(please CC me in replies, as i'm not subscribed) hi again,while i'm at reading the ml-policy, a question arose, which i would like you to clarify for me.
§4.0.2 says: > [Dependency] A package that includes a pre-trained machine learning > model, must also include the corresponding training program, or > Suggests: the package that provides the corresponding training > program. now, i'm a bit puzzled by this. consider a source package "foo", that includes a pre-trained model. the model was obtained by running "bar" on some input data.also consider that "src:foo" builds a binary package "libfoo1" which includes the pre-trained model ("statically linked" into the library).
the way i read the policy, "libfoo1" now has to "Suggests: bar".but what is the point of this? since the model is statically linked into libfoo1.so, there's no way to do anything useful with "bar".
i think, "bar" only makes sense for "src:foo" (and not "libfoo1"), but afaict the "Suggests" relationship is only valid for binary packages (at least Debian policy §7.2 speaks of "Suggests" only in the context of "Binary Dependencies"). my intuition would have suggested something like "Build-Suggests" (which of course does not exist)
am i missing something?like: "Suggests" being actually allowed for source packages (at least lintian does not complain). or that this stanza in the ML-policy is targetted towards binary packages that contain a model in distinct file (and load it at runtime), so the model could be updated independently of the binaries.
cheers, gdmasr IOhannesPS: sorry if this has already been discussed to death. if so, it would be nice if the policy (or some document in the vicinity) could point to such a discussion(s).
PPS: sorry if this list inappropriate for such questions. if so, please direct me to a better place.
Attachment:
OpenPGP_0xB65019C47F7A36F8.asc
Description: OpenPGP public key
Attachment:
OpenPGP_signature.asc
Description: OpenPGP digital signature