Trying my way into machine learning

[SheenBR]

Hello guys. So I want to learn about machine learning. As I spent some time swiping faces in Tinder, I thought, why not make a bot to learn my preferences and maybe start swiping for me, right or left?

I was not the only one to think about this and I came across this project:

You must be registered to see links

I can fully understand how it works: You must farm a bunch of photos, manually classify them as "Yes" or "No" and then it will learn from the "Yes" images so it can distinguish if an unknow image would be "Yes" or "No".

I understand everything until they process the image, making it grayslace and trying to detect only the person from the image, removing background noise. But when that Python code starts using Tensorflow and making all that magic it kinda explodes me.

I find a lot of material for Tensorflow using Python, seems to be the best language for this but I wanted to use C# so I have to translate the Python code that I dont understand to C#. :

My goal is to make that machine learning on the images. So with an unknow is given it can say if it would be an "Yes" or "No". Can anybody give me a light on this?

 

[GHOST107]

You can do that using a pattern detection algorithm and a classifier, you can extract unique features from images, like shape texture, etc then feed them to a classifier like SVM, KNN, etc to train it.

The simplest patter detection and feature extraction is LPB, and multi-scale LPB variant is used for feature extraction, you can use a sobel filter to extract the contours and generate the shape, there are other algorithms, and works pretty well with a classifier:

You must be registered to see links

 

[ashleymeah]

I dont know about Kits and all that, but neural networks are actually very simple to make yourself. The virtual cells use math functions such as linear activation, sigmoid/logistics, ReLU activation. It may be easier to get into making your own small networks. Machine learning is working backwards applying a weight to the networks that most fit what is "learning".

If you have no experience, it might be worth while into looking into shape recognition. Another sidenote is, when judging by appearance colour may be important for judging the health of someone. There is a lot of situations where neural networks can "learn" or score simple things we may overlook at first.