Function reference • slise

All functions

auto_named_list(): Creates a named list where the names are taken from the input variables

data_container(): Wrapper for creating a C++ DataContainer that parses parameter names

dlog_sigmoid(): derivative of log-sigmoid function

dsigmoid(): derivative of sigmoid function

fast_ols(): OLS solver that falls back to an optimisation if ncol(X) is huge Also supports LASSO via optimisation

grad_opt_debug(): Print debug statement for how the graduated optimisation is going

graduated_optimisation(): Graduated Optimisation to solve the SLISE problem

limited_logit(): Computes the logits from probabilities

log_approximation_ratio(): Calculate the Logarithm of the approximation ratio (logarithms are used for numerically stable calculations) See Theorem 3 from the paper for more details

log_sigmoid(): log-sigmoid function

log_sum(): Computes log(sum(exp(x))) in a numerically robust way.

log_sum_special(): Computes log(sum(exp(x) * y)), or log(sum(exp(x))) if all(y == 0), in a numerically robust way.

loss_sharp(): Sharp Loss Function Exact loss function without gradients

loss_sharp_res(): Sharp Loss Function Exact loss function without gradients for when the residuals are already calculated

loss_smooth(): Smooth Loss A loss function for when you want gradients

loss_smooth_grad(): Smooth Loss Gradient Gradient for the smooth loss function

loss_smooth_res(): Smooth Loss A loss function for when you want gradients and the residuals are already calculated

matching_epsilon(): Find the matching *epsilon

next_beta(): Find the next beta according to: ¤ approximation_ratio(alpha, beta_old, beta_new) == max_approx ¤ beta_new >= beta_old + min_increase ¤ beta_new <= beta_max

owlqn_c(): OWL-QN for optimising loss_smooth Cpp implementation

owlqn_r(): OWL-QN for optimising loss_smooth R implementation

plot(<slise>): Plot the robust regression or explanation from slise

plot(<slise_2d>): Plot the robust regression or explanation from slise in 2D

plot(<slise_bar>): Plot the robust regression or explanation from slise as bar plots

plot(<slise_distribution>): Plot the robust regression or explanation from slise with distributions

plot(<slise_mnist>): Plot the robust regression or explanation from slise as an image

plot(<slise_prediction>): Plot the robust regression or explanation from slise based on predictions

plot(<slise_wordcloud>): Plot the robust regression or explanation from slise as a wordcloud

predict(<slise>): Predict with a SLISE

print(<slise>): Print the robust regression or explanation from slise

scale_identity(): A variant of `scale` that only adds the attributes

scale_robust(): Robust Scale A variant of 'scale' that is based on median and mad (instead of mean and sd). It can handle zero variance without producing nan:s.

sigmoid(): sigmoid function

simple_pca(): Calculate the PCA rotation matrix The implementation is based on stats::prcomp. Assumes the data has already been centered and scaled (if that is desired).

slise slise-package: slise: Sparse Linear Subset Explanations

slise.explain(): SLISE for explaining Black box models.

slise.explain_comb(): SLISE Black Box Explainer Use SLISE for explaining predictions made by a black box. BUT with sparsity from a combinatorial search rather than Lasso!

slise.explain_find(): SLISE Black Box Explainer Use SLISE for explaining predictions made by a black box. BUT with a binary search for sparsity!

slise.fit(): SLISE for robust regression.

slise.formula(): SLISE for robust regression (using a formula).

slise.object(): Create a result object for SLISE that is similar to other regression method results

slise.object_unnormalise(): Turn a `slise.object` result based on normalised data to a `slise.object` result with unnormalised data. The normalised results are retained, but with a 'normalised_' prefix.

slise.preprocess(): Preprocess the data as necessary before running SLISE

slise_initialisation_candidates(): Initialise the graduated optimisation by sampling candidates

slise_initialisation_candidates2(): Initialise the graduated optimisation by sampling candidates

slise_initialisation_lasso(): Initialise the graduated optimisation with a LASSO solution

slise_initialisation_ols(): Initialise the graduated optimisation with an "Ordinary Least Squares" solution

slise_initialisation_zeros(): Initialise the graduated optimisation with a zero-vector

sparsity(): Sparsity Count the non-zero coefficients

unscale(): Unscale a scaled matrix / vector

which_min_n(): Which min n Get the indecies of the n smallest values using partial sort