Im using the authexample, running local. And i have modified the project to use XGBoost instead of a neural net, I am also using a custom dataset imported from a CSV. It appears that the initial authentication is done correctly. But when i do flwr run clients just time out, so im guessing the client_app is not starting, since i left some logs there and they aren’t showing. I would greatly appreciate the help. Thank you!
Hi @dj0nty , you can see in the logs that indeed the ClientApp started but something went wrong. Note the
aggregate_fit: received 0 results and 2 failures
line in the ServerApp log.
My recommendation is to start with the quickstart-xgboost example (which makes use of a purposed-build XGBoost strategy). You can then follow the steps as you did to (1) run it with the deployment engine (i.e. flower-superlink with two flower-supernodes); and then (2) make use of your custom dataset.
I hope this helps!! I also got back to you on Slack but let’s continue the discussion here better
Hi! Thank you for the suggestion. I am trying to do so with a custom dataset, but getting the following error, importing the data like this. Thank you!
Possibly. I would suggest first getting your dataset partitioner to work. You can take a look at the Flower Datasets guide on how to use local data.
Note that Flower Datasets is built on top of Hugging Face Dataset class, so if the guide i link above isn’t detailed enough, you should be able to find more online. For example here is the documentation that Hugging Face gives for loading CSV datasets: Load tabular data
I’d recommend first to get it working outside a Flower project. For example by just having a few lines in a notebook as shown in the Flower Datasets docs: Use with Local Data - Flower Datasets 0.5.0
Apparently, it works if i do it not using superlink, But works if i just run the .pys. I did try the guide on the supernode version but couldn’t figure it out, any idea on how to import it properly?
Figured it has to be this line, but the examples provide no information on how to use my custon dataset in this method. And using csv’s has only worked when not using superlink, this is running client,and server .py’s on CMD as is.