Continuing the Tong Framework Series we are releasing a Neural Nets C# Framework and Neural Nets Unity Adapter Framework.
With this you can create amazing intelligent behaviours through simple scripts attached to a gameObject. Also you can use your own code to use the neural net like you want.
The framework comes with three examples of a intelligent entity that runs in a speedway avoiding walls in different ways:
- Train the network through inspector - Train the network dinamically through code - Save and Load the network in a binary file easely
**[/CENTER]**Also includes TongFramework´s Persistence Framework to easely persist anything you want.
Why ANNs aren’t largely used in games? Unreliable, difficult to train, overtraining, etc. Yep!
Maybe you should market this to more AI style solutions than games? Although I dont know if anybody uses ANNs for AI solutions anymore, as they have been superseded by more powerful classifcation techniques. To be fair they are fun toys but I’m not sure what else, until I become enlightened.
Hi CrazySi, thanks for the response, i understand what you are saying, but the neural nets compared with various other ways are more simple to use, the only you have to care its about the train cases correct values, there are some posible uses of the neural nets: Radar – O’Reilly.
Im teacher at a video game school in argentina and we use neural nets a lot to simulate different behaviours, navigation behaviours, prediction behaviours, mass crowd control. Yes, it takes a while get the correct train cases but the results are successful.
I’m instantly interested in buying this package.
I have a real curiosity with AI and when I saw that someone released a neural network AI, I was ecstatic.
Sadly though, I won’t be purchasing; It’s not open source, which would of probably been the only reason for me to purchase. Even if it does look interesting.
Hi Dabeh, yes, those are two separte products, the only difference is that one is compiled and the other is the source code. The source code version has been approved so its active in the asset store right now. Greetings!
I appreciate what you’ve done, don’t get me wrong but should you post a few more examples/demos of what you can achieve using this kit to help convince some of the more skeptical that this is a worthwhile tool. Also can we see some screen shots of how you setup and train the ANN in the editor? I presume this also comes with a tutorial on how to train your dragon…er neural network.
Hi CraziSi, i will make a video tutorial of how to configure, but i can tell that with the kit you can create train cases from the editor with no code, also you can read the readme files in the asset store of the three examples. Thanks a lot for the feedback.
Ok i got the money… and this is in my cart but my question is; what can this really do? where does your system shine? is it possible to use it as a traffic simulator? etc.
Hi shifty, the core of the neural nets its to train is to detect patterns and react. So the only thing to learn about them is to create the correct training cases, the training cases teach the neural net patterns and the correct response and then the neural net remember all those and then response the remembered answer for the pattern.
Thank you for the response, I’m wondering where this system would shine… I’m not very visual! give me a couple of examples where this would be useful in say in a modern time game, would one use it on people, cars, enemies, traffic control or somewhere else
Hi shifty, the real shine of the NN is that they can be used in almost all scenerios, it can be donde to do intelligente enemies to determine their actions i.e. in a gun battle, or to drive planes in a war game, maybe do simulations of aerial battles with a lot of planes on the air, drive cars in a race, simulate a population of a city on their regular tasks i.e. GTA people. The uses dont have limits, remember that the NN are patterns saving machines, so you learn it to recognize a pattern in an input to return the correct response, with that you can use them in any scenerio. Remember to take the necesary time to test the patterns you will teach them yo remember. NN requires a time to understand their working, but once learned, they are useful not only on games, they are used to recognize faces on images, and also the neural nets can be connected with another neural nets to do a bigger task, i.e. google its using neural nets (http://www.eetimes.com/electronics-news/4376128/Google-neural-network-identifies-cats, Google X creates 16,000-core 'neural network' for independent machine learning - The Verge). In resume, the beauty of neural nets is their hability to remember patterns, once that concept its fully understand, there are no limtis to the use of NN. Thanks!
There is no documentation whatsoever, the whole package feels so empty the examples are really about avoiding any collision, i mean i can do something as simple as cast a ray and get a better result … I don’t know where to start or end, this is a disappointment borro56, thanks anyway mate
Hi Shifty, the classes are documented and the readme has the tutorials explanation, please, feel free to mail me in any thing you want to ask about, i will help you in anything you need. Greetings!
When I purchased the source code, I was under the assumption that ALL source code was available, but rather I’m still forced to rely on .dlls which are not open source(IE: Anything with TongFramework except for IA.NeuralNets.dll).
Is there a reason that you haven’t made those things available? People are usually under the assumption when they buy open source you get access to the whole thing.