Change Agent.Heuristic to take a float[] argument #3765
Conversation
| @@ -31,7 +34,9 @@ public void RequestDecision(AgentInfo info, List<ISensor> sensors) | |||
| StepSensors(sensors); | |||
| if (!info.done) | |||
| { | |||
| m_LastDecision = m_Heuristic.Invoke(); | |||
| // Reset m_LastDecision each time. | |||
| m_LastDecision = new float[m_numActions]; | |||
There was a problem hiding this comment.
I couldn't decide whether or not to reuse the array here. I figure this approach is safer, and this isn't performance-critical enough to worry about the allocation/GC
There was a problem hiding this comment.
I think it depends on the number of agents and the size of the action array. If there is a large action space and a QA person is trying to make a demo file to use in training but they are hitting GC hitches every few seconds that could make it 1. hard to actually make a good demo. 2. frustrate the hell out of that person.
There was a problem hiding this comment.
Discussed offline - we need to do a deeper review of array sharing and/or copying around this (and the other policies). Logged as https://jira.unity3d.com/browse/MLA-892 for followup.
|
overall looks good. I would rather we re-use the array. If you feel strongly about no re-using it, let's talk. |
docs/Migrating.md
Outdated
| @@ -15,6 +15,7 @@ The versions can be found in | |||
| * The `play_against_current_self_ratio` self-play trainer hyperparameter has been renamed to `play_against_latest_model_ratio` | |||
| * Removed the multi-agent gym option from the gym wrapper. For multi-agent scenarios, use the [Low Level Python API](Python-API.md). | |||
| * The low level Python API has changed. You can look at the document [Low Level Python API documentation](Python-API.md) for more information. If you use `mlagents-learn` for training, this should be a transparent change. | |||
| * The signature of `Agent.Heuristic()` was changes to take the `float[]` actions as a parameter, instead of returning the array. This was done to prevent a common source of error where users would return arrays of the wrong size. | |||
There was a problem hiding this comment.
changes --> changed
the float[] --> float[] (?)
There was a problem hiding this comment.
Changed to
The signature of
Agent.Heuristic()was changed to take afloat[]as a parameter...
(and in the changelog)
Proposed change(s)
Changes the signature of Agent.Heuristic() to take a
float[]instead of returning one. This eliminate a common source of user error, where the returned array was the wrong size.Useful links (Github issues, JIRA tickets, ML-Agents forum threads etc.)
https://jira.unity3d.com/browse/MLA-768
#3719
Types of change(s)
Checklist
Other comments