KernelSmoother¶

rgram.KernelSmoother.fit(
    X,
    y,
    *,
    data=None,
    sample_weight=None,
)

Fit one numeric feature and one numeric response.

Parameters¶

X array-like of shape (n_samples,) or (n_samples, 1), or str

Numeric feature, or its column name when data is supplied.

y array-like of shape (n_samples,) or (n_samples, 1), or str

Numeric response, or its column name when data is supplied.

data pandas.DataFrame, polars.DataFrame or polars.LazyFrame, default=None

Source for named columns. Pandas columns are converted internally to Polars. Selected columns are snapshotted once; indices are ignored.

sample_weight array-like of shape (n_samples,), default=None

Finite nonnegative influence weights with positive total mass.

Returns¶

self estimator

Fitted estimator, with independent training snapshots.

Raises¶

ValueError

For incompatible shapes, unequal lengths, missing/non-finite values, invalid settings, or unsupported weighting conventions.

TypeError

For nonnumeric/complex input or invalid callback/data types.

Notes¶

Rows and duplicates are preserved. Unsorted input emits an advisory warning. A failed refit invalidates prediction access. Zero-weight rows still participate in feature-based bin or bandwidth selection. Normal fitting does not run CV; Regressogram(n_bins='cv') explicitly requests it.

Examples¶

>>> import numpy as np
>>> from rgram import KernelSmoother
>>> X = np.arange(10.0).reshape(-1, 1)
>>> model = KernelSmoother().fit(X, X[:, 0])
>>> model.predict(X).shape
(10,)

rgram.KernelSmoother.fit_predict(
    X,
    y,
    *,
    data=None,
    sample_weight=None,
)

Fit and evaluate at the original training feature values.

Parameters¶

X array-like or str

One training feature, or its column name when data is supplied.

y array-like or str

Aligned numeric response, or its column name when data is supplied.

data pandas.DataFrame, polars.DataFrame or polars.LazyFrame, default=None

Source for named columns.

sample_weight array-like of shape (n_samples,), default=None

Observation influence weights; see fit.

Returns¶

prediction ndarray of shape (n_samples,)

Predictions at original training rows, preserving their order.

Notes¶

Training predictions describe the fit, not held-out performance. Fit followed by predict evaluates new queries; use predict_interval for bootstrap confidence bounds.

rgram.KernelSmoother.get_metadata_routing()

Get metadata routing of this object.

Please check metadata routing guide on how the routing mechanism works.

Returns¶

routing MetadataRequest

A sklearn.utils.metadata_routing.MetadataRequest encapsulating routing information.

rgram.KernelSmoother.get_params(deep=True)

Get parameters for this estimator.

Parameters¶

deep bool, default=True

If True, will return the parameters for this estimator and contained subobjects that are estimators.

Returns¶

params dict

Parameter names mapped to their values.

rgram.KernelSmoother.get_weights(
    X,
    *,
    kind='prediction',
)

Inspect the dense matrix of local influences in training-row order.

Parameters¶

X array-like of shape (n_queries,) or (n_queries, 1)

Query feature values.

kind {'prediction', 'kernel'}, default='prediction'

Prediction coefficients or normalized nonnegative kernel/observation weights. Local-linear prediction coefficients can be negative.

Returns¶

weights ndarray of shape (n_queries, n_samples_in_)

Columns follow original training order. Unsupported rows contain zeros. Supported prediction rows combine y_ into predicted values. For local constants, prediction and kernel weights coincide.

Notes¶

This intentionally allocates a dense matrix. Use small query selections; ordinary prediction does not require storing this full matrix.

rgram.KernelSmoother.predict(X)

Evaluate the local-constant or local-linear response curve.

Parameters¶

X array-like of shape (n_queries,) or (n_queries, 1)

Query feature values. Named inputs must match the fitted feature.

Returns¶

prediction ndarray of shape (n_queries,)

One value per query in input order. Unsupported estimates are NaN unless the configured support policy raises.

Raises¶

sklearn.exceptions.NotFittedError

If no successful fit is available.

ValueError

For invalid queries, feature-name mismatch, or strict policies.

See Also¶

• predict_interval: Pointwise bootstrap bounds and draw-support diagnostics.

rgram.KernelSmoother.predict_diagnostics(X)

Return local support and numerical information for each query.

Parameters¶

X array-like of shape (n_queries,) or (n_queries, 1)

One query feature, with fitted feature name if named.

Returns¶

diagnostics polars.DataFrame

Columns: x, in_training_range, prediction, supported, n_neighbors, effective_n, local_linear_fallback. One row per query in original order. Unsupported rows have NaN prediction, zero neighbors and effective_n, and false supported, unless a strict policy raises.

Notes¶

effective_n = 1 / sum(normalized kernel/observation weights ** 2). This measures weight concentration, not confidence or signed-coefficient variance.

rgram.KernelSmoother.predict_grid()

Evaluate diagnostics on an evenly spaced training-range grid.

Returns¶

diagnostics polars.DataFrame

The predict_diagnostics schema at n_eval_samples locations spanning the observed feature minimum and maximum, endpoints included.

Notes¶

This changes evaluation locations, not smoothing. With a constant feature, all grid coordinates are equal. Fit must have succeeded first.

rgram.KernelSmoother.predict_interval(
    X,
    *,
    confidence_level=0.95,
    n_resamples=200,
    random_state=None,
    method='percentile',
    resampling='iid',
    groups=None,
    block_size=None,
    min_valid_fraction=1.0,
)

Compute pointwise paired-bootstrap confidence intervals for the curve.

Parameters¶

X array-like of shape (n_queries,) or (n_queries, 1)

Query locations; order and duplicates are preserved.

confidence_level float, default=0.95

Nominal interval level strictly between zero and one.

n_resamples int, default=200

Bootstrap refits, at least two. More draws improve Monte Carlo stability.

random_state int or None, default=None

Seed for resampling, independent of any estimator CV seed.

method {'percentile', 'basic'}, default='percentile'

Bootstrap quantiles, or quantiles reflected around the original estimate.

resampling {'iid', 'groups', 'blocks'}, default='iid'

Resample paired independent rows, whole independent groups, or contiguous moving blocks in original training order. No sorting occurs.

groups array-like of shape (n_samples_in_,) or None, default=None

Nonmissing group labels, required only for group resampling. At least two distinct groups are required.

block_size int or None, default=None

Block length from 2 to n_samples_in_ - 1, required only for blocks.

min_valid_fraction float, default=1.0

Fraction of draws that must support each query, in (0, 1]. At least two valid draws are always required. Lower fractions condition on valid draws.

Returns¶

interval polars.DataFrame

Columns: x, prediction, lower, upper, n_valid, bootstrap_coverage. Insufficiently supported bounds are NaN. Queries are never dropped.

Raises¶

sklearn.exceptions.NotFittedError

If no successful fit is available.

ValueError

For invalid options, fewer than two training rows, invalid group/block inputs, or errors from the estimator's strict fit/prediction policies.

Notes¶

These are curve intervals, not prediction intervals or simultaneous bands. IID draws assume independent random-design rows. Group and block draws require appropriate independent groups or ordered dependence. Smoothing settings are held fixed; bin boundaries are relearned. Selection uncertainty and smoothing bias are not corrected. Memory includes n_resamples * n_queries draw values. A draw with no positive observation weight counts as unsupported. Other fit errors propagate. Explicit neighbor-only algorithms can reject unsorted draws.

rgram.KernelSmoother.regression_diagnostics(X, y)

Inspect observed values, fitted values, and residuals in input order.

Parameters¶

X array-like of shape (n_samples,) or (n_samples, 1)

One evaluation feature. Named inputs must match the fitted feature.

y array-like of shape (n_samples,) or (n_samples, 1)

Observed response, aligned with x. May be training or held-out data.

Returns¶

diagnostics polars.DataFrame

Columns: x, observed, prediction, residual (observed minus prediction), supported (finite prediction). Unsupported rows remain present.

Raises¶

sklearn.exceptions.NotFittedError

If no successful fit is available.

ValueError

For invalid, misaligned inputs or strict prediction policies.

rgram.KernelSmoother.score(
    X,
    y,
    sample_weight=None,
)

Return coefficient of determination on test data.

The coefficient of determination, \(R^2\), is defined as \((1 - \frac{u}{v})\), where \(u\) is the residual sum of squares ((y_true - y_pred)** 2).sum() and \(v\) is the total sum of squares ((y_true - y_true.mean()) ** 2).sum(). The best possible score is 1.0 and it can be negative (because the model can be arbitrarily worse). A constant model that always predicts the expected value of y, disregarding the input features, would get a \(R^2\) score of 0.0.

Parameters¶

X array-like of shape (n_samples, n_features)

Test samples. For some estimators this may be a precomputed kernel matrix or a list of generic objects instead with shape (n_samples, n_samples_fitted), where n_samples_fitted is the number of samples used in the fitting for the estimator.

y array-like of shape (n_samples,) or (n_samples, n_outputs)

True values for X.

sample_weight array-like of shape (n_samples,), default=None

Sample weights.

Returns¶

score float

\(R^2\) of self.predict(X) w.r.t. y.

Notes¶

The \(R^2\) score used when calling score on a regressor uses multioutput='uniform_average' from version 0.23 to keep consistent with default value of sklearn.metrics.r2_score. This influences the score method of all the multioutput regressors (except for sklearn.multioutput.MultiOutputRegressor).

rgram.KernelSmoother.set_fit_request(
    *,
    data='$UNCHANGED$',
    sample_weight='$UNCHANGED$',
)

Configure whether metadata should be requested to be passed to the fit method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a :term:meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config). Please check the metadata routing guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to fit if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to fit.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others. Added in version 1.3.

Parameters¶

data str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for data parameter in fit.

sample_weight str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in fit.

Returns¶

self object

The updated object.

rgram.KernelSmoother.set_params(**params)

Set the parameters of this estimator.

The method works on simple estimators as well as on nested objects (such as sklearn.pipeline.Pipeline). The latter have parameters of the form <component>__<parameter> so that it's possible to update each component of a nested object.

Parameters¶

**params dict

Estimator parameters.

Returns¶

self estimator instance

Estimator instance.

rgram.KernelSmoother.set_score_request(*, sample_weight='$UNCHANGED$')

Configure whether metadata should be requested to be passed to the score method.

Note that this method is only relevant when this estimator is used as a sub-estimator within a :term:meta-estimator and metadata routing is enabled with enable_metadata_routing=True (see sklearn.set_config). Please check the metadata routing guide on how the routing mechanism works.

The options for each parameter are:

• True: metadata is requested, and passed to score if provided. The request is ignored if metadata is not provided.

• False: metadata is not requested and the meta-estimator will not pass it to score.

• None: metadata is not requested, and the meta-estimator will raise an error if the user provides it.

• str: metadata should be passed to the meta-estimator with this given alias instead of the original name.

The default (sklearn.utils.metadata_routing.UNCHANGED) retains the existing request. This allows you to change the request for some parameters and not others. Added in version 1.3.

Parameters¶

sample_weight str, True, False, or None, default=sklearn.utils.metadata_routing.UNCHANGED

Metadata routing for sample_weight parameter in score.

Returns¶

self object

The updated object.