Meeting Notes - 19 Sept 2012¶
Minutes¶
Present: Ian Dickson, Karen Kruzycka, Tim Kay, Mark P, Chris C
- Mark: we're wanting to set up some sort of thing in which people can attach research output "blobs"
- Rather than just sticking up our own work and hoping people will reproduce it, we want to get third parties putting up things that they have used (autonomously)
- Ian: We talked about Drupal because we thought there might be an EndNote/BibTeX importer -- and there is
- Mark sketches a diagram
BibTeX, Endnote etc ----> Paper [URL/DOI]
|
Experiment ---------- [Instructions/Workflow]
/ \
,----------' `----------.
/ \
Code [URL] Data [URL]
The relationships all need to be N-to-N.
- Code + Data constitute experiment. Experiment may need many bits of code and many bits of data, and each conversely may be in many experiments
- Likewise an experiment may appear in many papers and vice versa
- Data may be in various partitions, e.g. training + test
- Data might not be openly available, but would still require an entity (e.g. RWC dataset)
- Users -- is there a one-to-one relationship between code site and RR site? (Mark thinks no -- the code site should have a greater level of authorisation)
- Ian notes need for logging / versioning -- rolling back in the case of erroneous / malicious edits
- Tim notes that using a wiki to hold actual object / entity data seems infelicitous
- Tim: Metrics for number of people reporting reproducibility etc -- even if the figures are all zero, we need to know that. Tracking outlinks may also be necessary
- Ian: Would like some examples of the end points (what code & data look like) -- Mark refers to Wavelab and SMALLbox, rr.epfl.ch, RWC dataset, Million Song dataset, MARSYAS, Sonic Visualiser etc as examples of RR, RR repo that doesn't necessarily have the vitality we'd like, datasets, code, etc