Options

'Read ARFF' import operator does not pass data

katulkatul Member Posts: 3 Contributor I
edited November 2018 in Help
I want to import a Weka file using the 'Read ARFF' operator. Its 'out' outlet is connected to the 'exa' socket of a 'Set Role' operator. Unfortunately, no data arrives that 'Set Role' operator, neither before nor after starting the process once (no matter which breakpoints).

The large "tooltip" box which appears on hovering the 'Read ARFF's 'out' output label, shows, *before* running the process: 

Read ARFF.output (output)
Meta data: Data Table
Number of examples = 0
0 attributes:
Generated by: Read ARFF.output
After running the process, the box shows 

Read ARFF.output (output)
Meta data: Data Table
Number of examples = 0
0 attributes:
Generated by: Read ARFF.output Data: Simple
ExampleSet: 150 examples, 5 regular attributes, no special attributes
So, after running the process, there are the 5 attributes acknowledged. Still, the table in the lower part of the box remains empty, and no example data or attributes arrive at the 'Set Role' op. If I set 'SetRole's 'name' parameter manually ('class' -> 'label'), it will say that the 'class' attribute does not exist.

It seems to me that I am doing it the intended way, and it works with the 'Retrieve Iris' op, but not with the 'Read ARFF' import.

Grateful for help.

Answers

  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    Please note that the problems that are displayed under the process view are labelled as *potential* problems. Did you try to actually run the process after setting the name attribute by hitting the blue play button? Just give it a try, most probably it will work.

    Best regards,
    Marius
  • Options
    katulkatul Member Posts: 3 Contributor I
    Thanks, Marius, for the reply.

    In fact, I did run the process — also with all the possible breakpoint configurations.
    The result is what is shown in the lower Code-Box in my posting, where the 'ExampleSet' is shown in the text, but not in the table (I could provide screenshots if this is not clear). Also, it is not transported to the next operator.

    My .arff file is accepted by the Weka Explorer.
    In RapidMiner, it _does_ work with the 'Read CSV' operator (after stripping off the ARFF header), but only if I use the wizard.

    In case there are important paths and variables: I started Rapidminer both with 'java -jar' and the script 'RapidMinerGUI' under MacOS Mountain Lion.

    Best wishes,
    Oliver
  • Options
    MikolajMikolaj Member Posts: 2 Contributor I
    I have exactly the same problem. The ReadARFF operator reads the file correctly, but does not pass the meta-data to the output. The result is that no other operator can be applied in the workflow that selects any subset of attributes, because the list of available operators is not passed through.
    If I store the ARFF file using the Store operator and then retrieve it in the binary RM format from the repository, the meta-data is stored and retrieved correctly.

    Ubuntu 12.10, RM 5.3.005

      Best,
      Mikolaj
  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    If the problem is only the metadata, you can type in the attribute names manually in the subsequent attribute selection etc. operators.

    Or you use the approac posted by Mikolaj above.

    Best regards, Marius
  • Options
    katulkatul Member Posts: 3 Contributor I
    Sorry Marius,

    both Mikolaj and I report the same problem and have no workaround. It may well be a bug, too.

    Again, the problem is: The 'read ...' operators do not pass data on to other operators. No metadata problem. Two different platforms.

    It is cool that you offer free support, but frankly, misleading answers are just a waste of both your and our time.

    Best,
    Oliver
  • Options
    MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869 Unicorn
    Oliver,

    Mikolaj explicitly mentioned that the *metadata* is not passed correctly, and the tooltips that you mentioned also contain only the metadata by design. So please let's keep this discussion factual.

    I just tested the Read ARFF operator, and it works as expected, with the only restriction that it does not produce any metadata. Maybe there is a problem with the combination of your input data and the Read ARFF operator, so to test that, it would be great if you could post your data and your process setup.

    Furthermore, you say:
    So, after running the process, there are the 5 attributes acknowledged. Still, the table in the lower part of the box remains empty, and no example data or attributes arrive at the 'Set Role' op. If I set 'SetRole's 'name' parameter manually ('class' -> 'label'), it will say that the 'class' attribute does not exist.
    When exactly does it say that the attribute does not exist? Until you run the process, there will be one *possible* problem mentioned in the problems view. Running the process should be fine, however, if you typed the attribute name exactly as it is in the data. Set a breakpoint after the Read ARFF operator if unsure.

    Best regards,
    Marius
Sign In or Register to comment.