Addition of the functions get_optimizer_by_name and get_name for optimizer DT

[jvdp1]

Addition of the functions

• get_optimizer_by_name
• get_name for optimizer DT

I also set the function get_activation_by_name as pure

[milancurcic]

Thanks @jvdp1, the code looks fine.

What is the need for this? Is it for the user to be able to instantiate an optimizer by a string? Or does it add some new functionality that wasn't already possible?

[milancurcic]

For getting the optimizer name, what do you think about setting a character(*), parameter :: name as a member to each concrete optimizer type, so then the name could be optimizer % name.

[jvdp1]

What is the need for this? Is it for the user to be able to instantiate an optimizer by a string?

Yes, it is. In my case, the user provides the name of the optimizer in a TOML file.

[milancurcic]

That sounds good. And I realize now that it mirrors the same functionality for activation functions (get_activation_by_name), which was necessary for https://github.com/neural-fortran/nf-keras-hdf5.

Just let me know what you think about

what do you think about setting a character(*), parameter :: name as a member to each concrete optimizer type, so then the name could be optimizer % name.

and this can go ahead.

[jvdp1]

For getting the optimizer name, what do you think about setting a character(*), parameter :: name as a member to each concrete optimizer type, so then the name could be optimizer % name.

I think it is a good idea, and probably more efficient. I will implement it.

[milancurcic]

I'm sorry for the confusion and back and forth. I now see that activation_function also has get_name, so your current approach mirrors it. In that case, actually better to stay consistent now, and we can discuss later if it should change in the style/approach. So, please merge this as is when ready.