lpspline.spline.piecewise_linear.PiecewiseLinear#

class lpspline.spline.piecewise_linear.PiecewiseLinear(term: str, knots: int | ndarray, tag: str | None = 'pwl', by: str | None = None)[source]#

Bases: Spline

Piecewise Linear Spine framework built primarily around discrete ReLU knot bases.

add_constraint(*constraints)#

Add one or more shape constraints to the spline.

Not all splines can accept all constraints. This function validates the compatibility of the constraints with the current spline type.

Parameters:

*constraints (Constraint) – One or more Constraint objects to apply (e.g., Monotonic, Convex).

Returns:

Returns the spline instance to allow methodical chaining.

Return type:

Spline

Raises:

ValueError – If a given constraint is incompatible with this spline type.

add_penalty(*penalties)#

Add one or more regularization penalties to the spline coefficients.

Parameters:

*penalties (Penalty) – One or more Penalty objects to apply (e.g., Ridge, Lasso).

Returns:

Returns the spline instance to allow methodical chaining.

Return type:

Spline

Raises:

TypeError – If the supplied argument is not a Penalty instance.

property by: str | None#

Returns the grouping variable column name.

Returns:

The mapping group attribute column name.

Return type:

Optional[str]

property coefficients: ndarray#

Returns the fitted coefficients of the spline.

Returns:

The computed coefficient values from the CVXPY variables.

Return type:

np.ndarray

property constraints: List[Constraint]#

Returns the CVXPY constraints associated with the spline.

Returns:

A list of all constraints that apply to this spline.

Return type:

List[cp.Constraint]

eval(x: ndarray, return_basis: bool = False, by: ndarray | None = None) → ndarray#

Evaluates the fitted numeric spline values for the given input x.

Parameters:

  • x (np.ndarray) – The 1D input feature array to evaluate the spline on.

  • return_basis (bool, default=False) – Whether to return the raw basis matrix instead of the evaluated spline.

  • by (np.ndarray, default=None) – The 1D integer-encoded grouping array, if the by argument is specified.

Returns:

A 1D numpy array of shape (n_samples,) representing the predicted values.

Return type:

np.ndarray

Raises:

AssertionError – If the spline has not been fitted and coefficients are not available.

init_by(by: ndarray)#

Initialize internal tracking for the unique classes in the by variable.

Parameters:

by (np.ndarray) – A 1D array of the categorical/grouping values found in the data.

init_spline(x: ndarray, by: ndarray | None = None)[source]#

Initializes parameter bounds dynamically assuming knots were expressed as integer lengths.

Parameters:

  • x (np.ndarray) – The 1D input target observation set.

  • by (np.ndarray, default=None) – Assigned category classes if grouping interactively.

property knots: int | ndarray#

Returns the knot breaking parameters natively tracked within.

Returns:

The 1D locations sequentially mapping piecewise shift points.

Return type:

Union[int, np.ndarray]

property penalties: List[Expression]#

Returns the penalty expressions associated with the spline.

Returns:

A list of CVXPY expressions representing the penalties.

Return type:

List[cp.Expression]

property tag: str | None#

Returns the custom tag for this spline component.

Returns:

The tag assigned to this spline component, or None.

Return type:

Optional[str]

property term: str#

Returns the column name or term this spline models.

Returns:

The column name or term string.

Return type:

str

property variables: List[Variable]#

Returns the CVXPY variables representing the spline coefficients.

Returns:

The list of CVXPY variables constructed during fitting.

Return type:

List[cp.Variable]