WEKA is dealing with string values like integer values!!

[farhad323]

Hi,

I cannot use join statement of DB2 tables in WEKA, so instead of using direct db2 query I exported the result to CSV file. I used this CSV file for clustering
The problem is that there are some fields of CHAR data type but they are basically numeric values. I want WEKA to deal with them like nominal values.
I added a CHAR flag to those fields to force WEKA to deal with them like nominal values. For example ‘123456’ converted to ‘a123456’. I guess there is a better and technical solution for this.
How can I force WEKA to deal with those fields like nominal values?

Thanks

[Mark]

From the command-line help:

java -cp /Users/mhall/Documents/Pentaho/dev/myDev/wekasvn/branches/stable-3-6/weka/build/classes/ weka.core.converters.CSVLoader

Usage:
CSVLoader [options] <file.csv>

Options:

-N <range>
The range of attributes to force type to be NOMINAL.
'first' and 'last' are accepted as well.
Examples: "first-last", "1,4,5-27,50-last"
(default: -none-)
-S <range>
The range of attribute to force type to be STRING.
'first' and 'last' are accepted as well.
Examples: "first-last", "1,4,5-27,50-last"
(default: -none-)
-D <range>
The range of attribute to force type to be DATE.
'first' and 'last' are accepted as well.
Examples: "first-last", "1,4,5-27,50-last"
(default: -none-)
-format <date format>
The date formatting string to use to parse date values.
(default: "yyyy-MM-dd'T'HH:mm:ss")
-M <str>
The string representing a missing value.
(default: ?)

In the Explorer (Weka 3.7 branch) there is a checkbox "Invoke options dialog" when you bring up a open file dialog that gives you access to these additional parameters.

Cheers,
Mark.