Allow AgentRuns to track parent learn & evaluation runs (for generating training performance plots)
Closed this issue · 0 comments
andyk commented
Currently, AgentRun
can optionally track a parent_run
, see:
Line 65 in 07bc713
Example agents (e.g., SB3 and Acme agents) use this to link an AgentRun
(e.g. if a Learn run performs an evaluate sub-run) or a ComponentRun
(which holds a run command for reproducibility).
We should:
- rename the current dependency field from
parent_run
toouter_run
- add another attribute
model_input_run
Then the webapp (and users in general) will be able to follow the lineage of a model back through history and generate a plot of the agent performance.