Do I individually save every net for genetic algorithm? Because when I either call the hardcode() or savenet() function, it returns the array with the name of that network like {Gen_Net1,Gen_Net2} from hardcode() or [null,null] with savenet().
1 Like
For the genetic algorithm, you do not use hardCode() or saveNet(). For those, you literally just copy/paste and save the folder in question. If you want to send it to someone privately, you right click the folder and press āSave to Fileā.
hardCode() and saveNet() are used for individual networks that arenāt necessarily in object form.
1 Like
Oh. It was that simple? I was iterating through every network until I got answer which nearly crashed my computer lol. Thank you for the help
1 Like
Not only that, but it also gives more importance/weight to pixels assigned to the most significant digits, right?
2 Likes
On generation 197, I always get a C Stack overflow error, is there any reason this happens and how can I prevent it?
2 Likes
Never had that issue so I need way more information than that.
I will either need to know your settings or the script you wrote for this.
Any tutorials?
Iām just trying to make a āsimpleā NN that learns the correct distance to jump so it can get on an object.
Iāve been trying for a good 5 hours (1am to 5am) and I canāt seem to get it to work/I dont understand any of it even when i read the doc.
1 Like
Read the explanation, documentation, and examples sections thoroughly.
The explanation will give you all the information you need to know about NNs.
The documentation will give you all the information about what each specific function does and how to use them.
The examples will give you a look at what a NN system needs to work properly, whether youāre using a backpropagated or genetic system.
Iāll try that.
This stuff is really cool
1 Like
Now this is a big brain moment.
2 Likes
This. is. awesome! Studying this domain since a little while, and was thinking of making one in the next 6 months.
Thanks for sharing such a creation 
1 Like
Just wanted to say-- great work! This library has proven really useful to me. Iām actually using it for a science fair project, but I canāt say Iāve very much to show for now.
Question: Would it make sense to modify the genetic algorithm to be a little more selective and take less than 60%? Or is that the best possible solution?
Thanks for your hard work!
2 Likes
The selection process in the genetic algorithm is very subjective. In some processes, you want the majority to die. In others, you want only the worst to die off and everyone else to breed.
It is absolutely fine to edit the parameters in the genetic algorithm if you know what youāre doing.
Have fun!
3 Likes
Since I found your module, I have been experimenting around with it. It was very valuable experience, and I think I should share my stuff too. This is what I made out of your module : Walker AI - Roblox
4 Likes
Thatās really cool! Practical and you have the progress save from your studio.
Only suggestion I would give is to try to generate them in-game instead of studio if your PC is not that powerful; Robloxās servers have proven to be very strong processing-wise, much stronger than most PCs.
2 Likes
Thanks for the advice 
I will try hosting the training on the server instead of studio
1 Like
Amazing. I actually though Neural Networks were impossible in Lua. Or maybe I just donāt have some uses for NNsā¦
Edit: Fixed typo.
3 Likes
This is so amazing! As a person working on neural networks with HTTP requests as a service, I am impressed.
5 Likes
Hmmā¦ I wonder how thatās possible. What data do you send over the web?
Edit: Fixed typo lol
1 Like
I send data like generic obstacle courses as training data so the NN makes its own! Iāll do more stuff with it later, and Iām using HTTPService!
4 Likes