File Structure

Ideally, I'd like things to be fairly transparent whether you wanted to run your own recordbook on your local machine, or use the recordbook that is stored on Amazon's S3. Maybe it's possible that with a very simple abstraction layer we can accomplish this.

That would be pretty rad, because it might also mean that you could download a record from S3 and run your own local experiments on it.  Maybe.  There are a lot of directions to go in and I don't want to go too quickly in them all at the same time.

Here is a proposed file structure:

The Model

/root/event1
/root/event1/Environment.jar
/root/event1/Experiment.jar
/root/event1/AgentName/Agent.jar
/root/event1/AgentName/Summary.index
/root/event1/AgentName/Summary.data
/root/event1/AgentName/data/results_timestamp1_a.index
/root/event1/AgentName/data/results_timestamp1_b.index
/root/event1/AgentName/data/results_timestamp1_a.data
/root/event1/AgentName/data/results_timestamp1_b.data
... (more results) ...
/root/event2
...

A Realization (on Amazon S3)

/bt-recordbook/beta/
/bt-recordbook/beta/MountainCarE0001/MountainCar.jar
/bt-recordbook/beta/MountainCarE0001/MountainCarE0001.jar
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/btannerAgentLib.jar
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/Summary.index
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/Summary.data
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/data/results_1212735866201_a.index
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/data/results_1212735866201_b.index
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/data/results_1212735866201_a.data
/bt-recordbook/beta/MountainCarE0001/EpsilonGreedyCMACSarsa/data/results_1212735866201_b.data


Comments