Ml5-library: crossover and mutate functions

[lastnod]

Just a feature request ! :)

Could you implement a crossover function in order to perform reinforcement learning ? (genetic algorithm for me :) )
and obviously with crossover we need a simple mutate function to tune weight, maybe with possibility to put your own tuning function like gaussian distribution.

i hope it's not already implemented !

[Vooxelfr]

Hello, until you are waiting for something, perhaps This Work can help you
https://wagenaartje.github.io/neataptic/
It will be great for This two repo to merge don’t you think.

It will Also be great that an a3c asynchronous advantage actor-critic algorithm be implemented.

Just imagine some people creates Games in js and every computer with just a browser open for one of these Games training a single big AI.

And just imagine That we can interact with the game to help this AI to solve problems and so all the solutions will be usefull to all the games that can be add after and can benefit directly from solutions that Was discovered before...

And just imagine that we call it...

Skynet!

I’m jocking!

[cvalenzuela]

Totally something we want to add! @shiffman has some examples we want to port at some point: https://github.com/shiffman/The-Nature-of-Code-Examples/tree/master/chp09_ga

Also happy to help if you want to contribute this!

[LukeTheWalker]

How is the GA implementation going?

[shiffman]

These are implemented as part of #899 and #919! I think I'll close this issue! Please file any new ones or reply to those threads if you like!