Partial Dependence & ICE Plot Explorer (2D)

A partial dependence plot answers one question: on average, how does the model's prediction change as one feature moves, once every other feature is averaged out? This 2D simulator renders a real two-feature response surface as a pannable, zoomable heatmap, samples a set of data instances on it, and builds the PDP curve for whichever feature you choose to sweep by averaging that surface across all the sampled instances — exactly the formula used by real SHAP/PDP tooling. Individual conditional expectation (ICE) curves are drawn in a linked chart panel for every instance so you can see the average being built from real per-instance behaviour, and an interaction-strength control lets you watch the moment PDP's core assumption — that the swept feature doesn't interact with the others — starts to break down, with the ICE curves fanning out around a PDP average that increasingly misrepresents any single instance.