• run regression test and verify no regressions
• commit to SVN before posting experiments
• use that revision number in the template below
• send a commit message to the Lang ID folks
  • look into automating this (svn commit should trigger email)

A feature log template has been made that will automatically format your results. Here are its parameters:

• name: Name of the experiment/definition file (base filename, no extensions)
• date: Date experiment was performed
• experimenter: Name of experimenter
• purpose: Brief statement of Purpose of experiment, including rationale for new features
• path: Path to archived config file and results on Entropy (/home/data/experiments/langid)
• Metrics (all files referenced are in results/plot-nist_norm1/global):
  • Hard Decision (values at the chosen operating point):
    • cost: actual cost (nist.hd.points: “DCF”)
    • missrate: Miss rate (nist.hd.points: “missprob”)
    • farate: False alarm rate (nist.hd.points: “faprob”)
  • Minimum Possible Cost
    • mincost: (DET_nist.points: row 1, col 1 [first value of file])
    • minmissrate: (DET_nist.points: row 1, col 6)
    • minfarate: (DET_nist.points: row 1, col 7)
  • eer: EER (eer.txt)
• rev: Subversion revision number
• summary: English summary of notable outcome

• Placeholder for 3-gram (you shouldn't use == to manually add a header, the template will take care of that)

Spanish vs English Feature Log

Pitch Feature Log

Spoken Language ID