Skip to content

plot

Description

Part of the interactive API guide: effect.plot(feature, ...) on every engine, the heterogeneity views, centering, and plotting inside a subregion.

Reading time

Approx. 5' to read.

One verb, five engines

The running example is a model with a conditional interaction: the slope of x_0 flips with the sign of x_1.

def predict(x):
    """y = 10·x_0 if x_1 > 0 else -10·x_0, plus noise."""
    y = np.zeros(x.shape[0])
    ind = x[:, 1] > 0
    y[ind] = 10 * x[ind, 0]
    y[~ind] = -10 * x[~ind, 0]
    return y + np.random.normal(0, 1, x.shape[0]) * 0.3


pdp = effector.PDP(X_test, predict)
pdp.plot(feature=0)
Global PDP

rhale = effector.RHALE(X_test, predict, jacobian)
rhale.plot(feature=0)
Global RHALE

shapdp = effector.ShapDP(X_test, predict)
shapdp.plot(feature=0)
Global ShapDP

ale = effector.ALE(X_test, predict)
ale.plot(feature=0)
Global ALE

derpdp = effector.DerPDP(X_test, predict, jacobian)
derpdp.plot(feature=0)
Global DerPDP

⚠️ Every mean curve above is flat, and the model is anything but. That flat line is the average of a +10 slope and a -10 slope cancelling out. The band around it is screaming; the line is not. That is what heterogeneity is for, and why it is drawn by default.

The two arguments you will actually reach for

heterogeneity picks how the spread is drawn. Each method has its native view, on by default:

engine default other views
PDP / DerPDP "ice" (per instance curves) "std" (a band), "std_err", False
ALE / RHALE True (std band + per bin dy/dx bars) False
ShapDP "shap_values" (the per instance scatter) "std", False

centering picks the vertical anchor: False (raw), True ("zero_integral", zero mean), or "zero_start" (the curve starts at 0). Centering moves the curve up or down; it never changes its shape or its heterogeneity.

On real data

On the bike sharing rig, with the schema in the constructor, the same call renders in raw units and real level names:

pdp.plot("hr")

Global PDP of hr

The mean curve shows the two commute peaks; the ICE cloud around the evening one is wide, and that spread is what the next pages will chase.

Plot inside a subregion

Every plot accepts mask= (a boolean (N,) array) or rule= (an effector.Rule or a string). Rules speak your schema, level names included:

pdp.plot("hr", rule="workingday == no")

PDP of hr on non-working days

On non working days the commute peaks flatten into one midday dome. Same cached effects, re summarized within the subregion, zero model calls. This is the drill down you will use constantly once find_regions hands you rules worth looking at.

Cosmetics

nof_points (grid resolution), nof_ice/nof_shap_values (how many instances the heterogeneity view draws), scale_x/scale_y (display scaling when you have no schema), y_limits/dy_limits, show_avg_output (a horizontal line at the model's mean prediction), feature_label (axis title override), and show_plot=False to script figures without a display.

👉 With a schema in the constructor, ticks, level names, and units render correctly without any of the scale_* arguments.

Where to next