adelie.state.pinball#

adelie.state.pinball(A: MatrixConstraintBase32 | MatrixConstraintBase64, y_var: float, S: ndarray, penalty_neg: ndarray, penalty_pos: ndarray, kappa: int, max_iters: int, tol: float, screen_set_size: int, screen_set: ndarray, is_screen: ndarray, screen_ASAT_diag: ndarray, screen_AS: ndarray, active_set_size: int, active_set: ndarray, is_active: ndarray, beta: ndarray, resid: ndarray, grad: ndarray, loss: float)[source]#

Creates a pinball state object.

Parameters:

A(m, d) Union[MatrixConstraintBase32, MatrixConstraintBase64]: Constraint matrix. It is typically one of the matrices defined in adelie.matrix submodule.
y_varfloat: Variance of \(y = S^{-\frac{1}{2}} v\) equivalent to \(\|y\|_2^2\).
S(d, d) ndarray: Positive semi-definite matrix.
penalty_neg(m,) ndarray: Penalty on the negative part of \(\beta\).
penalty_pos(m,) ndarray: Penalty on the positive part of \(\beta\).
kappaint: Violation batching size.
max_itersint: Maximum number of coordinate descents.
tolfloat: Coordinate descent convergence tolerance.
screen_set_sizeint: Number of screen groups. screen_set[i] is only well-defined for i in the range [0, screen_set_size).
screen_set(m,) ndarray: Screen set buffer. screen_set[i] is the i th screen variable that is in the range [0, m).
is_screen(m,) ndarray: Boolean vector indicating whether the j th feature is screen.
screen_ASAT_diag(m,) ndarray: \(A_j^\top S A_j\) where feature j is screen.
screen_AS(m, d) ndarray: \(A_j^\top S\) where feature j is screen.
active_set_sizeint: Number of active groups. active_set[i] is only well-defined for i in the range [0, active_set_size).
active_set(m,) ndarray: Active set buffer. active_set[i] is the i th active variable that is in the range [0, m).
is_active(m,) ndarray: Boolean vector indicating whether the j th feature is active.
beta(m,) ndarray: Coefficient vector.
resid(d,) ndarray: Residual \(v-SA^\top\beta\).
grad(m,) ndarray: Internal buffer that is implementation-defined.
lossfloat: The current loss \(\frac{1}{2} \|S^{-\frac{1}{2}} v - S^{\frac{1}{2}} A^\top \beta\|_2^2\).

Returns:

wrap: Wrapper state object.