Author Topic: help!!!  (Read 23750 times)

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
help!!!
« on: March 15, 2011, 04:12:56 »
hi,
when  i use learning wizard get a  BN model,How to evaluate the model is a good one?
thank you!
« Last Edit: March 15, 2011, 04:23:15 by znjsdu »

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: help!!!
« Reply #1 on: March 15, 2011, 16:37:42 »
The best way to evaluate a model depends on the task you are trying to solve (e.g., classification versus prediction).

The Learning Wizard produce the AIC, BIC, and log-likelihood scores of the model constructed. This information can be found in the log. Also, the Analysis Wizard has different measures for evaluating a model.
HUGIN EXPERT A/S

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
Re: help!!!
« Reply #2 on: March 17, 2011, 02:54:24 »
thank you for your response!
 :)
i want to use  the model to infer,which is the best way to evaluate  and how to verify the model?
looking forward your response!
thank you in advance!
« Last Edit: March 17, 2011, 03:00:01 by znjsdu »

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: help!!!
« Reply #3 on: March 18, 2011, 11:03:34 »
This thread discusses model evaluation is some detail: http://forum.hugin.com/index.php?topic=116.0

HUGIN EXPERT A/S

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
Re: help!!!
« Reply #4 on: March 18, 2011, 13:40:00 »
thank you very much!
I would like to ask you:
How to perform cross-validation  with hugin researcher 7.0?
Thank you for your time!

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: help!!!
« Reply #5 on: March 18, 2011, 14:14:58 »
To perform cross validation open "Wizards->Analysis Wizard" and select the "Data Accuracy" tab.
HUGIN EXPERT A/S

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
Re: help!!!
« Reply #6 on: April 17, 2011, 04:11:25 »
To perform cross validation open "Wizards->Analysis Wizard" and select the "Data Accuracy" tab.
when i perform cross validation,should i use all data that i use model the BBN  or use some of them?
How i can do Groups outside prediction?
« Last Edit: April 17, 2011, 04:32:02 by znjsdu »

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: help!!!
« Reply #7 on: April 18, 2011, 09:47:48 »
Quote
when i perform cross validation,should i use all data that i use model the BBN  or use some of them?

You learn the model on one subset of your data and test it on a different subset of the data (a subset which has not been used to build the model)

Quote
How i can do Groups outside prediction
I'm not sure what you are referring to here? The tool does not support splitting up data into learn and test sets. It is assumed that you have already split the data
HUGIN EXPERT A/S

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
Re: help!!!
« Reply #8 on: April 18, 2011, 13:45:58 »
You learn the model on one subset of your data and test it on a different subset of the data (a subset which has not been used to build the model)
I'm not sure what you are referring to here? The tool does not support splitting up data into learn and test sets. It is assumed that you have already split the data
thank you very much!
I am confused!you said that the hugin doesn't support splitting up data into learn and test sets,but how i can perform the cross validation!
thank you for your time,and looking forward your answer!
« Last Edit: April 18, 2011, 14:15:50 by znjsdu »

Offline Anders L Madsen

  • HUGIN Expert
  • Hero Member
  • *****
  • Posts: 2295
    • View Profile
Re: help!!!
« Reply #9 on: April 19, 2011, 10:01:13 »
Quote
I am confused!you said that the hugin doesn't support splitting up data into learn and test sets,but how i can perform the cross validation!

If you have 10000 cases, then you could, for instance, use 8000 cases for learning and 2000 cases for testing.

If there is no information in order of the cases in the data file, then you can store the first 8000 for learning in one file and the last 2000 for testing in a separate file.

If there may be information in the order of cases (for instance, if they are recorded over time), then you may first want to randomize the data cases (i.e., change the order at random) and then use the first 8000 as learning data and the last 2000 as testing data.

You have to create two separate files. One for learning and one for testing. This is done outside HUGIN.

HUGIN EXPERT A/S

Offline znjsdu

  • Newbie
  • *
  • Posts: 14
    • View Profile
Re: help!!!
« Reply #10 on: April 20, 2011, 02:38:16 »
Thank you for your time!