Explain leaf prediction path

[tlapusan]

Kind of similar with explain_prediction_path(x).

When looking at the tree structure (or leaf visualisations), you can observe some interesting nodes/leaves and you want to know what characteristics have all the training samples from that node. We can use the split values from the root until the interested node and build some similar output like in explain_prediction_path(x)

[lchakkei]

You can use view() to show the leaf prediction path e.g viz_model.view(x=X_testset[0])

[tlapusan]

Indeed, but it is not convenient/easy to find an x sample which belong to a specific leaf/node.

what I thought was something like 'explain_node_path(node_id=3)'

[parrt]

I guess that makes sense to me except I'm not sure how we know what the node id is. Do we have a way to display that?

[tlapusan]

I just put the mouse over a node and little pop-up appear with the node id. We could include also other info in this way.

[parrt]

cool! Didn't know about that. kind of like the feature earlier we talked about where in leaf display we could pop up node id in bar chart, right?

[tlapusan]

yes, could be the same... but it's kind of a hidden functionality. The user has to know it upfront.