Visualization and analysis of LightGBM and FastTree

[torronen]

trafficstars

As I remember in some comparisons, there is a note that fastTree, LightGBM and FastForest could be converted to code. LightGBM would appear to have some methods to convert it to C++ code.

I would be interested in inspecting the trees to understand the predicted issue better. There are already some ways for explainability, like PFI and GFI, but it would be interesting to dig deeper into the models.

Are there tools that would allow inspecting the contents of treebased models either in simple-to-read format or even visually?

[michaelgsharp]

@luisquintanilla @JakeRadMSFT any ideas here?

[luisquintanilla]

I believe we have some samples for this. I agree it would make sense to make it easier though because right now it's a very manual process and the APIs are not as easy to use.

At minimum though, documenting the current process would be helpful here instead of it living in the issue linked below.

https://github.com/dotnet/machinelearning/issues/4275