Submitted by Tom Schut on Mon, 2008-02-18 12:55
Hi,
I want to run an experiment with neurons which have a bias which should
be evolved along with the weights. But I do not want any other neural
properties to be evolved aswell. Is this possible in some way. I have
tried to set the flags to 32 but then the values cannot be set from the
genotype which seems nicer. A possible way is of course to hardcode the
properties in the neuron but it seems a bit artificial.
Regards,
Tom
Forums:
Re: flags
> I want to run an experiment with neurons which have a bias which should
> be evolved along with the weights. But I do not want any other neural
> properties to be evolved aswell. Is this possible in some way. I have
> tried to set the flags to 32 but then the values cannot be set from the
> genotype which seems nicer. A possible way is of course to hardcode the
> properties in the neuron but it seems a bit artificial.
It is not possible to make some "published" properties evolvable
while other published properties are not.
If a neuron property is available in the genotype, it is also subject to
mutations.
You could use f0 special flags to restrict mutations for some neurons
or neural connections, but this is something a bit different.
Anyway, if you are using f1, consider f0 as it can give you more
control over mutations.
As far as I understand, you would like to have a flag that
publishes a neural property so that you can set its value in the
genotype, but at the same time you would like to disallow mutations of
this property (while allowing other properties to be mutated).
There is no such possibility at this moment. If this is crucial for
your experiments, I could add a list of properties excluded from
mutations so that you can set it according to your needs,
e.g. "Ns:in,Ns:fo".
macko