ReportJob
Compute one evaluation report for a run over one split.
Rebuilds the requested split, predicts with the trained model, and hands the truth and the predictions to the report. The model's inputs are never passed on: a report compares predictions against the truth and nothing else, which is what separates it from an explainer.
Methods
get_job_name(self) -> str
ReportJobBuild a descriptive name for this job.
Returns
- str
- The report's name when it can be read, a generic label otherwise.
run(self) -> None
ReportJobCompute and persist the report's artifacts.
set_status_as_delivered(self, session_factory: 'sessionmaker' = <function ReportJob.<lambda> at 0x7f3903f0f9c0>) -> None
ReportJobMark the report as queued.
Parameters
- session_factory : sessionmaker
- Factory producing a SQLAlchemy session.
set_status_as_error(self, session_factory: 'sessionmaker' = <function ReportJob.<lambda> at 0x7f3903f0fb00>) -> None
ReportJobMark the report as failed.
Parameters
- session_factory : sessionmaker
- Factory producing a SQLAlchemy session.
on_cancel(self) -> None
BaseJobCalled in the consumer process after the worker subprocess is killed.
report_progress(self, fraction: Optional[float], message: Optional[str] = None) -> None
BaseJobReport the job's progress to the job queue.
Parameters
- fraction: Optional float
- Completion in the range 0-1, or None when the total work is unknown (the frontend then shows an indeterminate bar).
- message: Optional str
- Short description of the current phase.