Search Unity

Search Results

Search Again
Page 1 of 14
  1. Luke-Houlihan
    Post

    Question Can BufferSensor take int arrays instead of float arrays as observations?

    Every number type you send to the network will be converted to a float no matter what type you sent it as, neural networks are made of and only...
    Post by: Luke-Houlihan, Aug 16, 2023 in forum: ML-Agents
  2. Luke-Houlihan
    Post

    Question Should using the BufferSensor to record varying numbers of other agents include self as an entry?

    Buffer sensor observations pass through an attention layer before reaching the policy network so I wouldn't put anything the can't potentially be...
    Post by: Luke-Houlihan, Aug 16, 2023 in forum: ML-Agents
  3. Luke-Houlihan
    Post

    Question When and how often do Agents collect observations (with CollectObservations)?

    Stacked observations are just all observations from the previous n steps (n being the number of stacked obs you've set) being sent to the model on...
    Post by: Luke-Houlihan, Aug 16, 2023 in forum: ML-Agents
  4. Luke-Houlihan
    Post

    Question When and how often do Agents collect observations (with CollectObservations)?

    If there's no memory task and no stacked observations then there would likely not be much benefit to giving the agent observations between...
    Post by: Luke-Houlihan, Aug 11, 2023 in forum: ML-Agents
  5. Luke-Houlihan
    Post

    Question When and how often do Agents collect observations (with CollectObservations)?

    Yup, that's your answer. If you turn off automatic stepping and invoke the EnvironmentStep() method manually, you'll get the behavior you're...
    Post by: Luke-Houlihan, Aug 9, 2023 in forum: ML-Agents
  6. Luke-Houlihan
    Post

    Question Couldn't start socket communication because worker number 0 is still in use

    There's probably a group of services you can restart if you don't want to do a full reboot, but I've never been bothered by the issue enough to...
    Post by: Luke-Houlihan, Jul 19, 2023 in forum: ML-Agents
  7. Luke-Houlihan
    Post

    Question Couldn't start socket communication because worker number 0 is still in use

    This is an identified issue with how sockets are assigned by linux. ML Agents cant close the open socket connection (for whatever reason, I'm not...
    Post by: Luke-Houlihan, Jul 19, 2023 in forum: ML-Agents
  8. Luke-Houlihan
    Post

    Bug Cannot spawn more than 32 envs

    I've done 35 environments and not had any issues so my guess would be something specific to your environment. How many cpu cores do you have?...
    Post by: Luke-Houlihan, Jul 19, 2023 in forum: ML-Agents
  9. Luke-Houlihan
    Post

    Question Agent not learning, but mean reward is working

    I also see you are encoding the goal position directly in the observations and not the enemy's position. You can try a sanity check by encoding...
    Post by: Luke-Houlihan, Jun 29, 2023 in forum: ML-Agents
  10. Luke-Houlihan
    Post

    Question Agent not learning, but mean reward is working

    I don't see the logic for the enemy cube, I'm guessing from your results that it is randomly placed. It seems the random placement doesn't...
    Post by: Luke-Houlihan, Jun 29, 2023 in forum: ML-Agents
  11. Luke-Houlihan
    Post

    Question The details of the self-play algorithm implementation

    MLAgent's implementation of self-play largely follows this openai paper - https://arxiv.org/pdf/1710.03748.pdf Here's the blog post too if you...
    Post by: Luke-Houlihan, Jun 29, 2023 in forum: ML-Agents
  12. Luke-Houlihan
    Post

    Question Train configuration setting in Self Play

    Yes your interpretation is correct, a larger buffer_size than team_change would mean the agent would change sides and continue adding experiences...
    Post by: Luke-Houlihan, Jun 29, 2023 in forum: ML-Agents
  13. Luke-Houlihan
    Post

    Bug Potato performance when run inference

    Yeah stacked observations have never been useful in my experiments. The technique I've previously used to avoid the overhead of a memory task...
    Post by: Luke-Houlihan, Jun 29, 2023 in forum: ML-Agents
  14. Luke-Houlihan
    Post

    Bug Potato performance when run inference

    Yeah those numbers make sense to me, that is an extremely large network and LSTMs tend to be poor performers. Running VR rendering and a large NN...
    Post by: Luke-Houlihan, Jun 25, 2023 in forum: ML-Agents
  15. Luke-Houlihan
    Post

    Question Problem on CollectObservations() of agents for learning to form the formation of UAVs.

    You're transforming the position into a local-relative directional vector which does not convey a local-relative position. You probably meant to...
    Post by: Luke-Houlihan, Jun 23, 2023 in forum: ML-Agents
  16. Luke-Houlihan
    Post

    Bug Training freezing every 10.000 steps for 60-80 seconds

    That's your buffer hitting full size and the model weights being updated. When the buffer_size is hit you'll run through a gradient descent for...
    Post by: Luke-Houlihan, Jun 23, 2023 in forum: ML-Agents
  17. Luke-Houlihan
    Post

    Bug Potato performance when run inference

    Are you doing inference on the gpu or cpu?
    Post by: Luke-Houlihan, Jun 23, 2023 in forum: ML-Agents
  18. Luke-Houlihan
    Post

    Question buffer_size in yaml & stacked vectors in Behavior Parameters

    AFAIK the stacked vector only contains observations from previous time steps and is fed into the current step along with current observations....
    Post by: Luke-Houlihan, Jun 23, 2023 in forum: ML-Agents
  19. Luke-Houlihan
    Post

    Question Package ml-agents: how reward system works, observed parameters, statistics manipulation, network co

    Yes but not through the rewards, just use statsrecorder when the reward is assigned -...
    Post by: Luke-Houlihan, Jun 12, 2023 in forum: ML-Agents
  20. Luke-Houlihan
    Post

    Question Package ml-agents: how reward system works, observed parameters, statistics manipulation, network co

    The agents needs to know everything you would need to know in order perform the same task yourself. This depends entirely on the complexity of...
    Post by: Luke-Houlihan, Jun 12, 2023 in forum: ML-Agents
Showing results 1 to 20 of 266 Next >
Search Again
Page 1 of 14