Agent update #8
Agent update #8
Conversation
+minor reformat
using autoprep8 to reformat code for readability
|
|
||
| def get_new_action(self, prev_step: AAIAction) -> AAIAction: | ||
| """ | ||
| Provide a vector of 9 real values, one for each action, which is then softmaxed to provide the probability of selecting that action. Relative differences between the values is what is important. |
There was a problem hiding this comment.
Will this need to be maintained in the long run? For example, we may add new actions to the agent so will this script need to be maintained accordingly?
There was a problem hiding this comment.
Yes, will require updates.
| @@ -20,6 +20,7 @@ def main(): | |||
| rayMaxDegrees=30, | |||
| raysPerSide=(n_rays - 1) // 2, | |||
| play=False, | |||
| inference=True, | |||
There was a problem hiding this comment.
why was inference set true here and was not the case before the PR?
There was a problem hiding this comment.
Because presumably the user would want to watch the agent completing the task to confirm expected behaviour.
animalai/raycastparser.py
Outdated
| @@ -28,6 +28,7 @@ class RayCastObjects(enum.Enum): | |||
| SIGNPOSTER = 11 | |||
| DECAYGOAL = 12 | |||
| DECAYGOALBOUNCE = 13 | |||
| HOLLOWBOX = 14 | |||
There was a problem hiding this comment.
Added this with this PR as otherwise it would have been delayed until the next major PR for python codebase. Standalone mod.
There was a problem hiding this comment.
I don't think this is best practice.
| randomAgent = RandomActionAgent(step_length_distribution=lambda: np.random.normal( | ||
| 5, 1)) # a Rayleigh walker (sampling from normal distribution) | ||
|
|
||
| # by default should be AnimalAI?team=0 |
There was a problem hiding this comment.
What you mean by this?
There was a problem hiding this comment.
Which part? a Rayleigh Walker is a type of random walker in stochastic process theory, where angles and saccades are drawn from normal distributions. The default to AnimalAI?team=0 is the standard behaviour for the AAI low-level API.
| actions = AAIActions() | ||
|
|
||
| # Run two episodes | ||
| for _episode in range(2): |
There was a problem hiding this comment.
do the episodes start at index 0? so, arena 0, arena 1, ... , n arenas.
Did i misunderstand this loop?
There was a problem hiding this comment.
It plays the config twice over.
| @@ -7,11 +7,11 @@ arenas: | |||
| - !Item | |||
| name: Agent | |||
| positions: | |||
| - !Vector3 { x: 5, y: 1, z: 5 } | |||
| - !Vector3 { x: 5, y: 0, z: 5 } | |||
There was a problem hiding this comment.
Just to let you know the y value constraints are corrected in unity (as heads up note)
|
have you also considered the versioning? Is this a breaking change (becomes version 5.0.0) to the braintenberg agent? Or can we go to 4.1.0 OR 4.0.2 depending on the scale of the changes @kozzy97 |
|
This is not a breaking change (the behaviour of the braitenberg is slightly different and more reliable but all scripts should work the same), but the addition of the random action agent is significant. I would recommend this, plus the hollowbox change, be included in a 4.1.0 release. |
Proposed change(s)
I have added a random action agent implemented with the low-level API (as per Braitenberg) and fixed the reliability of the braitenberg agent.
Types of change(s)
Checklist
Other comments
Screenshots (if any)