Genetic algorithm in netket

[Ashu-20]

Can I use genetic algorithm for optimization in netket? If yes how?
For reference I am trying to optimize the parameter of feed forward neural network used to calculate the ground state of Heisenberg model using genetic algorithm.

[gcarleo]

No, not implemented. I advice against it

…

On Thu, Mar 30, 2023, 12:43 Ashu-20 ***@***.***> wrote: Can I use genetic algorithm for optimization in netket? If yes how? For reference I am trying to optimize the parameter of feed forward neural network used to calculate the ground state of Heisenberg model using genetic algorithm. — Reply to this email directly, view it on GitHub <#1450>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGWYRBFWVN5GSSETNBAANZLW6VPWVANCNFSM6AAAAAAWNDM4CU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[Ashu-20]

Can you please explain why you are against it?

[gcarleo]

state of the art deep learning models are trained with gradients, these are consistently better than genetic algorithms. I am not aware of a single instance where genetic algorithms are a superior alternative to gradients, especially in a context such as deep learning where you have gradients (more or less) for free!

[PhilipVinc]

(And genetic algorithms are in general used in contexts where you either have few parameters or can't compute the gradients. As Giuseppe says, if you have them for almost free, not using them is not exactly a 'computationally efficient' choice.)