bootphon/wordseg: wordseg-0.7.1
Description
-
New evaluation metrics in wordseg-eval:
-
adjusted rand index:
This requires the prepared text to be computed (whereas the other metrics only rely on segmented and gold texts), so it is implemented as an option
--rand-index <prep-file>
in wordseg-eval.An easiest implementation would have been to change the specifications of wordseg-eval to take the prepared text instead of the gold one, but we prefered the optional
--rand-index
for backward compatibility. -
segmentation errors summary:
Detailed report of segmentation errors, may be undersegmentation, oversegmentation or missegmentation. Implemented as the option
--summary <json-file>
in wordseg-eval.
-
-
In wordseg-dibs, renamed baseline algorithm to gold, so as to avoid confusion with wordseg-baseline. See #48.
-
tools/wordseg-qsub.sh
renamedtools/wordseg-sge.sh
and newtools/wordseg-{slurm, bash}.sh
to submit jobs on SLURM based clusters and locally using bash. -
Bugfix in
tools/wordseg-{sge, slurm, bash}.sh
: wordseg-dibs is correctly handled (was a problem with the train file). Those tools now included full pipeline, including statistics and text preparation.
Files
bootphon/wordseg-v0.7.1.zip
Files
(273.7 kB)
Name | Size | Download all |
---|---|---|
md5:4b010f189daabeade2b7249c8970e015
|
273.7 kB | Preview Download |
Additional details
Related works
- Is supplement to
- https://github.com/bootphon/wordseg/tree/v0.7.1 (URL)