mining closed sequential patterns

Ask any questions, discuss or post suggestions for SPMF
Post Reply
Ndrez
Posts: 4
Joined: Fri Apr 29, 2022 5:15 pm

mining closed sequential patterns

Post by Ndrez »

Good morning,
for a project I need to mine closed sequential patterns, so I tested all the algorithms provided in the SPMF library (i.e., ClaSP, CM-ClaSP, CloFAST, CloSpan, BIDE+). Since all these algorithms mine closed sequential patterns, why does BIDE+ report a different number of closed sequential patterns with respect to the other algorithms? (of course using the same data and same parameters).

Thank you in advance.
Andrea
User avatar
admin
Site Admin
Posts: 121
Joined: Tue Apr 05, 2022 12:47 am
Location: China
Contact:

Re: mining closed sequential patterns

Post by admin »

Dear Andrea,

Thanks for using SPMF.

As noted in the download page of SPMF under "known limitations", there is a bug in the BIDE+ implementation, such that the result is incorrect especially in the case where sequences contain more than one item per itemset. I have spent a week to investigate this problem in BIDE+ and try to fix it a while ago, but it is a very complex algorithm so i finally decided to give up on fixing BIDE+. I currently keep BIDE+ in SPMF just for reference while mentioning the limitation on the website, but I have been thinking of removing BIDE+ from SPMF for a long time.

The other closed sequential pattern mining algorithms in SPMF are all correct and have been tested extensively.

If I can answer any other questions, feel free to ask in the forum. :-)

Best regards,
Ndrez
Posts: 4
Joined: Fri Apr 29, 2022 5:15 pm

Re: mining closed sequential patterns

Post by Ndrez »

Thank you for your answer.
Since I did not find the "known limitations" section in the download page, are there any known bugs for the VMSP algorithm? It returned me some sequential patterns (called here A,B,C) such that B and C are subsequences of A (and this already violates the definition of maximal sequential patterns if I am not wrong), but A has a larger support w.r.t. the ones of B and C (that it is not possible since B and C are subsequences of A). Since I am postprocessing the patterns after the mining, it is possible that the bug is from my side, I am just asking to check if there are already known issues.
Thanks,
Andrea
Ndrez
Posts: 4
Joined: Fri Apr 29, 2022 5:15 pm

Re: mining closed sequential patterns

Post by Ndrez »

Directly checking the output of VMSP, I think that the issue is in the algorithm.
I got this output (beside other patterns):
1 -1 2 -1 3 -1 4 -1 5 -1 6 -1 #SUP: 12
1 -1 3 -1 4 -1 5 -1 6 -1 #SUP: 11
1 -1 2 -1 4 -1 5 -1 6 -1 #SUP: 11
1 -1 2 -1 3 -1 5 -1 6 -1 #SUP: 11
If it can help, I am using the "maxGap" and "show sequences ids?" parameters.
Post Reply