Integration with Fed-BioMed
jeandut opened this issue · 3 comments
From @AyedSamy:
"Regarding the integration, and to facilitate the way we handle the link between the federated classes of FLamby and the nodes of Fed-BioMed, I thought about including a common naming for these federated classes, in common.py file. For example for fed ixi, I have added a simple line in it
FedClass = FedIXITiny
It makes the automation easier working with all the datasets.
On an other hand, I think we could guess these federated classes from the dataset folder name, the problem is we called ours fed_ixi instead of fed_ixi_tiny, what do you think about it ?"
@AyedSamy on the naming of the IXI dataset the fact that it is named IXITiny introduces a lot of issues and forces to have extra logic so I would be in favor of renaming.
In case of the integration I do not really know how to answer as I do not have seen the specifics of your code yet. I would use the mean that makes the more sense to you and that is easier. However of course having in mind that the logic we add for one dataset would need to be added to all of them eventually.
"Guessing" from a folder seems legitimate we are only doing that in several places in the code. But frankly it is hard to answer wo seeing the code. Please take a look at the data_path argument that was recently added maybe it can help your use-case ?
@AyedSamy did my answer satisfy you ? Do not hesitate if you need help or more details.
@jeandut Hi Jean sorry for the late response, yes I totally agree with your point!
Renaming to IXITiny would introduces a lot of issues, so I would go with the simple solution, which is just adding a line in common.py:
FedClass = FedIXITiny (with the required import as well).
As you said this logic would be done with all the datasets.
So the object FedClass could be used for each of them without difficulty through Fed-BioMed