Welcome, Guest. Please Login.
YaBB - Yet another Bulletin Board
09/05/10 at 11:32:39
News: Registration is required for posting a message. Please send a note to e l a d @ i e e e . o r g (remove spaces) to obtain a login.
Home Help Search Login


Pages: 1
Send Topic Print
Sequential feature selection params and targets (Read 1186 times)
tshifhie
YaBB Newbies
*


I irrelevant YaBB 2!

Posts: 4
Gender: female
Sequential feature selection params and targets
01/20/10 at 13:36:37
 
How do I use params in sequential feature selection, I have tried Sequential_Feature_Selection(patterns,targets,['Backward',2,'LS',[]]) it gives errors  in line 16 which is the line about process_params.
 
Does the target only use 0 and 1? How then do i represent my 15 indeces, because i would have liked to represent them in strings.
 
Iv only just started using both matlab and patterns so i wud also like to know,how do I pass my data to  the Sequential _Feature_ Selection function from another M_file or from the command window
 
Back to top
 
 
Email tshifhie   IP Logged
Elad Yom-Tov
YaBB Administrator
*****


There must be a
pattern somewhere
out there!

Posts: 14
Gender: male
Re: Sequential feature selection params and target
Reply #1 - 01/25/10 at 06:23:21
 
Hi,
 
The parameters should be like this:
Sequential_Feature_Selection(patterns,targets,'[''Backward'',2,''LS'',[]]')  
 
The targets should indeed be numerical, with 0/1 for binary classification.
 
Passing parameters is done by calling the function (you can do this from your other functions).
 
Regards,
 
Elad
Back to top
 
 
Email WWW   IP Logged
tshifhie
YaBB Newbies
*


I irrelevant YaBB 2!

Posts: 4
Gender: female
Re: Sequential feature selection params and target
Reply #2 - 01/27/10 at 09:34:56
 
I have tried the solution you suggested, when i enter  this statement Sequential_Feature_Selection(patterns,targets,'["Backward",2,"LS",[]]') there is a message saying parse error at string invalid matlab expression before i even compile
 
When i do compile it gives >> Sequential_Feature_Selection(patterns,targets,'["Backward",2,"LS",[]]')
??? Undefined function or method 'process_params' for input arguments of type
'char'.
 
Error in ==> Sequential_Feature_Selection at 15
[type, final_Ndims, Type, Params] = process_params(params);
 
I am using matlab R2009b
Back to top
 
 
Email tshifhie   IP Logged
Elad Yom-Tov
YaBB Administrator
*****


There must be a
pattern somewhere
out there!

Posts: 14
Gender: male
Re: Sequential feature selection params and target
Reply #3 - 02/07/10 at 12:48:47
 
Please copy and paste my code. You may have been using " instead of two '.
Back to top
 
 
Email WWW   IP Logged
tshifhie
YaBB Newbies
*


I irrelevant YaBB 2!

Posts: 4
Gender: female
Re: Sequential feature selection params and target
Reply #4 - 02/07/10 at 13:07:12
 
Thank you very much, it works. but it works on the clouds.mat data, please tell me how do i organise my own data in the same format as clouds.mat
 
I have 15 columns of features and 72 observations.
 
How is the data on clouds.mat organised,are the 5000 columns features or observations
Back to top
 
 
Email tshifhie   IP Logged
Elad Yom-Tov
YaBB Administrator
*****


There must be a
pattern somewhere
out there!

Posts: 14
Gender: male
Re: Sequential feature selection params and target
Reply #5 - 02/08/10 at 10:59:21
 
In clouds, there are 5000 observations and 2 features.
 
In your data, you'll need to have a patterns matrix of size 15x72, and a targets (labels) vector of size 1x72.
Back to top
 
 
Email WWW   IP Logged
tshifhie
YaBB Newbies
*


I irrelevant YaBB 2!

Posts: 4
Gender: female
Re: Sequential feature selection params and target
Reply #6 - 02/20/10 at 08:51:19
 
is it possible for me to use the sequential feature selection algorithm and not specify the dimension, because i would like the feature selection algorithm to select the optimum features without me specifing the demension number. I would like it to tell me the optimum feature
 
How do i use the feature selection algorithm with a classifier, I would like to use sequential feature selection with kmeans or with Genetic algorithm
Back to top
 
 
Email tshifhie   IP Logged
Pages: 1
Send Topic Print