikipedia article I linked in previous post?
safaa
Posts: 31
Joined: Sat May 24, 2014 7:00 am

Re: validation,inference,learn

Post by safaa »

okay. thank you
Post Reply