libmoon.solver

GradBaseSolver

class GradBaseSolver

Base: object

All gradient solvers should implement the methods in the base class GradBaseSolver

__init__(self, step_size, n_iter, tol)
Parameters:

  • step_size (int) – The step size for updating directions.

  • n_iter (int) – Maximal iterations.

  • tol (float) – Stop when the norm of gradients is less than the value of tol.

solve(self, x, prefs)
Parameters:

  • x (ndarray (K, n)) – The initial solution.

  • prefs (ndarray (K, m)) – The preference matrix.

class GradAggSolver

Base: GradBaseSolver

__init__(self, problem, step_size, n_iter, tol, agg_fun)
Parameters:

  • problem (str) – Existing problems or the given n_obj.

  • agg_fun (str) – agg_fun from ls, mtche, tche, pbi, cosmos, invagg, softtche, softmtche

solve(self, x, prefs)
class EPOSolver

Base: GradBaseSolver

EPOSolver, published in:

  1. “Multi-Task Learning with User Preferences: Gradient Descent with Controlled Ascent in Pareto Optimization”

  2. “Controllable Pareto Multi-Task Learning”

solve(self, x, prefs)
MOO-SVGDSolver

Base: GradBaseSolver

MOO-SVGDSolver, published in: “Profiling Pareto Front With Multi-Objective Stein Variational Gradient Descent”

solve(self, x, prefs, problem, n_prob, n_obj)
Parameters:

  • n_obj (int) – The number of objectives.

  • n_prob (float) – The number of problems.

class MGDAUBSolver

Base: GradBaseSolver

MGDAUBSolver, published in:

  1. “Multiple-gradient descent algorithm (MGDA) for multiobjective optimizationAlgorithme de descente à gradients multiples pour lʼoptimisation multiobjectif”

  2. “Multi-task learning as multimnist-objective optimization.”

solve(self, x, prefs, problem, n_prob, n_obj)
class PMGDASolver

Base: GradBaseSolver

PMGDASolver, published in:

“PMGDA: A Preference-based Multiple Gradient Descent Algorithm.”

solve(self, x, prefs)
class PMTLSolver

Base: GradBaseSolver

solve(self, x, prefs, problem, n_prob, n_obj)
class HVGradSolver

Base: GradBaseSolver

solve(self, x, prefs, problem, n_prob, n_obj)

SimplePSLSolver

class SimplePSLSolver

Base: object

All pareto set learning solvers should implement the methods in the base class SimplePSLSolver

__init__(self, n_obj, n_var, lr=1e-3)
Parameters:

  • n_obj (int) – Number of objectives.

  • n_var (int) – Number of variables.

  • lr (float) – Learning rate. Default is 1e-3.

forward(self, prefs)
Parameters:

prefs (ndarray (n_prob, n_obj)) – The preference matrix.

Returns:

The solution matrix of shape (n_prob, n_var).

optimize(self, problem, epoch)
Parameters:

  • problem (ProblemClass) – The problem class instance to optimize.

  • epoch (int) – Number of epochs for optimization.

evaluate(self, prefs)
Parameters:

prefs (ndarray (n_prob, n_obj)) – The preference matrix.

Returns:

The decision_variables matrix of shape (n_prob, n_var).

class SimplePSLLoRAModel

Base: SimplePSLModel