Cool. Thanks for sharing this. I will file a jira issue over this.

Robin



On Mon, Feb 15, 2010 at 9:52 PM, Neal Richter <[email protected]> wrote:

> I have no problem with the repetition!
>
> I'll have to poke at this a bit more, but I like the switches ideas.
> I often use Christian Borgelt's itemset implementations for playing
> with data.  He's implemented a nice set of switches, see below.
> Setting a minimum support threshold and mimimum itemset size are both
> convenient and tend to make the algorithm run a bit faster.
>
> http://www.borgelt.net/software.html
>
> ne...@nrichter-laptop:~$ fpgrowth_fim
> usage: fpgrowth_fim [options] infile outfile
> find frequent item sets with the fpgrowth algorithm
> version 1.13 (2008.05.02)        (c) 2004-2008   Christian Borgelt
> -m#      minimal number of items per item set (default: 1)
> -n#      maximal number of items per item set (default: no limit)
> -s#      minimal support of an item set (default: 10%)
>         (positive: percentage, negative: absolute number)
> -d#      minimal binary logarithm of support quotient (default: none)
> -p#      output format for the item set support (default: "%.1f")
> -a       print absolute support (number of transactions)
> -g       write output in scanable form (quote certain characters)
> -q#      sort items w.r.t. their frequency (default: -2)
>         (1: ascending, -1: descending, 0: do not sort,
>          2: ascending, -2: descending w.r.t. transaction size sum)
> -u       use alternative tree projection method
> -z       do not prune tree projections to bonsai
> -j       use quicksort to sort the transactions (default: heapsort)
> -i#      ignore records starting with a character in the given string
> -b/f/r#  blank characters, field and record separators
>         (default: " \t\r", " \t", "\n")
> infile   file to read transactions from
> outfile  file to write frequent item se
>
> On Mon, Feb 15, 2010 at 9:14 AM, Robin Anil <[email protected]> wrote:
> > Hi Neal,
> >             I know there is repetition. I tried sticking true to the
> > original algorithm that is finding closed patterns and using the longest
> > one.
> >
> > Say if 68 and 12 occurs 1000 times
> > and 68 12 17 also occurs 1000 times, there so information that former
> > pattern gives you. So, you can remove it. Therefore you say that 68 12 17
> is
> > a closed pattern and all the patterns it is enclosing are removed.
> >
> > had 68 alone occurred 2000 times. It no longer becomes a closed pattern..
> >
> > Things could be made configurable by having a flag to remove closed
> patterns
> > within a percentage of the support Or mine only patterns > 3 items in
> > length. These are tricky but could be done.
> >
> > Robin
> >
> >
> > On Mon, Feb 15, 2010 at 9:34 PM, Neal Richter <[email protected]>
> wrote:
> >
> >> Grant:  Chapter 5 of Han and Kamber (Data Mining: Concepts and
> >> Techniques) detail itemset mining and the fpgrowth alg.  Han is a
> >> co-inventor of it.
> >>
> >> There is a bit of repetition in the output compared to other itemset
> >> mining packages, though this structure is convenient for relational
> >> indexing by key.
> >>
> >> - Neal
> >>
> >> On Mon, Feb 15, 2010 at 6:49 AM, Robin Anil <[email protected]>
> wrote:
> >> > Ok.. A bit more background..
> >> >
> >> > An Itemset is a subset I1, I2, I3... In
> >> >
> >> > so [I2, I4, I7] is an itemset and the support(no of times its visible
> in
> >> the
> >> > dataset) is say Y
> >> >
> >> > A Pattern is Pair<Itemset, support>
> >> >
> >> > Take a look at in this format
> >> >
> >> > 68:
> >> >     ([68],90692),
> >> >     ([17, 68],90683),
> >> >     ([12, 68],90490),
> >> >     ([17, 12, 68],90481),
> >> >     ([18, 68],90291)
> >> >
> >> > these are top patterns containing 68 and their support in descending
> >> order
> >> > 68 occurs with 12,  90490 times
> >> >
> >> > Robin
> >> >
> >> >
> >> > On Mon, Feb 15, 2010 at 6:27 PM, Grant Ingersoll <[email protected]
> >> >wrote:
> >> >
> >> >>
> >> >> On Feb 14, 2010, at 11:37 PM, Robin Anil wrote:
> >> >>
> >> >> > Each key is a feature and each attribute is the topK frequent
> patterns
> >> >> where
> >> >> > the feature exist
> >> >>
> >> >> Still a bit confused.
> >> >> Given:
> >> >> Key: 68: Value: ([68],90692), ([17, 68],90683), ([12, 68],90490),
> ([17,
> >> 12,
> >> >> 68],90481), ([18, 68],90291), ([17, 18, 68],90282), ([12, 18,
> >> 68],90229),
> >> >> ([17, 12, 18, 68],90220), ([31, 68],89071), ([17, 31, 68],89062),
> ([12,
> >> 31,
> >> >> 68],88874), ([17, 12, 31, 68],88865), ([18, 31, 68],88681), ([17, 18,
> >> 31,
> >> >> 68],88672), ([12, 18, 31, 68],88619), ([17, 12, 18, 31, 68],88610),
> >> ([16,
> >> >> 68],87933),
> >> >>
> >> >> So, 68 is the feature in question.  That makes sense.  Then, what is
> the
> >> >> significance of the [] areas, as in [68],90692 or [17,12,68], 90481.
> >>  Why
> >> >> all the repetition?
> >> >>
> >> >> -Grant
> >> >
> >>
> >
>

Reply via email to