I apologize if I’ve formatted this wrong or posted it in the wrong area, I’m new to the forum.
I’ve been browsing the tutorials of both tensorflow and tensorflow.js today, and saw that with tensorflow, policies can be written (and in this way, known and extracted). This is documented here → Policies | TensorFlow Agents. However, I didn’t see this described in tutorials on tensorflow.js. As I want to use tensorflow.js for my dissertation (as it has to be web-based), I was hoping someone could advise me as to whether policies could still be written.
I guess the need is more on the TF Agents side to support this and they can own to ensure API is aligned with their future goals etc? However “both” is probably preferred choice as they may need advice from the JS pov too to account for needs of web engineers and js devs etc and to ensure it works well with existing web tech and may need changes to tfjs to make some things work?
I have spoken to the TF agents team about JS support but they do not have any JS eng on their team to take it on right now but yes it’s possible for a SIG effort if they can help guide as im not an expert in this area.
Thank you for all the advice! I apologize for the slow response, I’ve been super busy with coursework.
If I’m reading your comments correctly, tensorflow.js does not currently support policies, but it may be up for discussion/implementation in the future?
We are fully open source so welcome contributions to add such functionality to the project in the future. We also have a special interests group which is where people like yourself from other companies/community work together on key future areas of TensorFlow.js. Reinforcement learning could be one of them if enough people interested in hacking on that together and willing to commit to milestones etc.