1
$\begingroup$

I generated a csv file with 167 attributes and around 5000 entries. One is a nominal attribute, two are dates and the rest is numerical.

I can import the file into the weka explorer without problems. But all clustering methods and most attribute selectors are greyed out. This still happens if I delete around 50 of my attributes. I only found ideas how to make it work with classifier methods but I need clustering or at least attribute selection. I used similar data before but with only 20 attributes and it worked, so I guess I need to further reduce the amount of attributes? But I'd really like to make some algorithm chose instead of me.

Can you advise me what I could change? Are there other tools that work with this many attributes? Or did I do a obvious mistake I am not finding?

$\endgroup$
0
$\begingroup$

Many clustering algorithms only work with continuous values.

Therefore, they are greyed out until you convert all your attributes to numeric. But beware that the methods will still assume e.g. a similar scale. So mapping a date and a count to comparable values may simply be impossible.

Don't blindly use these algorithms as black box methods. This will produce useless / meaningless results. Instead, your job is to understand the algorithms and identify the way to transform the data (and often: how to modify the algorithm) that the result has desirable properties for your problem. You don't gain anything if the algorithm solves a different problem...

$\endgroup$
  • $\begingroup$ I understand what you are saying and I agree. But I don't think that this is the problem in my case. I understand the algorithms I want to work with (in this case I figured DBScan should be the most promising) and I know that - at least in theory - my features are fitting to be used with it. I still feel it is more guessing than knowing because I lack the experience but I have an idea of what to do. I am just dumbfounded because I wasn't able to use ANY algorithm at all and couldn't figure out why. Turns out, I screwed up in the export and had some badly formatted numbers. Thanks though :) $\endgroup$ – Anita219 Jun 8 '17 at 8:29
  • $\begingroup$ Weka DBSCAN is really slow. Use ELKI DBSCAN instead. $\endgroup$ – Anony-Mousse Jun 8 '17 at 18:39
0
$\begingroup$

For future references: Turns out I had a poorly formatted csv file because I made some changes in it through excel and it decided to place additional points in my numbers. Generated a new export and now it works.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.