[Clam-devel] Annotator demo guide (aggregation part)
David García Garzón
dgarcia at iua.upf.edu
Wed Apr 15 02:36:48 PDT 2009
I guess i closed enough the aggregation part for Xavi to train that part of
the demo. SemWeb is still WIP.
Aggregation demo consists on constructing a project by assembling the output
of several extractors.
Before the guide, what does not work:
- You should run it in linux and from the work directory clam/Annotator. The
path to the aggregator script is still hardcoded and relative.
- You should specify different pool suffix for every extractor to avoid
collisions. No check is done, and it will crash silently if you don't. The top
level project has also a suffix that must be different to the ones in
extractors even if you have a single extractor.
- Modifying the schema of a live project is cool, but if you have already
computed a song, modifying the schema and selecting the song again it is very
likely to crash. So either remove the pool files or, more pragmatic, every
time you change the aggregation, change the top level suffix.
- Child scopes relations are not mapped, so don't rename child scopes (frame,
segments) Do not remove or rename attributes and scopes that define hierarchy
(FrameDivisions and Segmentations)
- You should ignore the schema field. I kept it because i want to support
extractorless sources, but now is completely ignored as it is queried to the
extractor.
That said a little guide:
- Open the annotator
- Create a new project
- Choose a new file example-data/demo1.pro
- You will be presented a 'project editor' dialog without extractors
- Choose the suffix ".demo1"
- Click the 'Add' button to add an extractor:
Source name 'chords'
Source command 'ChordExtractor' (if you have it in the path or full path)
File suffix: '.chord'
- When you accept the extractor the program will validate the binary by:
- Executing it parameterless
- Executing it asking the schema
- Validating the schema
- When back to the project editor, in the maps box all the descriptors of the
extractor appear, all with an marked checkbox.
- Unmark some final attribute at will (final means no frames or segments)
- Rename some final attribute at will by double clicking and editing. Renamed
items are marked in bold.
- Add another extractor:
Source name 'chords'
Source command 'ChordExtractor' (if you have it in the path or full path)
File suffix: '.chord'
- Accept the project
- Go to the schema browser and check the resulting schema
- Add a file to the project
- Click on it, and run the extractor
- Notice that both extractors are run
- Add a second segment views so that you can see the onsets and the chords
side by side
- Side note: you can now compare results from different extractors or an
extractor with a ground truth.
Do I miss something, Jun?
--
David García Garzón
(Work) dgarcia at iua dot upf anotherdot es
http://www.iua.upf.edu/~dgarcia
More information about the clam-devel
mailing list