Skip to content
GitLab
Projects Groups Snippets
  • /
  • Help
    • Help
    • Support
    • Community forum
    • Submit feedback
    • Contribute to GitLab
  • Sign in
  • bob.bio.base bob.bio.base
  • Project information
    • Project information
    • Activity
    • Labels
    • Members
  • Repository
    • Repository
    • Files
    • Commits
    • Branches
    • Tags
    • Contributors
    • Graph
    • Compare
  • Issues 15
    • Issues 15
    • List
    • Boards
    • Service Desk
    • Milestones
  • Merge requests 1
    • Merge requests 1
  • CI/CD
    • CI/CD
    • Pipelines
    • Jobs
    • Schedules
  • Deployments
    • Deployments
    • Environments
    • Releases
  • Packages and registries
    • Packages and registries
    • Package Registry
    • Infrastructure Registry
  • Monitor
    • Monitor
    • Incidents
  • Analytics
    • Analytics
    • Value stream
    • CI/CD
    • Repository
  • Activity
  • Graph
  • Create a new issue
  • Jobs
  • Commits
  • Issue Boards
Collapse sidebar
  • bobbob
  • bob.bio.basebob.bio.base
  • Issues
  • #65
Closed
Open
Issue created Feb 27, 2017 by Manuel Günther@mguentherMaintainer

evaluation script is too complicated and does not make use of the internal structure

By now, we have to specify --dev-files and --eval-files using full paths. However, as most parts of the paths are common, it should be easily possible to simplify the parameters such that you only need to specify the --result-directory that you want to evaluate. The remaining parameters (check for the presence of scores-dev and scores-eval files), and even the --protocol or --experiments can automatically be estimated from the presence inside the --results-directory.

The current options --dev-files and --eval-files should IMHO still be present, so that we can plot together experiments from different directories. An exemplary evaluation call might be:

./bin/evaluate.py --result-directory results/mobio --protocol male --algorithms gmm ivector --dev-files other-results/mobio/isv/male/nonorm/scores-dev --eval-files other-results/mobio/isv/male/nonorm/scores-eval

In this case, the --algorithms could be automatically determined by the sub-directories inside results/mobio, and even the protocol might be automatically estimated, when only one common sub-directory is present inside results/mobio/*. In non-obvious cases, an error including proper instructions might be issued.

Another option would be to be able to provide several --result-directories, so that the same command line as above would be:

./bin/evaluate.py --result-directories results/mobio/gmm results/mobio/ivector other-results/mobio/isv --protocol male

This might be less confusing, but is longer to write, especially when you have several algorithms to evaluate.

@amohammadi @tiago.pereira Which option would you prefer?

Assignee
Assign to
Time tracking