felisat / federated-learning Goto Github PK

Hello,

In the configuration file of federated_learning.json, I got the error message shown as the title when I assign value to net with_"cnn"_. There is a missing part when the object is trying to inheriting "fedlearnCNN". Can you please provide that piece of code?

I am trying to run experiments with kws. Which version of kws should I download and how to download it? Is http://download.tensorflow.org/data/speech_commands_v0.02.tar.gz the right file?

When I am trying to run experiments with the model of VGG, including "vgg11" and "vgg11s" specified in the configuration file of "federated_learning.json", I will get the error message as follows:

"RuntimeError: Given groups=1, weight of size [64, 3, 3, 3], expected input[1, 1, 32, 32] to have 3 channels, but got 1 channels instead"

The compression of the gradients takes place in the compress_weight_update_up using the compress function with and without the accumulation of errors. However, upon applying the compression, the new values of the gradient are not applied to the server or the client model. So, how are the models assessed?

For a simple test, I set the hyperparameter as in the picture above and ran the test, but why does the accuracy not exceed 0.12?
Does the number of clients have a huge impact on learning? Setting the communication rounds to a large constant produces the same result.

how to compress dw? there is no such code