You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We have this idea of having something similar to carl.distributions.Sampler but connected to a REST API (instead of sampling from a fixed given array).
@lukasheinrich I think you already have something along these lines. Could you comment on the expected REST API? or an example?
Since we don't have an API yet, we can design it around an actual usecase. what would the needed API endpoints look like? Do you need the events streamed as a file? Or would something else work as well (database?)
Would something like curl api.aretha.io/<sampleid> > file work?
Also what format would the events need to be in?
I think it can be quite simple. We would need to able to implement the following methods:
rvs(n_samples): this should return an array of n_samples samples. Under the hood, this should call the REST api to fetch these.
pdf(X): X is an array of samples, where each row is an individual sample. This should return an array of size len(X), where is each value is pdf(X[i]). (This is optional; as we will usually assume the likelihood-free setup)
nnlf(X): same as pdf(X) but for the negative log-likelihood. (Again, this is optional)
In some way or another, the generator behind the REST api should also be tied to model parameters that can be specified when making requests.
We have this idea of having something similar to
carl.distributions.Sampler
but connected to a REST API (instead of sampling from a fixed given array).@lukasheinrich I think you already have something along these lines. Could you comment on the expected REST API? or an example?
CC: @cranmer
The text was updated successfully, but these errors were encountered: