common_libs | Common libraries for Viewer 's projects

I am assuming you hit the error you describe here.

It seems that the symbol is not in the 0.18 release. You can either depend on the nightly version (pip install tensorflow-federated-nightly), or modify the line to construct the new object directly, instead of using the update_struct helper. That is, the linked command could change to:

The Helm template define action creates a "function". It implicitly takes a single "parameter", using the special variable name ., and .Values is actually a lookup in .. It does not "capture" .Values at the point where it is defined; it uses the Values property of the parameter that's passed to it.

This means the template will behave differently when it's called in different contexts. As the Helm documentation on Subcharts and Global Variables describes, when executing the subchart, the top-level . parameter will have its Values replaced by the subchart's key in the primary values.

There's three ways to work around this:

1. If you're using Helm 3, you can directly import a value from the subchart into the parent. (I'm not clear what version of Helm exactly this was added, or if the syntax works in a separate requirements.yaml file.) Declare the subchart dependency in your Chart.yaml as

The new dataset Python class will need to support serialization. This is necessary because TensorFlow Federated is designed to be run on the machines that are not necessary the same as the machine that wrote the computation (e.g. smartphones in the case of cross-device federated learning). These machines may not be running Python, and hence not understand the new subclass that is created, hence the serialization layer would need to be updated. However, this is pretty low-level and there maybe alternative ways to achieve the desired goal.

Going out on a limb: If the goal is to provide metadata along with the dataset for a client, it maybe easier to alter the function signature of the iterative process returned by fed_avg_schedule.build_fed_avg_process to accept a tuple of (dataset, metadata structure) for each client.

Currently the signature of the next computation is (in TFF type shorthand introduced in Custom Federated Algorithms, Part 1: Introduction to the Federated Core):

The model_fn argument of tff.learning.build_federated_averaging_process needs to be a callable (What is a callable?) that takes no arguments and returns a tff.learning.Model.

From the code (reproduced here for readability):