Plot gallery¶

Visual diagnostics

Inspect governed objects without changing them¶

gp3mlpy plot methods visualize audit, validation, robustness, threshold, shift, reproducibility, and portability objects. The documentation build regenerates every featured figure from the current Python package.

16registered plot contracts

4featured walkthroughs

0hidden model selection

The gallery uses deliberately small synthetic documentation fixtures. These figures demonstrate rendering contracts only; they are not scientific results and should not be interpreted substantively.

Featured diagnostics¶

Decision layerThreshold evaluationInspect declared candidate thresholds and their consequences.

TransportabilityDataset shiftSeparate predictor shift from calibration and performance change.

PortabilityEngine capabilitiesExpose available core and optional modelling backends.

GovernanceEvidence profileSurface controls with evidence and controls still requiring review.

Decision-threshold evaluation¶

Balanced-accuracy curve across explicit decision thresholds

Decision governance

Explicit candidate thresholds¶

Candidate thresholds are declared by the analyst. The plot visualizes the resulting metric surface rather than choosing a hidden optimum.

Show Python example

import numpy as np
import gp3mlpy as gp

thresholds = gp.evaluate_gazepoint_thresholds(
    truth=["control", "control", "target", "target"],
    probability=[0.15, 0.42, 0.63, 0.88],
    positive="target",
    thresholds=np.linspace(0.2, 0.8, 7),
)

fig = thresholds.plot(metric="balanced_accuracy")

Predictor-distribution shift¶

Horizontal bars showing predictor shift magnitude

External validation

Keep shift distinct from outcome performance¶

Numeric predictors are summarized with standardized differences; categorical predictors use a distribution statistic. Shift is reported separately from outcome prevalence, calibration, and predictive performance.

Show Python example

shift = gp.audit_gazepoint_dataset_shift(
    development,
    external,
    predictors=["tracking_ratio", "fixation_duration", "condition"],
)

fig = shift.plot()

Engine portability¶

Engine availability by gp3ml engine label

Runtime capability

Make backend assumptions visible¶

Core engines and optional backends are displayed explicitly so portability assumptions are not buried inside model fitting or silently changed across environments.

Show Python example

import gp3mlpy as gp
from gp3mlpy.plotting import plot_engine_capabilities

capabilities = gp.gp3ml_engine_capabilities()
fig = plot_engine_capabilities(capabilities)

Governance evidence¶

Evidence review

Show what is evidenced and what still needs review¶

Controls with available evidence and controls requiring review are surfaced directly rather than collapsed into one opaque score.

Show Python example

profile = gp.create_gp3ml_governance_profile(evidence)
audit = gp.audit_gp3ml_governance_profile(profile)
fig = audit.plot()

The complete plotting surface¶

Decision + uncertainty
Threshold evaluation · abstention · conformal coverage

Validation + robustness
Dataset shift · model robustness · environment comparison

Governance + planning
Governance profile · analysis-plan deviations · API stability

Artifacts + reproducibility
Handoffs · model artifacts · research bundles · RO-Crate · checksums · reproducibility

Plot methods are intentionally thin: they visualize the state of an existing governed object. They do not mutate the analysis, refit models, select models, or change validation decisions.

Need the function behind a plot? Use the workflow API map for stage-oriented discovery or the complete API reference for exact-name lookup.