Partition of genre pieces into segments

Questions, answers, discussions related to ISMIS 2011 Contest

Partition of genre pieces into segments

Postby dslate » Tue Feb 22, 2011 10:28 am

The description of the genre data says that "All music pieces are partitioned into 20 segments and parameterized." However, the numbers of records in the training data for the 6 genres do not seem to be multiples of 20. Does that mean that not all segments for each piece are present in the data?

Please clarify,

Thanks,

-- dslate
dslate
 
Posts: 15
Joined: Mon Jul 05, 2010 4:35 am

Re: Partition of genre pieces into segments

Postby Marcin » Mon Mar 07, 2011 1:53 pm

All segments are present in the data.

Regards,
Marcin Wojnarski
Marcin
 
Posts: 115
Joined: Fri Oct 09, 2009 6:45 pm

Re: Partition of genre pieces into segments

Postby jw81 » Sun Mar 20, 2011 7:46 am

So then the statement "All music pieces are partitioned into [exactly] 20 segments and parameterized" is not correct?

What should should it say?
jw81
 
Posts: 1
Joined: Mon Mar 14, 2011 7:53 pm

Re: Partition of genre pieces into segments

Postby Marcin » Tue Mar 22, 2011 9:47 pm

Both are correct:
* the pieces were partitioned into 20 segments, and ...
* the number of records is NOT a multiplicity of 20.
;)
I'm sure some guys here can explain this miracle... Anybody to help?
Marcin
 
Posts: 115
Joined: Fri Oct 09, 2009 6:45 pm

Re: Partition of genre pieces into segments

Postby wahoo » Thu Mar 24, 2011 3:18 am

I remember finding duplicates and wondering if any other surprises were lurking in data that weren't described in the overview!
wahoo
 
Posts: 3
Joined: Thu Jan 07, 2010 1:15 am

Re: Partition of genre pieces into segments

Postby Marcin » Thu Mar 24, 2011 7:03 pm

Yep, there were duplicates. This wasn't intended, initially it was just a bug in data generation process. We discovered it shortly before contest launch and wanted to fix the data, but later on we decided to leave it as it is, because in a real-world setting the data are never clean, rather opposite - they contain all different kinds of impurities and without laborious investigation and cleansing one have no chances to get any close to optimal accuracy. So, this was a kind of an exercise. :) Do you know if duplicates had any influence on model building and results?

As to other surprises - we'd be surprised ourselves if there are any. Or maybe you found something?

-m
Marcin
 
Posts: 115
Joined: Fri Oct 09, 2009 6:45 pm


Return to ISMIS 2011 Contest: Music Information Retrieval

Who is online

Users browsing this forum: No registered users and 1 guest

cron