Skip to main content

Built-In SFDs

Limen ships a small set of foundational SFDs under limen.sfd.foundational_sfd. These are the packaged decoders you can run immediately without authoring your own experiment module first.

They are the fastest way to learn how Limen is shaped in practice because each one already combines:

  • params()
  • manifest()
  • a matching reference-architecture model surface

The Current Catalog​

SFDTask shapeNotes
logreg_binarybinary classificationthe main manifest-driven logistic-regression reference flow
random_binarybinary classification baselineuseful for sanity checks and control comparisons
xgboost_regressorregressiontree-based regression workflow
tabpfn_binarybinary classificationoptional, available only when tabpfn is installed

Foundational SFD Versus Reference Architecture​

Each built-in SFD has a matching model module in Reference Architecture.

The split is:

LayerOwns
foundational SFDsearch space plus manifest pipeline
reference architectureclass-based model contract and function wrapper

So, for example:

  • limen.sfd.foundational_sfd.logreg_binary owns the packaged experiment
  • limen.sfd.reference_architecture.logreg_binary owns the model implementation

This separation is what lets Trainer reconstruct a finished experiment and retrain the matching ReferenceModel.

logreg_binary​

logreg_binary is the standard manifest-driven binary classifier in the package.

It currently combines:

  • indicators such as roc, atr, ppo, and wilder_rsi
  • features such as vwap and kline_imbalance
  • a fitted quantile-based target
  • scaler selection from params (logreg, robust, rank_gauss)
  • the LogRegBinary reference model
  • CalibrationBuilder with sklearn_probability_calibrator and grid_threshold_optimizer

The classifier parameter surface mirrors the sklearn LogisticRegression constructor through manifest params: solver, penalty, dual, tol, C, fit_intercept, intercept_scaling, class_weight, random_state, max_iter, multi_class, verbose, warm_start, n_jobs, and l1_ratio.

The calibration search space includes use_calibration, use_threshold, cal_method, threshold_min, threshold_max, and threshold_step, giving a full grid of calibration modes within a single experiment run.

On a live local smoke run over the bundled test dataset in this repo, it prepared:

  • 24 training features
  • 3610 training rows

random_binary​

random_binary is the baseline binary classifier. It is deliberately simple and deliberately stochastic.

Use it when you want:

  • a control run
  • a smoke-test decoder
  • a deliberately weak comparison point

On a live local smoke run in this repo, it prepared:

  • 18 training features
  • 2999 training rows

Because it is stochastic, it is a poor fit for deterministic reconstruction in Trainer.

xgboost_regressor​

xgboost_regressor is the regression-oriented foundational SFD.

Use it when the target is better treated as continuous rather than binary.

On a live local smoke run in this repo, it prepared:

  • 49 training features
  • 3615 training rows

It requires xgboost.

tabpfn_binary​

tabpfn_binary is an optional packaged SFD. It only becomes available when tabpfn is installed.

It uses CalibrationBuilder with the same probability calibration and threshold optimisation wiring as logreg_binary, so its results also include optimal_threshold and val_score when calibration is active.

That optional status matters at import time and in local documentation examples. In a live local smoke pass in this repo, it was unavailable because tabpfn was not installed.

Running One Immediately​

The simplest way to use a built-in SFD is:

import limen

uel = limen.UniversalExperimentLoop(
    sfd=limen.sfd.logreg_binary,
)

uel.run(
    experiment_name='built-in-logreg',
    n_permutations=5,
    prep_each_round=True,
)

If you omit data=, the manifest fetches data using fetch_data(). Pass test_mode=True to UEL to use the test data source instead.

How To Choose​

  • Choose logreg_binary when you want the clearest canonical Limen path.
  • Choose random_binary when you want a baseline or smoke-test decoder.
  • Choose xgboost_regressor when the target is continuous and tree-based regression is the better fit.
  • Choose tabpfn_binary only when that dependency is installed and you specifically want the TabPFN workflow.