Dear author,

First of, thanks for your contribution thus far! As a personal project I am currently looking into FL and I would like to use TF Lite for this. As the current TF Federated framework cannot be deployed yet, I think that your work might make a fine start of my project. Now I am wondering if you could give me any pointers to this and if know if e.g. we can simply get the weights that are trained on the device and send them back to the server to then, once more, be used by a Python interpreter?

Thanks!

Dear apibsaeed,

I'm a student and also a beginner in Machine Learning and Deep Learning. I'm studying your project and post about HAR. However, when I clone your project, I cannot find the file "x.npy" and "y.npy". So, how can I get them?

Sincerely.

Thank you for your post.

Transfer learning cannot be applied to the model you prepared. This is the error received when using 'tfltransfer' which causes trouble to extract the trained model.

When the model is about to be converted by

converter = TFLiteTransferConverter(num_classes, base, heads.KerasModelHead(head), optimizers.SGD(learning_rate), train_batch_size=batch_size)

, the following error pops up.

File "main.py", line 66, in <module> heads.KerasModelHead(head), File "/converter/tfltransfer/heads/keras_model_head.py", line 52, in __init__ input_def = next(self._predict_signature.inputs.values().__iter__()) AttributeError: 'list' object has no attribute 'values'

I acknowledge that the 'tfltransfer' you used was the older version, and recently there have been some updates on it. I already tried with both versions and still, the error exists. I am not sure what causes the error but I have tried a few different ways to mitigate the issue but have not been successful yet.

Would you please give me some comments on how to convert the model?

I look forward to hearing from you soon.