Expand description
Auto-generated module
🤖 Generated with SplitRS
Functions§
- adam
- Adam adaptive-moment optimizer.
- augmented_
lagrangian - Augmented Lagrangian method for equality-constrained minimisation.
- backtracking_
line_ search - Backtracking line search satisfying the Armijo (sufficient-decrease) condition.
- bfgs
- Full-memory BFGS (Broyden-Fletcher-Goldfarb-Shanno).
- bisection
- Bisection method for root finding on
[a, b]. - brent
- Brent’s method for 1-D minimization given a bracketing triple
(a, b, c)wherebis betweenaandcandf(b) < f(a),f(b) < f(c). - cholesky_
lower - Compute the lower-triangular Cholesky factor L of an n×n matrix stored row-major. Returns L such that L * L^T ≈ A.
- clip_
gradient - Clips the L2 norm of
gradtomax_norm(in-place). - conjugate_
gradient_ minimize - Nonlinear conjugate gradient minimization using the Fletcher-Reeves update.
- constrained_
min_ box - Box-constrained minimization via projected gradient descent.
- coordinate_
descent - Cyclic coordinate descent (derivative-free).
- forward_
solve_ lower - Forward substitution: solve L * x = b for lower-triangular L (n×n, row-major).
- genetic_
algorithm - Genetic algorithm for real-valued continuous minimisation.
- golden_
section - Golden-section search for a unimodal minimum on
[a, b]. - gradient_
descent - Gradient descent with fixed learning rate.
- gradient_
descent_ clipped - Gradient descent with gradient clipping.
- gradient_
descent_ momentum - Gradient descent with heavy-ball (Polyak) momentum.
- lbfgs
- L-BFGS (Limited-memory Broyden–Fletcher–Goldfarb–Shanno).
- lbfgsb
- L-BFGS-B: box-constrained variant of L-BFGS.
- nelder_
mead - Nelder-Mead simplex method (derivative-free).
- numerical_
gradient - Central-difference numerical gradient.
- numerical_
hessian - Finite-difference Hessian matrix (central differences, second order).
- particle_
swarm - Particle Swarm Optimization (PSO).
- powell
- Powell’s method (derivative-free multidimensional minimisation).
- sgd_
cosine_ annealing - SGD with cosine annealing learning-rate schedule.
- simulated_
annealing - Simulated annealing for global minimisation.
- wolfe_
line_ search - Strong Wolfe conditions line search.