#### Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

# Artificial Inteligence

Mod
edited November 2019 in General Posts: 891

I’ve started to prepare a simple code structure here (deleted). I’m currently googling, searching, graving up old posts about that. Some questions I’d like to ask you:

• What do you want to have in an AI API?
• Are there any must-read posts that you found and like to share?
• Right now, what do you think about the code? (Functionality, once there?)
• Will you use it?
• Are you interested in helping to develop it? (Fair Credits)

Please comment, because it will help me learning quicker)

Tagged:

• Posts: 305

I was trying to get one to work once ago, but didn't works at the end... One of the best resource was : http://neuralnetworksanddeeplearning.com/
That explain all of the things you need to know, but there is a need to convert python examples in lua.

Think that the first best demo is to recognize handwritten digits as explained in the resource.

I will try to share my current code, but it's old, and I don't know why it doesn't work, so maybe it's useless...

• Mod
Posts: 891

Below two photos attached. I decided to rework everything from scratch. Something happens.. first of all a tendency between the expects and outputs becomes different, not sure if that’s just because 01 and 10 is just the digits rotated. But.. progress.

@Jmv38 (https://codea.io/talk/discussion/1569/artificial-intelligence-ai-competition-proposal) you had created an algorithm already, maybe you can answer: In a neuronal net, how can be inputs non-numeric? Been struggling with it for a while.

• Posts: 791

@Anatoly the inputs can’t be non numeric. It’s just that the non-numeric input will be converted to a numeric representation. For example a picture would be converted into n by m inputs, each with a numeric representation of the pixel at that point.

• Mod
edited November 2019 Posts: 891

@West hmm. How about strings or other large objects? Strings don’t have a fixed size...

• Posts: 791

@Anatoly strings, sentences, documents etc are often represented by a bag of words approach where counts or instances of each word are used as the input to the ANN. More complex approaches also use the relative positions of features (for example in coding that one word follows another). Wikipedia does a reasonable job of introducing this https://en.wikipedia.org/wiki/Bag-of-words_model but it’s use will depend on what your application is.

Also it is worth noting that ANN’s are only one part of the larger field of AI