Skip to content

Check how we calculate metrics to be displayed in the UI #3408

Description

@slinlee

@johnnyL7 will provide more details.

When we collect metrics from a multi-agent training the results seem different than what happens back in AL.

One theory is that for multi-agent, if an agent is skipped, we aren't adjusting for that. If there are 10 agents, and 9 are skipped, we still divide by 10 to calculate that metric, rather than dividing by 1 (the one agent that actually performed an action)

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Labels

Type: QuestionFurther information is requested

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions