>>52047958
>Not consumerist shilling or muh ganoo
>>>/out/

Did anyone try te re-create dna sequencing and creation in a pc and run it for a long time to see what happens?

>>52048249
Please stop shitting up the board.

>>52047958
>Post interesting and really good ones sources to learn.
I'm also interested in this.
Can anyone recommend noob-friendly introduction to machine learning, neural networks, natural language processing etc?

>>52047958
You mean do a type of beam search (if I remember correctly)? Because always picking the best risks going to local maxima and not the global one

>>52048319
bump this

>>52048344
I've read about picking up chromosomes by "Russian roullete method" which is based on a wheel where best chromosomes have more space than worse. So there is little probability that the chosen ones will be the worst ones.
I'm wondering what will happen when we will try to create new genotype by creating based on, lets say, 2 most fittest chromosomes. Or, pick 2 best without changing them and the rest of population create by typical random way mutating and crossing with each other.
I can't check it by myself because I didn't finished my implementation yet. And also I want to have strong, teorethical basics.

>>52048416
>pick 2 best without changing them

Thats used to preserve the nth best solutions just so they don't go away.

Tbh, there's not much depth to NN and genetic algorithms since they are quite simple, the interesting stuff is as to what are you applying it.

>>52049515
I know they are not most complicated thing in programming, but for me it's totally new (about 4 days ago I had no slightly idea what to do and how to do it).
My goal is unbeaten tetris bot. Tetris(pure) is my programming elementary project at computer science college. I decided to make this a little bit complicated and there is this ai idea.

>>52049672

If you want to try something on the same vein you could always try to implement grammatical evolution.
>http://www.grammatical-evolution.org/

>>52048416
When I learned it all you did was keep the best x successors and randomly breed them

>>52047958
I don't quite understand what you mean, but genetic algorithms have many ways of implementing the selection stage. They don't generally pick simply at random, usually there is some element that makes it more likely for the fitter individuals to 'breed' and there is also the concept of 'elitism' which ensures that the best individual always survives.

Some multiobjective genetic algorithms (like the ones that use Pareto fronts, NSGA/NSGA II for instance) may select based on other criteria still and work somewhat differently from the classic GA.

Also, it's actually very important to maintain population diversity for most problems. If your population loses diversity it's very likely that instead of converging towards the objective function's global maximum/minimum, your algorithm would get stuck on a local maximum/minimum point and provide poor optimization results.

>>52047958
Machine learning major here.
You don't select the "best" candidate because its probably not the best candidate, it might just be luck and you will end up propagating a bias all along the all along the learning process.
To be more sure to select the best candidate to derive from, you'll want to repeat the experiment around a hundred times with each candidate, average their scores and select the best candidate. In practice this is slower than keeping all the passing candidates and carrying on with the next generation.

Even then, the early best candidate might not let you spawn actual best strategy in the late stages of the process. So if you pruned all the early passing but not best candidates, you will end up with a suboptimal solution.

>>52049948
Also what he says >>52049946

>>52048319
You might find some stuff in the /sci/ sticky/wiki
https://sites.google.com/site/scienceandmathguide/