Machine learning where to start (sets of numbers)?

Question

Machine learning

I’ve been working as a programmer for several years now, I want to discover machine learning for myself, but somehow it doesn’t work out, I scrolled through the Habr, everything is fished out, then fish, then people, and so on and so forth. But here I found an interesting task that has nothing to do with graphics.
The essence of the problem, there is a data set at the moment, it is about 40 records, divided approximately equally into 9 categories, although not wrong. There is a game, there are objects in it, let's say everyone has their own set of properties, there are only about 15 properties, they can be combined in different ways, but they are connected between categories (types of characters) well, like a magician needs 'intelligence' and not dexterity.
And so actually, I have already existing sets of things, with a set of properties, the value of the property depends on:
1. Item level
2. Quality (there are only 6 of them, of which type: simple, normal, unusual, unique, legendary)
What I want to learn from the available data, and get new items depending on the specified quality and level.
And now the questions themselves:
is machine learning suitable for this?
is there enough data?
and, perhaps most importantly, where can one look at such examples? (I'm a practitioner)
and finally, which library to choose, which language, what is more or less relevant now? (well familiar with languages ​​php, java, c++, js but heard that everyone loves python)

Answer 1

You should look into GANs , specifically WGAN( 1 , 2 ) and CGAN . As examples of code, I can recommend three repositories: 1 , 2 , 3 .
If you still decide to pick GANs, I strongly recommend you this repository. It contains some tricks to improve the quality of the generated examples.
UPD. I also recommend taking a look at Chainer GAN and a very fresh article Progressive Growing of GANs .