Sequential Rule Mining - Now What?

btibert

After we apply the GSP operator to a proper dataset, it is not obvious to me what we can do with the results passed to the port pat, or, the GSPSet.  

Simply, how can we analyze the resulting rules for support, confidence, etc, after applying the GSP operator?  It's not obvious what operator would go next, and Create Association Rules is looking for a Frequent Itemset generated by FP Growth.

land

If you want to look at the patterns, you simply can connect the pat(tern) port to the result port of the process and take a look at the visualization.

If you want to post process them, there's a little trick necessary: You cannot directly access the table as data. In the Advanced Reporting Extension, there's an operator called "Extract Tables". It should be able to get the data from the visualization and convert it into a standard Example Set so that you can apply any operator that can process regular data.

Greetings,

Sebastian

MartinLiebig

Hi,

i usually just use it for feature generation in classification to find frequent sub-sets.

CCing @land who is i think the expert on the matter.

BR,

Martin

land

Hi,
it's really the question what you are trying to achieve. Could you give us some hints?

The answer will really differ, if you (as human) want to get insights into something or if you want to train a machine intelligenz to automate something.

Greetings,

 Sebastian

btibert

In other tools and languages, we use Sequence mining to look at how patterns emerge.  Unlike Association Rules which are looking at co-association of items, which are used to create rules with various metrics like support, confidence, lift, etc, I want to take this one step further and retain the sequential nature of the data while building rules.  In short, I see the operator, but it's not clear what operator(s) can be connected to the results from GSP. 

I appreciate the replies @mschmitz and @land , but I guess I lost as to what operators can be  used after GSP in RapidMiner.  

btibert

Thanks @land, I see a table of results (that somehow I overlooked) but no charts.