openquant.sb_bagging

Compiled from Rust (crates/pyopenquant). Signatures and types are from the shipped .pyi stubs; descriptions are the functions’ own docstrings, as help() shows them. All Python modules.

Classes

SequentiallyBootstrappedBaggingClassifier

class SequentiallyBootstrappedBaggingClassifier(
    n_estimators: int = 10,
    max_samples: float = 1.0,
    max_features: float = 1.0,
    bootstrap_features: bool = False,
    oob_score: bool = False,
    random_state: int = 42,
)

Sequentially bootstrapped bagging classifier (AFML sections 4.5 and 6.3), as an object.

It has fit, predict and predict_proba, so it can be fitted on one set of rows and scored on another (for example, out of fold under purged cross-validation).

predict_proba returns one [P(y = 0), P(y = 1)] row per input row, in the order of classes_ ([0, 1]), as scikit-learn classifiers do. The probability of class 1 is the share of estimators voting 1, so predict is 1 exactly when it is at least 0.5.

**The sampling is real; the base learner is a sketch.** Each estimator draws max_samples labels with the sequential bootstrap (AFML section 4.5), which favours labels that overlap little with those already drawn, draws max_features feature indices and keeps only the first, and fits a fixed one-feature learner on the drawn rows: a stump whose threshold is the weighted mean of the feature over the drawn rows, predicting 1 on whichever side had the higher weighted rate of y == 1. The out-of-bag score stays optimistic when labels overlap, because a held-out label still overlaps drawn labels in time; score the model under purged cross-validation instead.

Parameters

n_estimators : int, default 10

Number of estimators; must be positive (checked by fit).

max_samples : float, default 1.0

Labels drawn per estimator, as a fraction of the rows of x (rounded down); must be positive, at most 1.0, and resolve to at least one row (checked by fit).

max_features : float, default 1.0

Feature indices drawn per estimator, as a fraction of the columns of x (rounded down); must be positive, at most 1.0, and resolve to at least one column (checked by fit). Only the first drawn index is used.

bootstrap_features : bool, default False

Draw feature indices with replacement instead of as a random permutation.

oob_score : bool, default False

Compute the out-of-bag accuracy (oob_score_) during fit.

random_state : int, default 42

Seed; the same seed and inputs reproduce a fit.

SequentiallyBootstrappedBaggingClassifier.classes_

classes_: list[int]

The class labels, [0, 1]: the column order of predict_proba.

SequentiallyBootstrappedBaggingClassifier.estimators_samples_

estimators_samples_: list[list[int]]

Row indices each estimator was trained on, one list per estimator.

SequentiallyBootstrappedBaggingClassifier.fit

fit(
    x: Sequence[Sequence[float]],
    y: Sequence[int],
    ind_mat: Sequence[Sequence[int]],
    sample_weight: Sequence[float] | None = None,
) -> Self

Fits the ensemble and returns the model, so calls can be chained.

Parameters

x : list[list[float]]

Features, one inner list per label (observation) and one column per feature.

y : list[int]

Class per row of x: 1 is the positive class and any other value counts as 0.

ind_mat : list[list[int]]

Bars x labels indicator matrix (as sampling.get_ind_matrix returns); its columns correspond one-to-one with the rows of x.

sample_weight : list[float] | None, default None

One non-negative, finite weight per row of x, not all zero; None weighs rows equally.

Returns

SequentiallyBootstrappedBaggingClassifier

The fitted model (self).

Raises

ValueError

If the inputs are empty, ragged or disagree on the number of rows, or a setting or sample weight is out of range. A failed fit leaves the model unfitted.

SequentiallyBootstrappedBaggingClassifier.n_estimators

n_estimators: int

Number of estimators set at construction.

SequentiallyBootstrappedBaggingClassifier.n_features_in_

n_features_in_: int | None

Column count seen by the last successful fit; None before one.

SequentiallyBootstrappedBaggingClassifier.oob_score_

oob_score_: float | None

Out-of-bag accuracy from the last fit when oob_score=True, else None.

SequentiallyBootstrappedBaggingClassifier.predict

predict(x: Sequence[Sequence[float]]) -> list[int]

Predicted class (0 or 1) for each row of x, by majority vote; ties go to 1.

Parameters

x : list[list[float]]

Rows to score, with the column count of the matrix passed to fit.

Returns

list[int]

One class per row of x.

Raises

ValueError

If the model has not been fitted, or x is empty, ragged, or has a different number of columns from the matrix passed to fit.

SequentiallyBootstrappedBaggingClassifier.predict_proba

predict_proba(x: Sequence[Sequence[float]]) -> list[list[float]]

[P(y = 0), P(y = 1)] for each row of x; the columns follow classes_.

P(y = 1) is the share of estimators voting 1.

Parameters

x : list[list[float]]

Rows to score, with the column count of the matrix passed to fit.

Returns

list[list[float]]

One [P(y = 0), P(y = 1)] row per row of x; each row sums to 1.

Raises

ValueError

If the model has not been fitted, or x is empty, ragged, or has a different number of columns from the matrix passed to fit.

SequentiallyBootstrappedBaggingClassifier.random_state

random_state: int

Seed set at construction.

Functions

fit_predict_sb_classifier

fit_predict_sb_classifier(
    x: Sequence[Sequence[float]],
    y: Sequence[int],
    ind_mat: Sequence[Sequence[int]],
    n_estimators: int = 10,
    max_samples: float = 1.0,
    max_features: float = 1.0,
    random_state: int = 42,
    sample_weight: Sequence[float] | None = None,
) -> Any

Fits a sequentially bootstrapped bagging classifier and predicts the rows it was fitted on.

A one-call shortcut for SequentiallyBootstrappedBaggingClassifier(..., oob_score=True) followed by fit(x, y, ind_mat) and predict(x); use the class to score rows it was not fitted on. Predictions are a majority vote of the estimators, ties going to 1.

**The sampling is real; the base learner is a sketch.** Each estimator draws max_samples labels with the sequential bootstrap (AFML section 4.5), which favours labels that overlap little with those already drawn, draws max_features feature indices and keeps only the first, and fits a fixed one-feature learner on the drawn rows: a stump whose threshold is the weighted mean of the feature over the drawn rows, predicting 1 on whichever side had the higher weighted rate of y == 1. The out-of-bag score stays optimistic when labels overlap, because a held-out label still overlaps drawn labels in time; score the model under purged cross-validation instead.

Parameters

x : list[list[float]]

Features, one inner list per label (observation) and one column per feature.

y : list[int]

Class per row of x: 1 is the positive class and any other value counts as 0, so map a -1/+1 encoding to 0/1 first.

ind_mat : list[list[int]]

Bars x labels indicator matrix (as sampling.get_ind_matrix returns): one inner list per bar, one 0/1 entry per label. Its columns correspond one-to-one with the rows of x.

n_estimators : int, default 10

Number of estimators; must be positive.

max_samples : float, default 1.0

Labels drawn per estimator, as a fraction of the rows of x (rounded down); must be positive, at most 1.0, and resolve to at least one row.

max_features : float, default 1.0

Feature indices drawn per estimator, as a fraction of the columns of x (rounded down); must be positive, at most 1.0, and resolve to at least one column. Only the first drawn index is used, so values above one column change only the random stream.

random_state : int, default 42

Seed; the same seed and inputs reproduce a fit.

sample_weight : list[float] | None, default None

One non-negative, finite weight per row of x, not all zero; None weighs rows equally. A row drawn twice counts twice.

Returns

dict[str, Any]

predictions: the predicted class (0 or 1) for each row of x (in-sample). oob_score: out-of-bag accuracy over the rows that at least one estimator did not draw, each predicted by the estimators that did not draw it; None if every row was drawn by every estimator.

Raises

ValueError

If x is empty or ragged; y, sample_weight or the rows of ind_mat do not match the rows of x, or ind_mat is empty; n_estimators is zero; max_samples or max_features is out of range; or a sample weight is negative or not finite, or all weights are zero.

fit_predict_sb_regressor

fit_predict_sb_regressor(
    x: Sequence[Sequence[float]],
    y: Sequence[float],
    ind_mat: Sequence[Sequence[int]],
    n_estimators: int = 10,
    max_samples: float = 1.0,
    max_features: float = 1.0,
    random_state: int = 42,
    sample_weight: Sequence[float] | None = None,
) -> Any

Fits a sequentially bootstrapped bagging regressor and predicts the rows it was fitted on.

Predictions are the mean of the estimators' predictions.

**The sampling is real; the base learner is a sketch.** Each estimator draws max_samples labels with the sequential bootstrap (AFML section 4.5), which favours labels that overlap little with those already drawn, draws max_features feature indices and keeps only the first, and fits a fixed one-feature learner on the drawn rows: a weighted least-squares line. The out-of-bag score stays optimistic when labels overlap, because a held-out label still overlaps drawn labels in time; score the model under purged cross-validation instead.

Parameters

x : list[list[float]]

Features, one inner list per label (observation) and one column per feature.

y : list[float]

Target per row of x.

ind_mat : list[list[int]]

Bars x labels indicator matrix (as sampling.get_ind_matrix returns): one inner list per bar, one 0/1 entry per label. Its columns correspond one-to-one with the rows of x.

n_estimators : int, default 10

Number of estimators; must be positive.

max_samples : float, default 1.0

Labels drawn per estimator, as a fraction of the rows of x (rounded down); must be positive, at most 1.0, and resolve to at least one row.

max_features : float, default 1.0

Feature indices drawn per estimator, as a fraction of the columns of x (rounded down); must be positive, at most 1.0, and resolve to at least one column. Only the first drawn index is used, so values above one column change only the random stream.

random_state : int, default 42

Seed; the same seed and inputs reproduce a fit.

sample_weight : list[float] | None, default None

One non-negative, finite weight per row of x, not all zero; None weighs rows equally. A row drawn twice counts twice.

Returns

dict[str, Any]

predictions: the predicted value for each row of x (in-sample). oob_score: out-of-bag R^2 over the rows that at least one estimator did not draw, each predicted by the mean of the estimators that did not draw it (0 when the scored targets are constant); None if every row was drawn by every estimator.

Raises

ValueError

If x is empty or ragged; y, sample_weight or the rows of ind_mat do not match the rows of x, or ind_mat is empty; n_estimators is zero; max_samples or max_features is out of range; or a sample weight is negative or not finite, or all weights are zero.