sparse_ho.models.SimplexSVR

class sparse_ho.models.SimplexSVR(estimator=None)

The simplex support vector regression without bias The optimization problem is solved in the dual.

It solves the SVR with probability vector constraints: sum_i beta_i = 1 beta_i >= 0

Parameters
estimator: sklearn

An estimator that follows the scikit-learn API.

__init__(estimator=None)

Methods

__init__([estimator])

generalized_supp(X, v, log_hyperparam)

Generalized support of iterate.

get_L(X)

Compute Lipschitz constant of datafit.

get_dual_v(mask, dense, X, y, v, log_hyperparam)

Compute the dual of v

get_full_jac_v(mask, jac_v, n_features)

TODO

get_jac_residual_norm(Xs, ys, n_samples, ...)

get_jac_v(X, y, mask, dense, jac, v)

Compute hypergradient.

get_mat_vec(X, y, mask, dense, log_C)

Returns a LinearOperator computing the matrix vector product with the Hessian of datafit.

proj_hyperparam(X, y, log_hyperparam)

Project hyperparameter on an admissible range of values.

reduce_X(X, mask)

Reduce design matrix to generalized support.

reduce_y(y, mask)

Reduce observation vector to generalized support.

sign(beta, log_hyperparams)

Get sign of iterate.