MT Studio test index

MT Studio 4.1

I have a 100 node classifier, against which there are 10 training documents per node, all properly referenced.

If I try and view the referenced training document by clicking on the link, should I be able to see the document? (I can't - returns a blank screen). As well, if I build/commit the index and then attempt to test it by putting in a full document to text (by copy-pasting into the window), clicking "Run Test" does not return any results whatsoever.

Thoughts?

Find more posts tagged with

Comments

Migrateduser

Oop - previous response was too quick. Results depend on thresholds. If no similar documents were found, then you should expect no results. Have you considered using iwmtdebug?

Clark

82434a.PNG

82434b.PNG

Migrateduser

The 'getting blank when clicking on the training doc', does the document name have accented characters,ex:ï,â,ä..., in them?

Migrateduser

Also - what happens when you run iwmttransconverter on the training docs - is there any text coming out?

Parameter.jpg

ShovelFace

The behaviour is inconsistent.

When loading a vocabulary from and XML file, all nodes are properly created. I then load a training set from another XML file and build the index (which takes close to 10 minutes due to the number of training files), it appears to work. When clicking on a training file from within the GUI, a blank window is displayed.

When I build the index, I am receiving a number of errors similar to the following:

Warning: empty file in record /apps/mt-home/conf/mtstudio_home/USERFILES/DD/MetataggerType/001/english/abstract/abstract1_e.txt

. I originally had this error with MTStudio 4.11(Windows XP)/MTServer 4.11 (Solaris 9), and I upgraded both to 4.12 and the problem went away. On the upgrade, the "USERFILES/DD" part of the path (which is wrong) went away.

Just getting back to it since the upgrade and the same behaviour is re-appearing.

Any ideas? MT bug or other?

Regarding accented characters, a one node classifier with a non-accented training document does not work either. For the brief period of time where it did work, it also worked on accented characters.

ShovelFace

Ok, managed to resolve issue by rebuilding the MTServer as a file was corrupt through our build process.

Ottawa_IWOV

Seems like some experts on this thread, so I wanted to drop a line ...

We currently parse a MT Recognizer Vocabulary and show some nodes as drop-down choices in a DCT.

The only was I have been able to add terms and have them available is to "Export" the taxonomy to somewhere my DCT code can pick it up.

What I would like is to have a workflow control this, but I can seem to find the vocabulary file in MetaTagger. Is there a way to find this file and have workflow "deploy" it to another location, basically mimicking the "Export"?

Thanks for any help on this one!!

Migrateduser

you can export the taxonomy with the iwmtbuild CLT or using MTStudio

Migrateduser

You can also retrieve walk the taxonomy using the Java API (or via the CIWeb sample app).