Edit: Geni AI is now available on the store!
Hi everyone!
I’m excited to announce a new plugin I worked on.
Introducing Geni - an artificial intelligence library. Using the power of neural networks and machine learning, Geni delivers clever AI to your game or project.
Under the hood, Geni uses Tensorflow, a robust and powerful graph API developed by Google. This means it is able to handle large datasets with no problem. It is also incredibly fast, letting you evaluate complex graphs every frame in realtime.
One of the best features of Geni is the flexibility you have for training. I’ve provided easy setups for training through gradient descent, genetic algorithms, and reinforcement learning.
Right now Geni only runs on Windows x86 and x64, but I am looking to broaden that in the future. If you have needs on another platform, please let me know!
I know that neural networks is a deep and complex field of research that can feel overwhelming. I’ve tried to make things as simple as possible through an easy-to-use API, example projects you can tinker with, and plenty of documentation. The documentation includes the thought process developing the demos, helping you understand the ‘why’ behind the structure.
I am available for any questions you might have about integrating neural networks in your project, including network structure and training strategies.
Features:
-
Capable of handling hundreds of thousands rows of data with ease
-
High performance for realtime evaluation
-
Flexible training through gradient descent, genetic algorithms, and reinforcement learning (experimental)
-
Many types of graph nodes (linear layers, LSTM, convolution, concat, random choice, along with a few others and a variety of activation functions)
-
Easy-to-use API, including JSON definition of network graphs
-
Distributed training across different machines and Unity instances
-
Three demo projects with documentation
-
Windows x86 and x64 only at the moment
-
Unity version agnostic, but demos were built with 2018
-
Unlimited email support to help integrate Geni into your project
Planned Features:
- More platforms
- Better reinforcement learning support. It’s experimental and subject to change right now without an associated demo project. Current techniques include A2C, D4PG, PPO, and a curiosity module. I hope to expand this set and include the latest research in the field.
- GPU training and evaluation
- Visual graph editor
- More demo projects
I leave you with a video describing the demo projects that come with the library as well as the demo itself.
Thank you!
~ Jon Christen
http://geni-ai.com/demo/GeniDemo.zip (Windows x64)