The output I get looks very weird to me

Trying prefixspan-0.5.2 with python 3.6.8

I am running prefixspan-cli frequent 5 --closed ids.txt > seqs.ids

ids.txt: https://gist.github.com/johann-petrak/9d07e3bacd167639c26defb822dbe6aa seqs.ids: https://gist.github.com/johann-petrak/7db1b94153816075556798db1d068069

If you look at the line 4 from the bottom, this is “34 1 0 2 : 6” and the next line is “34 0 2 : 8”.

If you look at the input you will notice that “34 0 2” does not occur anywhere in the input, so why is it included in the output with frequency 8?

[LoLei]

@johann-petrak I suggest you read this survey about sequential pattern mining [Fournier-Viger et al., 2017]. The introduction describes what sequential patterns are in general, and why gap constraints are not universally needed, and there are also short descriptions of the various algorithms and their parameters.

Also the papers of PrefixSpan [Pei et al., 2004], BIDE [Wang et al., 2007] and FEAT [Gao et al., 2008] are of course interesting.

This repository merely provides efficient implementations of these algorithms.

Adding gap constraints arbitrarily to an existing algorithm is not trivial, as the possibility heavily depends on how that specific algorithm works. For some, gap-constraint versions have been proposed after the original algorithm was published, e.g. Gap-BIDE [Li and Wang, 2008].

[johann-petrak]

My expectation would have been that unless the size and number of gaps are specified, no gaps would be allowed. In many situations one just wants proper subsequences without gaps, in other situations e.g. subsequences with at most 1 gap of at most 2 elements etc. But this algorithm does not seem to have a command line option to control this and I did also not see a mention about what the default settings are.

But clearly id does not make much sense in many applications to allow any number of gaps of any size.