Deep Learning Tracking Issue

[luisquintanilla]

Work Items

• [x] #6485
• [ ] #6486
• [ ] #6487
• [x] #6488
• [x] #6489
• [ ] #6490

[GeorgeS2019]

@luisquintanilla

We need ML.NET to support a deep reinforcement learning framework, preferably based on TorchSharp for the entire .NET community, especially usable in game engine like e.g. Godot etc.

We need FOR ONCE a .NET only (without Python) ML.NET-based ML RL Agent usable in e.g. Godot (we now have many coming from Unity) and Unity.

The Godot communities now have ONNX driven RL agent based on industry phython deep RL framework. This means thare are many Godot use cases.

Others are working on Unity ML Agent like concepts BUT in Godot. We are working to replace the python part to completely TorchSharp.

A PPO TorchSharp using Open AI Gym .NET framework is also shared in the link above.

We need a ML.NET Deep RL framework compatible with what the python communities are doing.

Most important!

If you search Godot and c#, you will find > 20+ Godot c# examples written recently by Unity users who are coming over to Godot. Many of them involve codes how to port Unity to Godot c#

ML.NET Users asking for Reinforcement Learning

[asieradzk]

Sounds great. How can we get some financial support for this issue? I'm working on Godot plugin similar to unity's ML agents but I'll probably release it as a paid plugin. My electricity goes up by 20% and insurance 10% next year.

[GeorgeS2019]

@luisquintanilla The 2021 Survey clearly showed that the community clearly requested deep learning support. It would be great if the community see the significant progress that has been made in this aspect by having your team MORE generously use labels such as deep-learning or TorchSharp

Using TorchSharp label will remind clear commitment to this perhaps the most disruptive deep learning .NET attempt!

[luisquintanilla]

Closing this issue and tracking TF / Image API / BYO model separately.