B-Spline Kolmogorov-Arnold Networks in Native R 'torch'


[Up] [Top]

Documentation for package ‘splineKANr’ version 0.5.1

Help Pages

kan_activation_check Fraction of layer inputs outside the spline domain
kan_benchmark Multi-seed convergence and accuracy benchmark
kan_bspline_basis B-spline basis in torch (Cox-de Boor, batched)
kan_bspline_basis_r B-spline basis by the Cox-de Boor recursion (base R reference)
kan_bspline_deriv_r First derivative of the B-spline basis (base R reference)
kan_edge_function Evaluate one edge function on a grid of inputs
kan_edge_importance Edge importance (mean absolute edge activation)
kan_extend_grid Extend a domain grid by uniform knots on both sides
kan_fd_gradient_r Central finite-difference gradient of the MSE loss (base R)
kan_fit Fit a B-spline Kolmogorov-Arnold Network
kan_forward_r Forward pass of a KAN in base R
kan_gradient_r Mean squared error loss and analytic gradients in base R
kan_init_state Initialize KAN parameters
kan_layer A single B-spline KAN layer (torch module)
kan_load Save and load a fitted KAN
kan_make_grid Build an extended uniform knot grid
kan_metrics Regression accuracy metrics
kan_module_from_state Build a torch module from a state
kan_net Multi-layer B-spline KAN (torch module)
kan_partial_dependence Partial dependence of the prediction on one predictor
kan_plot_history Plot the training history
kan_plot_network Plot the network structure
kan_prune Prune edges and hidden nodes with small activations
kan_refine Refine the grid of a model (grid extension)
kan_regularization_r Regularization penalty in base R
kan_reg_control Regularization settings
kan_save Save and load a fitted KAN
kan_scale Apply or invert a scaler
kan_scaler Fit a column scaler
kan_spec Specify a B-spline KAN architecture
kan_synthetic Synthetic regression problems
kan_unscale Apply or invert a scaler
kan_update_grid Update spline grids from data (adaptive grid)
kan_validate_all Run all numerical validations
kan_validate_basis Validate the B-spline basis
kan_validate_derivative Validate the first derivative of the B-spline basis
kan_validate_gradient Validate parameter gradients of the network
plot.kan_model Plot the edge functions of a KAN layer
predict.kan_model Predict from a fitted KAN