Re: How to classify unseen multi instance data?

Am 06.02.24 um 22:58 schrieb Peter Reutemann:
> 
> RELAGGS uses SQL aggregation methods like COUNT, MIN, MAX, AVG, SUM,
> etc on the relational data to flatten it into a propositional, single
> table.
> A long time ago, when I was working with relational/multi-instance
> data, algorithms trained on the RELAGGS-flattened data tended to
> outperform other multi-instance algorithms that worked on the
> multi-instance data.

I have now tested the RELAGGS filter extensively and at first the 
results were encouraging. The models I created with different algorithms 
predicted decent accuracies, but when I used those models to actually 
classify unseen data, I found that they did not work at all. They either 
predicted "true" all the time or "false" all the time.

I tried J48, IBk and RandomForest as algorithms after applying the 
RELAGGS filter. Those same algorithms worked a lot better when used as a 
base classifier with SimpleMI on multi instance data (i. e. without 
applying the RELAGGS filter). In those cases the performance predicted 
in the training phase turned out to be rather similar to the actual 
performance achieved in validation.

So maybe the RELAGGS filter is not appropriate for my data (which is 
essentially timeseries sensor measurements).

_______________________________________________
Wekalist mailing list -- [email protected]
Send posts to [email protected]
To unsubscribe send an email to [email protected]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html