Missing Label after SVD Reduction

AnneGAnneG Member Posts: 8 Contributor II
edited November 2018 in Help
Hi there,

I really like the opportunity of SVDReduction to visualize the clusters, especially for comparing predefined classes to clusters. However, today I seem to have a problem and I hope you can help me on that: my data consits of a label and a few metric attributes. I just use Example source, KMeans and SVDResuction in line. Setting a breakpoint to KMeans, my label (values 1,2 or 3) is still to be seen in the example set. But in SVDReduction the label is still there but in the Meta Data View I see the following:
Statistics: mode = unknown; Mode: 3 (0), 2 (0), 1 (0); Unknown: 310.0
In the data view there is always "?" insead of the label's value. What have I done wrong? Do I need some kind of explicit index (like in the tutorial example) to keep a link between cluster and class?

Greetings,
Anne

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi Anne,
    are label and cluster attribute set to the special roles label and cluster? Normally these special attributes are kept as they are when the examples are transformed.

    Greetings,
      Sebastian
  • AnneGAnneG Member Posts: 8 Contributor II
    Hi Sebastian,

    in the meta data output there is my label (of type label), the cluster, an id and d0 and d1. I have done this before on other data and it worked perfectly - now I am a bit puzzled about why the access to the label values seems to be lost.

    Thanx so far,
    Anne
  • IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder
    Hi,

    I am not sure if I got you right: is the label only missing in the meta data or also in the real data? If it's only in the meta data, this could be a bug in the meta data transformation of the SVD. If it's also in the real data, I must admit that I have no idea what the reason might be without seeing the data. But maybe a workaround could help: Just add an additional ID (or use the existing one), multiply the data, cluster and reduce it and join the label back to the data set again.

    Cheers,
    Ingo
  • AnneGAnneG Member Posts: 8 Contributor II
    Hi,

    in my output and the 'card' ExampleSet of the SVDReduction, there is still a variable with the name of my class label (that is contained in the original data), it is of type nominal, the actual values are just gone. the data view in the SVDReduction is like that

    RowNo. | cluster     | myLabel | id |       d0 |     d1
              1 | cluster_0 |            ? | 1  | 0.089 | -0.04
              2 | cluster_1 |            ? | 2 | 0.04    | 0.03

    My original input is very simple, just the label (nominal) and a few real-value attributes.
    By the way, the values also get lost when I leave out the clustering and directly apply SVDReduction on the data.
    Thank you so far.
  • IngoRMIngoRM Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, Community Manager, RMResearcher, Member, University Professor Posts: 1,751 RM Founder
    I have split the topic from the KMeans discussion and moved it into this problems board. Maybe here somebody can help you.

    Cheers,
    Ingo
  • text_minertext_miner Member Posts: 11 Contributor II
    AnneG,

    I have previously run across this same issue and submitted a bug report (http://bugs.rapid-i.com/show_bug.cgi?id=307).

    As a workaround, try renaming your label attribute (named "myLabel") to "label".

    Hopefully this fixes the problem.
  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi,
    I have fixed this bug. It will become available in the repository tomorrow or will be delivered with the next update.

    Greetings,
      Sebastian
  • AnneGAnneG Member Posts: 8 Contributor II
    Thank you very much!
    Best regards,
    Anne
Sign In or Register to comment.