added past code #1

emilyzfliu · 2020-12-04T07:38:22Z

i tried pushing a branch directly to neuronets but it says access denied, would it be preferable to still use branches in the future?

satra · 2020-12-04T18:35:01Z

PRs are the best way to review code instead of pushing directly. the branch can live in your fork.

fedavg.py

satra · 2020-12-04T18:54:00Z

fedavg.py

+        await load_weights(client_id)
+
+async def fed_avg():
+    await load_data()


Suggested change

await load_data()

you can load the weights in the for loop below.

satra · 2020-12-04T18:54:58Z

fedavg.py

+    self.assertAllClose(model.model.trainable,
+                        iterative_process.get_model_weights(model).trainable)
+
+    for _ in range(num_clients):


load the weights for the latest submitted model from each client inside this loop.

emilyzfliu · 2020-12-04T21:55:18Z

i think i implemented most of the changes from fedavg.py onto dwc_server.py, right now my main question is how do i make sure the dwc code only runs after there's enough client models loaded and how to save/load the priors from the models

past code

b334e70