LoginLogin
Might make SBS readonly: thread

Very Simple Artificial Neural Network

Root / Submissions / [.]

amihartCreated:
Version:0.1Size:6.6 KB
(Removed to make space for better programs.) This is an artificial neural network that you can specify a truth table for a 2-bit function with a 1-bit output and the the neural network will use back propagation to learn the function. It has one hidden layer containing three nodes. There are three input nodes (1 for each bit plus 1 for the bias), and there's a single output node. It uses the sigmoid activation function. It can learn every function, including XOR which is non-linear. I might update this in the future with something more complicated. Currently it just serves as an example of how you could go about making an ANN in SmileBASIC.

Instructions:

Run the program ".RUNME". When it asks for the outputs for the truth table, specify a 1 or a 0 and press A in between each. When it asks you to test it, give it two values each being a 0 or a 1 and again press A between each of them.
ProgramsToolsTech Demo
Simeon#1
Yes yes!
hakke#2
oh yeah mr krabs
Defaultio#3
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
hakke#4
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
A neural network like that would be extremely slow in SB. (i think)
amihart#5
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
Not sure why the random insult, I mean you have no clue what my programming background is. I also have no what Jump-Kun is.
hakke#6
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
Jump-Kun is GAME3JUMP
amihart#7
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
Ah. It may be possible with some serious optimizations and spending a lot of time with it, but I'm not sure, training an ANN can be very resource intensive. I've seen someone train an ANN to play a very similar game using a genetic algorithm and I believe he said it took over night just to figure out the first level, and that's running on a desktop computer. You'd probably want to go for a genetic algorithm and not back propagation like this tutorial uses as that requires you having a list of inputs and desired outputs from the get-go. It'd be a pretty hefty project, not something I or probably anyone else cares to invest into a dead platform. As I've said about several of my other projects, I probably won't be putting too much more effort into SmileBASIC unless there's a Switch version announced. I'd prefer to work on software for the computer.
hakke#8
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
TL;DR: 1. Would require too many resources. 2. Not investing in an a dead platform
amihart#9
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
Sorry, I tend to write more than is necessary. That's about it, yeah.
hakke#10
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
no prablamo
12Me21#11
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
if SB came out on the switch, only like 5 people would buy it lol
RNGesus#12
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
And I would be one of them.
amihart#13
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
It would surely gain massive support after awhile when people realize they can make games easily with the power of a home console, I'm sure this community would grow much much larger. Given that SmileBASIC never came out in Europe or on the Wii U I'm doubtful it will come to the Switch, though, but if it does I will definitely be one of the first to buy it. SmileBoom needs to realize that for a development platform to attract developers, you need to make the platform seem viable. Not releasing it worldwide was probably the first big hit to that.
HTV04#14
Replying to:Defaultio
How about a slightly more complex neural network? Something for Jump-Kun? That may be too much for an amateur like you though. No offense.
SmileBASIC did come out on the Wii U, but just in Japan. Also I would totally get it for the Switch.