uf3.regression.least_squares

This module provides the WeightedLinearModel class for fitting UF potentials from featurized DataFrames using regularized least squares.

Functions

apply_weighted_gram	Deprecated utility function for weighting gram matrix.
apply_weights	Deprecated utility function for weighting inputs/outputs.
arrange_coefficients	Arrange coefficients by degree of interaction.
batched_moore_penrose	Batched evaluation of gram matrix (x^T x) and ordinate (x^T y).
batched_prediction	Convenience function for optimization workflow.
dataframe_to_tuples	Extract energy/force inputs/outputs from DataFrame.
find_pair_potential_well	Identify coefficient index corresponding to possible potential well.
freeze_columns	Freeze coefficients of the solution (e.g.
freeze_regularizer	Apply freezing mask to regularizer, eliminating masked columns.
get_freezing_mask	Freezing mask is the set difference between the range of feature indices and the indices to be excluded (col_idx).
linear_least_squares	Solves the linear least-squares problem Ax=y.
lu_factorization	LU factorization for least-squares solution using np.linalg.solve().
mae_metric	Mean-absolute error metric.
moore_penrose_components	Compute gram matrix (x^T x) and ordinate (x^T y).
postprocess_coefficients_2b	Postprocess 2B coefficients to enforce repulsive core.
revert_frozen_coefficients	Reverse freezing operations by arranging learned coefficients and frozen coefficients using the mask.
rmse_metric	Root-mean-square error metric.
subset_prediction	Convenience function for optimization workflow.
validate_regularizer	Check for consistency between regularizer matrix and number of features.
weighted_least_squares	Solves the linear least-squares problem with optional Tikhonov regularizer matrix and optional weighting.

Classes

BasicLinearModel	Base class for linear regression.
VarianceRecorder	Convenience class for computing online variance and mean
WeightedLinearModel	Handler class for regularized linear least squares using energies and forces and basis set provided by bspline.BsplineBasis.