When running discriminate.pl (especially with cluster stopping) you may notice an error message of the following form:
free(): invalid pointer 0x816d1d8! As horrible as this message looks, it does not actually have any significant effect on processing. This is an error from Cluto's vcluster or scluster program that is caused when the --showtree option is on when there is just 1 cluster discovered. This --showtree option causes an ascii dendogram tree to be generated, and apparently cluto has some trouble with this for the 1 cluster case. We have reported this error to the Cluto folks, and for now have removed --showtree option from our invocations of scluster and vcluster when there is only 1 cluster. Note that I say this is something you will typically see when using cluster stopping. The reason for this is not any flaw in cluster stopping, but rather the fact that it is most likely when you specify the number of clusters manually (not using cluster stopping) you will not specify 1. Automatic cluster stopping will sometimes find 1 cluster, which is when that error message is generated. It also generates a core dump, which is also a scary thing to see. Below is an example of what you might see... discriminate.pl --showargs --crfun i2 --verbose --space vector --clmethod direct --token /home/cs2/tpederse/SenseClusters-v0.93/Demos/token.regex --target /home/cs2/tpederse/SenseClusters-v0.93/Demos/target.regex --prefix line-n --context o1 --feature bi --remove 5 --window 2 --stop /home/cs2/tpederse/SenseClusters-v0.93/Demos/stoplist.new --cluststop gap --stat ll --stat_score 3.841 line-n-test.xml Preprocessing the input data ... Computing Bigram Counts ... Performing Statistics on Word Pairs ... Finding Feature Regex/s ... Building 1st Order Context Vectors ... Clustering in Vector Space ... Finding number of clusters ... free(): invalid pointer 0x816d1d8! Creating Cluster Labels ... ================================================================= Output when #clusters = 1 (GAP measure) ================================================================= Clusters of given contexts can be found in file: line-n.clusters.gap ======================================================================== ======================================================================== The specific invocation of vcluster (which was created and submitted by discriminate.pl in this case) that caused this error was: vcluster --showtree --rlabelfile line-n.rlabel --clmethod direct --crfun i2 --nfeatures 10 --clabel line-n.clabel --rowmodel none --colmodel none --clustfile line-n.cluster_solution.gap --plotclusters line-n.gap.dendogram.ps --plotformat ps line-n.vectors 1 In future it will run like this: vcluster --rlabelfile line-n.rlabel --clmethod direct --crfun i2 --nfeatures 10 --clabel line-n.clabel --rowmodel none --colmodel none --clustfile line-n.cluster_solution.gap line-n.vectors 1 Note that in the case of 1 cluster we do not generate a dendogram plot, as Cluto does not support that. We do however generate dendogram plots for 2 or more clusters (and we will also continue to provide -showtree output in those cases too.) Thanks! Ted -- Ted Pedersen http://www.d.umn.edu/~tpederse ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys -- and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ senseclusters-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/senseclusters-users
