Can someone help me figure out how to code a system where the

Recognizing numerals is like the hello world of the machine learning field.

Here's an example using tensorflow to train a neural net to identify handwritten digits on the MNIST dataset
https://www.tensorflow.org/versions/r0.10/tutorials/mnist/beginners/index.html

>>56122112
>Can someone help me figure out how to code a system where the user draws in a simple shape, and the computer recognizes it within the context of a pre-determined set of shapes?
You mean, handwriting recognition? You correctly recognized that this is indeed a very advanced problem. It is a very classic case of a classification problem (say, a set of images -> {0,1,...9})

>but I can't imagine that a game like Brain Age was all that advanced to code
What exactly makes you think so? First, as difficult as it is, if you know your shit, it's actually rather straight-forward. Second, who says they implemented it themselves and didn't just use some framework for this. There are hundreds out there, even specifically for recognizing handwriting, even free ones.

Have a look at this. It's a really fucking cool visualization of the MNIST data set, which is just a bunch of small black and white images that contain numbers. Maybe that explains to you what kind of a problem this is and what kind of methods can be used to deal with it.
http://colah.github.io/posts/2014-10-Visualizing-MNIST/

>>56122112
you could also use a kernel
https://en.wikipedia.org/wiki/Kernel_(image_processing)

>>56122187
Thanks! This really helps.

>>56122243
Damn cool read.

>>56122112
Brain Age is pretty bad at recognizing numbers if you don't write them how the devs intended

If you're like me and write "4" in 1 stroke instead of 2 strokes, then it's almost always going to pick it up as a number that takes 1 stroke.

It leads me to believe that they don't use handwriting recognition. I think it just keeps track of your stylus' directional movement and stroke count for numbers and letters, and uses some basic pattern/object recognition for drawings, because it never EVER guessed correctly what I focused on in my drawings...

Besides, the DS is pretty weak for machine learning and Brain Age is very snappy and never lags, so they're probably using something very simple. That's my guess anyway.

>>56122665
related
https://youtu.be/mSFHKAvTGNk?t=16m43s

>>56122665
Yeah lol. Thats why I figured it couldn't be so difficult to get something working. I'm not trying to code something that needs to be accurate.

A better example of what I am trying to do is from Ni No Kuni

https://youtu.be/ucs8dDig47o?t=1950

Just a simple recognition of a basic shape. I don't need advanced AI learning and it doesnt need to be accurate. The symbols will be large and very different (unlike say a 1 and a 7).

>>56122665
>Brain Age
Shit, it's called Dr. Kawashima's Brain Training where I live.
Kinda feels weird...

>>56122980
Didn't' realise time links don't work on 4chan. Time is 32:30

>>56122980

If you just want to recognise a few shapes then just draw the perfect shape as a white on black image/template, capture what the user draws another white on black image, then multiply the pixel values of what the user drew with each preset drawing and decide what they drew based on which picture comes out with the most white pixels, or something to that effect.

>>56123574
But then if someone drew a really small shape they could potentially get false positives on every sample if black-white pixel ratio was within acceptable range. And if you tweaked it to expect a near-perfect drawing then you might get a ton of false negatives with somewhat accurately drawn shapes.