Migrate tutorial-series-get-started-with-flower-pytorch to FDS #2480
Conversation
adam-narozniak
requested review from
danieljanes and
tanertopal
as code owners
| @@ -640,9 +620,21 @@ | |||
| "toc_visible": true | |||
| }, | |||
| "kernelspec": { | |||
| "display_name": "flower-3.7.12", | |||
| "display_name": "flwr-rtwXnbAq-py3.10", | |||
There was a problem hiding this comment.
@adam-narozniak just to be sure, is this intended?
There was a problem hiding this comment.
that change probably comes from my side when re-running the notebook. i think this is not important. Can be discarded or kept.
There was a problem hiding this comment.
@tanertopal both "display_name" and "name" are required. I changed it to "flwr" now.
When they are not present I get "[E 09:57:23.746 NotebookApp] Notebook JSON is invalid: 'display_name' is a required property". And in UI, when starting the notebook I get the error (if not present). Similarly when "name" was not present in google colab, it complained and defaulted to "python3"
| }, | ||
| "language_info": { | ||
| "codemirror_mode": { | ||
| "name": "ipython", | ||
| "version": 3 | ||
| }, | ||
| "file_extension": ".py", | ||
| "mimetype": "text/x-python", | ||
| "name": "python", | ||
| "nbconvert_exporter": "python", | ||
| "pygments_lexer": "ipython3", | ||
| "version": "3.10.6" |
There was a problem hiding this comment.
@adam-narozniak do we need this change?
There was a problem hiding this comment.
No, we don't need it. Our current nbstripout removed it. That was added in Javier's run and I didn't notice it.
Issue
The tutorial series uses the custom dataset partitioning which might be implied.
Proposal
Migrate the 1st tutorial to use the Flower Datasets.
Discussion
I've updated the example so the structure remains the same - there is a list of dataloaders. There might be an alternative approach to just create a FederatedDataset abstraction and use it in the client function to fetch a partition, split train test and transform to DataLoader.