From: Chris Y. <Chr...@io...> - 2011-10-23 13:39:36
|
Hi Alexander, There are a number of ways to determine variable importance, either directly or indirectly. i) Some algorithms lend themselves to direct interpretations of variable importance by examining the log output or model file. For example, the pca based algorithms such as ENFA or CSM you can examine the variable loadings for each component and the component weightings. In OM v 1.2 these values are part of the standard log output for the ENFA algorithm. I'm sure similar things can be done with the model files to extract variable weightings for other algorithms, although I'm not familiar with all of them. ii) Indirectly we can examine variable importance by a jackknife/bootstrap procedure. This is implemented in om-desktop as a pre-analysis tool (although its always been a bit flakey for me). If you start with N variables, this runs your analysis N times but each time with N-1 variables to determine the model performance (i.e. AUC) without each variable in turn. Additionally you can run N models with just a single input variable each time to see the performance with a single variable (not possible with every algorithm). iii) Another indirect way to see the impact of a variable is to do variable response curves. Select some random points (say 1000) drill these into your projected model and the original variable layer. Doing a x,y plot of variable against model score will show how the model responds to that variable. If its a flat line then you can say that this variable doesn't appear to influence the model (at least in isolation). I've promised someone that I'd write a script to manage a jackknife analysis (ii) from the command line. I'd be happy to pass this on when I've written it. I hope this is helpful. - Chris Dr Chris Yesson Postdoctoral Researcher Institute of Zoology Zoological Society of London Regent's Park London NW1 4RY 020 7449 6267 http://www.zsl.org/chrisyesson/ http://eu-fp7-coralfish.net/ -----Original Message----- From: Alexander Bruy [mailto:ale...@gm...] Sent: Sun 2011-10-23 10:44 AM To: openModeller Subject: [om-devel] How to find which layers are most important Hi all, is it possible to find which layers from input layers set are most important for model creation? For example, is some Windows programs after generating model I can see which layers have greater weight in determining the likehood of occurence. Is it possible to get such information with openModeller? I use console tools, not openModeller Desktop Thanks -- Alexander Bruy ------------------------------------------------------------------------------ The demand for IT networking professionals continues to grow, and the demand for specialized networking skills is growing even more rapidly. Take a complimentary Learning@Cisco Self-Assessment and learn about Cisco certifications, training, and career opportunities. http://p.sf.net/sfu/cisco-dev2dev _______________________________________________ openmodeller-devel mailing list ope...@li... https://lists.sourceforge.net/lists/listinfo/openmodeller-devel This message has been scanned for viruses by MailControl - www.mailcontrol.com Click https://www.mailcontrol.com/sr/wQw0zmjPoHdJTZGyOCrrhg== to report this email as spam. The Zoological Society of London is incorporated by Royal Charter Principal Office England. Company Number RC000749 Registered address: Regent's Park, London, England NW1 4RY Registered Charity in England and Wales no. 208728 _________________________________________________________________________ This e-mail has been sent in confidence to the named addressee(s). If you are not the intended recipient, you must not disclose or distribute it in any form, and you are asked to contact the sender immediately. Views or opinions expressed in this communication may not be those of The Zoological Society of London and, therefore, The Zoological Society of London does not accept legal responsibility for the contents of this message. The recipient(s) must be aware that e-mail is not a secure communication medium and that the contents of this mail may have been altered by a third party in transit. If you have any issues regarding this mail please contact: adm...@zs.... ___________________________________________________________________________ This message has been scanned for viruses by MailControl, a service from BlackSpider Technologies. |