ocel-transform

Recipe-driven OCEL 2.0 log transformation: clean, rename, and filter event logs before mining — noise is best removed in the data, not worked around in the algorithms.

ocel-transform --in fd.sqlite --recipe clean.json --out fd.clean.sqlite

A recipe is a named, ordered list of steps; the output is a new, valid OCEL log (every step preserves referential integrity, and the result is re-validated). Each step reports what it changed — events and objects before and after — on stderr and as contract-v2 NDJSON progress on stdout, so ocel-studio can run recipes as data sources with a live progress bar.

{
  "name": "clean",
  "steps": [
    { "dropEventTypes": ["subscribed", "mentioned"] },
    { "dropEventsWhere": { "eventType": "comment", "attr": "body",
                           "matches": "(?i)^(thanks|thank you|lgtm)\\W*$" } },
    { "renameEventTypes": { "labeled": "triage", "unlabeled": "triage" } },
    { "timeWindow": { "from": "2024-01-01", "to": "2024-12-31" } },
    { "keepObjectTypes": ["issue", "user"] },
    "dropObjectsWithoutEvents"
  ]
}

Steps

Deterministic only: no step invents data. Semantic classification (e.g. ML-flagging low-value comments) belongs in annotation attributes written by other tools; recipes then filter on those attributes like any other.

The ocel family

License

MIT