Ren Li-An | 1 Nov 14:32 2002
Picon

RE: How apply a model to a test file?

>optionally save result of classification to MyResults.txt
java weka.classifiers.trees.j48.J48 -T MyInstance.arff -l MyTree.tree -p
0 > MyResults.txt
 
Thank you. But how can I output the results to a file? Maybe need other parameters?
 
I use weka-3-3-4, it doesn't work with "> MyResults.txt", and can I use the explorer to do the same task?
 
 
Best regards,
 
alex

	-----Original Message----- 
	From: Olumide [mailto:50295 <at> web.de] 
	Sent: 2002-11-1 (ζ˜ŸζœŸδΊ”) 6:45 
	To: Ren Li-An 
	Cc: wekalist <at> list.scms.waikato.ac.nz 
	Subject: Re: [Wekalist] How apply a model to a test file?
	
	

	>
	>
	>I have gotten the model by training a classifier, but how can I get the prediction results of a test file
which has not any target?
	>
	What I do - and I don't know if there is a better way -  is that I
	create another ARFF file with the data I wish to classify BUT with a
	"dummy" classification and then I pass this to the trained classifier e.g:
(Continue reading)

Olumide | 1 Nov 16:03 2002
Picon

Re: How apply a model to a test file?

java weka.classifiers.trees.j48.J48 -T MyInstance.arff -l MyTree.tree -p
0 > MyResults.txt

Thank you. But how can I output the results to a file? Maybe need other parameters?

I use weka-3-3-4, it doesn't work with "> MyResults.txt", and can I use the explorer to do the same task?

It should! If you are working on the DOS/Windows environment! Or are you working in the CLI?

- Olumide
Olumide | 1 Nov 16:34 2002
Picon

Re: How apply a model to a test file?

> It should! If you are working on the DOS/Windows environment! Or are 
> you working in the CLI?

I just checked. The redirect ">" does not work in the CLI version of 
Weka. You need to use the DOS prompt (if you are in Windows) or the 
Shell (if you are in UNIX/Linux). To learn how to do so (in case you do 
not know how) point your browser to the URL:
  http://www.cs.waikato.ac.nz/~ml/weka/tips_and_tricks.html

NOTE: You need to set the CLASSPATH environment variable

- Olumide
Saket Joshi | 2 Nov 22:57 2002

attribute selection

Hi Everyone,

I have a quick question.
I have an instances object that contains about 5000 instances and about a
hundred attributes (0 to 99). When I train my classifier on these
instances, I want it to ignore attributes 40 to 70 and only train on
attributes 0 to 39 and 71 to 99. One of the attributes from 40 to 70 will
be the class. How can I do this without deleting the attributes to be
ignored from the instances object?

Thanks in advance,
-Saket.
Olumide | 3 Nov 00:20 2002
Picon

Re: attribute selection

>
>
>  I want it to ignore attributes 40 to 70 and only train on
>attributes 0 to 39 and 71 to 99. One of the attributes from 40 to 70 will
>be the class. How can I do this without deleting the attributes to be
>ignored from the instances object?
>
In the Weka Explorer do one of the following:

- (De)Select the numbered check boxes for the attributes you wish to use 
or ignore
- The Simpler way: Use the AttributeFilter 
(weka.filters.AttributeFilter) in the Filters section of the Preprocess 
(default) Panel/Tab of Explorer. You can use attribute indices to select 
individual or a range of attributes and inverse selection to make things 
easier. (Thereafter you will have to:
        - Press the Add Button to add the filter to a list
        - Press the Apply Filters Button to run the filter
        - Press the Replace button to make the filtered data the active 
dataset in Weka
        - Alternatively you can save the filtered data to new file by 
pressing the save button ...

You can find all these buttons on the Top-Right corner of the Preprocess 
Panel/Tab (which is the default view) of the Weka Explorer.

In conclusion: as you know, all the above can be done in the CLI as well ...

Have Fun ... (I see I'm not the only one "at it" late on a saturday 
night ... :-)  )

- Olumide
Mithun Prasad | 3 Nov 13:00 2002
Picon

dataset

Hi,
    Is there a way to flag certain instances in the dataset by changing weka source-code...???

Cheers,
Mithun.


Surf the Web without missing calls! Get MSN Broadband. Click Here
Mithun Prasad | 3 Nov 13:41 2002
Picon

dataset

Hi,
    Is there a way to flag certain instances in the dataset by changing weka source-code...???

Cheers,
Mithun.


Unlimited Internet access -- and 2 months free!  Try MSN. Click Here
Mithun Prasad | 3 Nov 15:36 2002
Picon

Ignore attributes - problem

Hi,
     I was wondering if its possible to access the ignored attribute in the Instance.java code...
I have set 4th attribute as 'ignore attribute' in the weka gui... However, in my code I have,

System.out.println("Class index: " + toCluster.value(3));
where toCluster is of type Instance.

but I get the error message... :
java.lang.ArrayIndexOutOfBoundsException
        at weka.core.Instance.value(Unknown Source)

Cheers,
Mithun.

Surf the Web without missing calls! Get MSN Broadband. Click Here
gabi | 3 Nov 21:59 2002
Picon
Picon

Re: dataset

Your question is not quite clear to me. Should some instances be treated differently to others or should some be left out?
gabi

Mithun Prasad wrote:

 

Hi,
    Is there a way to flag certain instances in the dataset by changing weka source-code...???

Cheers,
Mithun.


Unlimited Internet access -- and 2 months free!  Try MSN. Click Here _______________________________________________ Wekalist mailing list Wekalist <at> list.scms.waikato.ac.nz http://list.scms.waikato.ac.nz/mailman/listinfo/wekalist
Shane Butler | 3 Nov 22:55 2002
Picon

FW: threshold values

Forwarding this on.  Please reply to conugent <at> cs.tcd.ie as Conor may not
be on the list.

-----Original Message-----
From: conugent <at> cs.tcd.ie [mailto:conugent <at> cs.tcd.ie] 
Sent: Monday, 4 November 2002 4:55 AM
To: sbutle <at> deakin.edu.au
Subject: threshold values

Hello Mr. Butler,

My name is Conor and I've just started as a research student at Trinity
College Dublin. I'm using Weka as part of some research that I'm doing.
In particular I've been using your neural net software and I have found
it extremely useful. I was just wondering if  there already existed an
easy way to set a threshold value on a regression type output node.
Basically the regression output would be converted to a nominal class
based on the threshold. We're interested in experimenting with different
threshold values to see how they might effect the true_positive,
false_negative (etc) rates in a classification problem we are studying.

Any help you could give me would be greatly appreciated

Conor

Gmane