Optional Engines and Cross-Platform Portability

Source-derived companion to gp3ml 0.3.0 vignette optional-engine-portability.Rmd. The runnable Python companion is under examples/optional-engine-portability.py.

Optional engines remain optional. Missing packages are reported explicitly rather than changing the scientific task or silently selecting another model.

Inspect the capability table first

import gp3mlpy as gp

capabilities = gp.gp3ml_engine_capabilities()
print(capabilities)

The table records the gp3ml engine label, corresponding Python package, supported task types, package availability, backend information where applicable, and portability notes.

Engine portability. Availability is visible before fitting and is not treated as a model-selection criterion.

For a visual check:

from gp3mlpy.plotting import plot_engine_capabilities

fig = plot_engine_capabilities(capabilities)

Core and optional backends

glm and lm are part of the core scientific stack. Other engine labels may depend on additional packages or adapters. The current Python port deliberately distinguishes interface/semantic correspondence from algorithmic identity:

• ranger uses a governed scikit-learn random-forest adapter and is not claimed to be algorithmically identical to R ranger;
• nnet uses a governed scikit-learn MLP adapter and is not claimed to be algorithmically identical to R nnet;
• xgboost requires the optional xgboost package;
• keras3 requires Keras and, when requested, an explicitly checked usable backend;
• custom requires an externally supplied engine plus safety declarations.

Fail explicitly when a requested engine is unavailable

gp.assert_gp3ml_engine_available("xgboost")

This check is preferable to silently substituting a different estimator because model family is part of the declared analysis plan.

Availability is not model selection

Engine availability does not authorize automatic selection. Candidate comparison remains metric-declared, direction-declared, partition-aware, and reviewable through the model-tuning and selection objects.

Key functions

Function	Role
gp3ml_engine_capabilities	Inspect supported tasks and backend availability.
gp3ml_available_engines	Enumerate available governed engines.
assert_gp3ml_engine_available	Fail explicitly when a requested backend is missing.
fit_gazepoint_model	Fit through the governed engine interface.
integrate_black_box_model	Integrate an externally supplied model under explicit declarations.

See the plot gallery for the generated portability figure and other diagnostics.