Just to understand correctly, ML-agent is not friendly yet with Sentis, and there is no direct way to import sentis while ML-agent package and barracuda 3.0 are in the project directly, correct?
Bec when I tried to do, there were +999 compile error coming from Sentis compared when importing sentis on empty project.
I just want to understand.
I’d recommend to be patient, we are going to release ml-agent support with latest version of Sentis soon
Super, thanks for update:))
Is the ML Agent’s support an extension of the Sentis 1.1 release or will there be some new ML Agents specific additions?
Aiming to get a package together for the Sep 14th Featured Projects submission, but would love to know the direction the ML Agents support is heading. Is it model training / integration / reading or even a full set of features? (Sentis was a nice surprise so I don’t want to be greedy!)
I’ve been focused on adding useability and a front-end to deploy MLA models in-game as AI players/actors.
Hi there - We are working on ML-Agents support. Unfortunately I don’t think it will come in time for Sept 14, but hopefully shortly after so in case you rely on it for a Unite demo you would still have time. For Sept 14, perhaps you can just submit your narrative (written and visual) for the experience, and we can still work with you
Hey there - We just updated ML-Agents with Sentis integration/support. Try it out and let us know if it works well for you, or if there are any improvements that we can make.
2023.3.0a8 it can’t seem to find the preview version of the package, all other AI/ML related packages can be found:
The same happens for Unity
2022.3.10f1 (latest LTS) which is said to be supported by the release page
I’ve fixed it by manually importing the packages, but it may be something that’s good to be aware of for the team, since I do have allow preview packages enabled so I don’t see why it shouldn’t be able to find it.
Just tested and worked just fine… didnt expect such smooth transaction.
Performance wise, I notice that the GC is dramatically reduced compared to Barracuda version, also the time spent on decision making also reduced dramatically.
Havent tested on low end device yet, but I will update the thread here once we test the webgl build on i5 intel mini mac
Thank you for reporting this issue! I’ve informed the ML-Agents team about it.
Hey @Walter_Hulsebos, this release was a github only release, and as such, is not currently available through the package manager. You can install it manually by cloning the repo and installing from disk, or by using the following git url: “git+https://github.com/Unity-Technologies/ml-agents.git?path=com.unity.ml-agents#develop” to get the latest version on develop or “git+https://github.com/Unity-Technologies/ml-agents.git?path=com.unity.ml-agents#release_21” to grab release 21.
Fantastic! Let us know any other performance gains that you find.
This is great work. Going to pack up a bunch of comparative data and send it your way.
Is it best to Private message anyone in particular with a Dropbox link or upload the data another way? (github etc)
Unfortuanatly, I cant confirm any thing because of the following reason
Our game is a webgl game, We have been forced to upgrade to unity version 2022.3 as minimum requirement, yet sentis already working with 2021.
When we upgraded to 2022, We forced to upgrade URP as well, which did effect the performance as well on low end devices.
Atm, Its difficult for me to say we got performance gain by upgrading to sentis or not due to that the downgrade in performance due to URP upgrade and changing in unity version… we are not sure.
I hope you guys allow 2021 to use ML agent package 3.0.0 … its difficult people upgrade to 2022 just to use it while sentis already support 2021.
Thanks for your patience reading my feedback
What version of 2021 are you using? I can patch ML-Agents to exclude features related to 2022.3 and have it available on the develop branch.
Will training ML Agents always require setting up a Python runtime environment, or will this become a more streamlined workflow that’s built into the Unity editor ?
Lets continue the disucssion on the issue on github
Really appreciated :))
I’ve just got a response from the ML-Agents developers, and unfortunately, they replied that there are no plans for in-editor training without Python as of right now.
Python Scripting for Unity (7.0.1) comes with Python 3.10.6, while ML-Agents release 21 requries Python 3.10.12 or higher, so maybe with a small version bump a lot of the installation could be done behind the scenes