Existing trials.md

Existing trials

If the same lab notebook is run a second time, experitur skips existing trials by default.

Running

P_to_run, P_existing

• Skip existing
  • Same trial_id? Not sufficient if parameters are added or deleted during trials
  • parameters == parameter_pre? Not sufficient alone, if default parameters change between runs or the value of a parameter deviates in a second run. Might skip too many.
  • parameters subseteq parameter_post? Not sufficient alone, if default parameters change between runs. -> We can't track changes of default parameters without actually running the trial. -> Able to detect deviation from default parameters. (subseteq =:= P_to_run.keys() is a nonempty subset of P_to_run.keys() and the values for each matching key are the same)
  • parameters == parameter_post? Wrong, because default parameters are added. Might skip none at all.
• Trial ID already taken? Include existing trial in calculation of varying parameters.
• Redo existing.

Cleaning

• All
• Failed (success is False or trial directory is empty)
• [Experiment]

Backend

• Retrieve trial data by id ('<experiment>/<trial>')
• Generate trial id given parameter set (experiment, parameters)
• Retrieve trial data by parameter set