-
Notifications
You must be signed in to change notification settings - Fork 10
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Regularisation Interface #16
Comments
ah - perfect place to discuss this. |
I've been mulling this over .... and I now think that what you already started (@casv2 ) is correct; for now, a regularizer should simply be an And then we put into ACE1.jl exactly the one-liner you had in the Slack to produce a simple regularizer which is diagonal. @casv2 I'd be grateful if you can make a PR to ACE1.jl to set this up in a way that is convenient for you and will work as expected with IPFitting.jl |
Ok, let me add a PR to ACE1.jl putting the oneliner in there such that we set up the |
I don't see why it shouldn't - we may have to modify the solvers though. |
We should also document this "interface" somewhere. Maybe the ACE1docs for now are ok, but eventually for ACE2 and ACEfit we need better "developer docs" that clearly specify those interfaces. |
once this is in ACE1 and there is documentation of this discussion, this can be closed |
To keep the agnostic nature we need to start adding regularisation specific functions to aPIPs/ACE separately. These will then return a Γ matrix, with which we solve the regularised LSQ problem here in IPFitting.
It seems this is really only an issue for Laplacian regularisation, right? For ridge regression the Γ matrix is just the identity (times some factor) and therefore independent of the basis.
More practically speaking should I look at adding this Γ function to
ACE/src/regularisers.jl
first? We can then merge later if we agree through a PR?@cortner @gabor1
The text was updated successfully, but these errors were encountered: