BFGS#
- class pyrost.BFGS(loss, x0, grad=None, epsilon=0.0001, c1=0.0001, c2=0.9, xtol=1e-14, beta=0.9, line_search='minpack')#
Minimize a function of one parameter using the the Broyden-Fletcher- Goldfarb-Shanno (BFGS) algorithm.
- Parameters
loss (
Callable[[ndarray],float]) – Objective function to be minimized.x0 (
ndarray) – Initial guess.grad (
Optional[Callable[[ndarray],float]]) – Gradient of the objective function.epsilon – If grad is approximated, use this value for the step size.
c1 (
float) – Parameter for Armijo condition rule.c2 (
float) – Parameter for curvature condition rule.xtol – Relative tolerance for an acceptable step in the line search algorithm.
beta (
float) – Exponential decay coefficient for inverse Hessian matrix.line_search (
str) –Choose the implementation of the line search algorithm to enforce strong Wolfe conditions. The following keyword values are allowed:
minpack : MINPACK line search algorithm.
scipy : SciPy line search algorithm.
- grad(x)#
Return the gradient value of the objective function for a given argument.
- loss(x)#
Return the objective value for a given argument.
- state_dict()#
Returns the state of the optimizer as a dict.
- Returns
fcount : Number of functions evaluations made.
gcount : Number of gradient evaluations made.
c1 : Parameter for Armijo condition rule.
c2 : Parameter for curvature condition rule.
xk : The current point.
fval : Objective value of the current point.
old_fval : Objective value of the point prior to ‘xk’.
gfk : Gradient value of the current point.
gnorm : Gradient norm value of the current point.
Hk : The current guess of the Hessian matrix.
epsilon : If grad is approximated, use this value for the step size.
xtol : Relative tolerance for an acceptable step in the line search algorithm.
- Return type
A dictionary with all the parameters of the optimizer
- step(maxiter=10, amin=1e-100, amax=1e+100)#
Performs a single optimization step.