Visualization¶

SWIFT provides two main visualization methods, both accessible directly from the SWIFTMonitor instance. All plots return (Figure, Axes) tuples for further customization.

Bucket profile¶

The bucket profile visualizes how a feature's SHAP response and observation density are distributed across buckets — the SHAP response curve is exactly the per-bucket mean SHAP fitted in Normalization.

Basic usage¶

# Reference density only
fig, ax = monitor.plot_buckets("feature_0")

Bucket profile with reference density only

The plot shows:

Left y-axis: Mean SHAP value per bucket (line + markers) with a shaded band showing mean +/- 2 standard deviations
Right y-axis: Observation density (fraction of samples in each bucket)
X-axis: Bucket intervals
Dashed grey line: SHAP = 0 reference

Comparing distributions¶

Pass X_compare to overlay a second distribution (e.g., monitoring vs. reference):

fig, ax = monitor.plot_buckets(
    "feature_0",
    X_compare=X_drifted,
    labels=("Reference", "Drifted"),
)

Reference vs. drifted density comparison — the density shift is clearly visible

Custom density source¶

By default, the density is computed from the reference data used during fit(). Pass X to use a different data source:

# Show monitoring density instead of reference
fig, ax = monitor.plot_buckets("feature_0", X=X_mon)

Natural x-axis¶

For a more interpretable x-axis, use x_axis="natural" to show bucket midpoints on the feature-value scale instead of bucket indices:

fig, ax = monitor.plot_buckets("feature_0", x_axis="natural")

Natural x-axis mode — bucket midpoints on the feature-value scale

When there are more than 20 buckets, the default mode automatically switches to compact labels (B0, B1, ...) for readability.

SWIFT score plot¶

The SWIFT score plot provides an overview of drift across all features.

Single result¶

fig, ax = monitor.plot_swift_scores(result, threshold=0.01)

SWIFT scores — drifted — SWIFT scores with drift-colored bars — red indicates drifted features

Shows:

Bars: One per feature, colored red (drifted) or blue (not drifted)
Horizontal lines: SWIFT max, SWIFT mean, and optional threshold

Comparing two results¶

fig, ax = monitor.plot_swift_scores(
    result_clean,
    result_compare=result_drifted,
    labels=("Clean", "Drifted"),
)

SWIFT scores — comparison — Side-by-side comparison of SWIFT scores between clean and drifted data

Shows grouped side-by-side bars with neutral coloring for direct comparison.

Customization¶

Both methods return matplotlib (Figure, Axes) tuples, so you can customize the plots further:

fig, ax = monitor.plot_swift_scores(result)
ax.set_title("My Custom Title")
ax.set_ylabel("SWIFT Score")
fig.savefig("swift_scores.png", dpi=300, bbox_inches="tight")

Standalone functions¶

The plotting functions are also available as standalone functions for advanced use:

from swift.plotting import plot_bucket_profile, plot_feature_swift_scores

See the Plotting API reference for full details.