Jack Gerrits

The side channel could be achieved by being more strict with what can be passed and formalizing the allowed things as supported "services" (via Workspace). Which really is just glorified...

Yeah that sounds good to me

Ideally, Python could hook into the structured logs that VW produces and forward them to the standard Python logging infrastructure. Then the user can filter or consume as need be

I implemented my take on this feature in the new set of python bindings (https://github.com/VowpalWabbit/py-vowpal-wabbit-next/blob/f5727409f9674168e7468865aa83b7e6ba8cdccd/src/main.cpp#L266). Essentially any internal log in VW is forwarded to Python's `logging` module so that the...

I'd recommend inspecting the generated MDX to see what the issue is. It could be many things and it's just way easier to inspect the MDX and go through a...

No worries, install Quarto in the way mentioned on [this page](https://microsoft.github.io/autogen/docs/contributor-guide/documentation#build-documentation-locally) Then, convert your notebook to MDX using the command `quarto render `. This should produce a MDX file. Now...

I am not familiar with what makes a databricks notebook unique and different. If you'd like to just skip it you can see how to here https://github.com/microsoft/autogen/blob/main/notebook/contributing.md#metadata-fields

IMO we should not be handling output and progress spinners at this layer of the library.

When training for CCB the chosen actions will be the same as the labelled actions (see [here](https://github.com/VowpalWabbit/vowpal_wabbit/blob/128fad36ee6b92e77f746cdc9cc8c27e2557362c/vowpalwabbit/core/src/reductions/cb/cb_sample.cc#L58)) In order for the sampling/exploration to behave as you are looking for you...

I'm not sure I follow - what would the difference between chosen actions and actual predictions be? The prediction is the set of chosen actions. For how the progressive validation...