Skip to content

Step semantics is confusing #166

@younik

Description

@younik

I was puzzled by getting Avg steps per trajectory: 17.0 after setting MAX_STEPS to 10.

This is because, we indicate as TrajectoryStep, both the EnvStep and AgentStep:
https://github.com/The-AI-Alliance/AgentLab2/blob/8252e484701c8b6be6b54ee94ecb084bbaddc547/src/agentlab2/core.py#L181-L184

Then, we compute the number of steps simply by counting the TrajectoryStep:
https://github.com/The-AI-Alliance/AgentLab2/blob/8252e484701c8b6be6b54ee94ecb084bbaddc547/src/agentlab2/experiment.py#L215-L216

In RL, the number of steps usually indicates the number of iteractions with the environment. For the max_steps variable we use this semantic, but for the internal code and computation we use a different one.

A common solution is to use the concept of Transition (instead of TrajectoryStep), which includes both the env step data and the agent one.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions