Model Tiers: Stable and Experimental

DeepTab separates production-ready models from research-stage models.

Tier	Import path	API expectation	Best use
Stable	`from deeptab.models import ...`	Public API intended to remain compatible within a major version.	Production, long-running projects, baseline suites.
Experimental	`from deeptab.models.experimental import ...`	May change as research implementations mature.	Prototyping, research comparisons, early feedback.

Stable Models

Stable models live directly under deeptab.models:

from deeptab.models import MambularClassifier, TabMRegressor, FTTransformerLSS

Stable model pages:

Stable models include MLP/ResNet/TabM baselines, Transformer models, Mamba-family models, neural tree models, and retrieval models. All stable models are available as *Classifier, *Regressor, and *LSS variants unless noted in the API reference.

Experimental Models

Experimental models use the explicit experimental import path:

from deeptab.models.experimental import TromptClassifier, ModernNCARegressor

The explicit import is intentional: it makes research-stage dependency risk visible in code review and experiment records.

Experimental model pages:

Custom Models

Beyond the stable and experimental tiers, you can plug in your own architecture and use it through the same scikit-learn API, preprocessing pipeline, and trainer as the built-in models. See Custom Models for the full guide.

Choosing a Tier

Consideration	Stable	Experimental
Primary use	Production and long-running projects	Prototyping and research comparisons
Reproducibility	Stable across minor releases	Requires pinning an exact version
API stability	Compatible within a major version	May introduce breaking changes
Maintenance burden	Lower; safe baseline for collaborators	Higher; tracks recent, evolving research
Goal	Reliable deployment	Early evaluation and research feedback

Note

Version pinning. For stable-only projects, pin a compatible range such as deeptab>=2.0,<3.0. For projects that use experimental models, pin the exact version (deeptab==2.0.0), since their APIs may change between releases.