Closed MattJBritton closed 5 years ago
Decided that the use case for filtering by range was too narrow, and too difficult to implement satisfactorily with Altair (because it is not currently possible to filter out lines by points which constitute them).
Otherwise this is implemented in visualize_datapoints()
Build a parallel coordinates plot where each axis is a component name and the value of a datapoint on that axis is the contribution of that component to the eventual prediction. Include two extra components at the end, one for the remainder of the prediction (not in current explanation) and another for the loss of the base model.
This should be constructed in Altair and allow selection/filtering by selecting a curve and (ideally) selecting a range on the Y and a particular feature on the X.