Problem objects#

These classes build a problem in Python; The Sellar problem shows them in use. The YAML builder creates the same objects.

class DMDO.variableData(name, sp_index, index, link, coupling_type, baseline, lb, ub, value, scaling, dim, type)#

One variable copy: its name, the subproblem that holds it (sp_index), its position in the global list (index, starting at 1), the subproblem it is linked to (link), its coupling type (COUPLING_TYPE), bounds, baseline and current value, scale, dimension and type ("REAL").

class DMDO.DA(inputs, outputs, blackbox, links, coupling_type)#

A disciplinary analysis: blackbox computes outputs from inputs (lists of variableData). blackbox is a callable working on lists, or a Discipline.

class DMDO.MDA(nAnalyses, analyses, variables, responses)#

The analyses evaluated, in order, for each design of a subproblem.

class DMDO.ADMM(nsp, beta, budget, index_of_master_SP, display, scaling, mode, M_update_scheme, store_q_io=False, index=None, gamma=0.5)#

The augmented Lagrangian coordinator of Background: nsp subproblems, penalty growth beta, stall threshold gamma, update scheme M_update_scheme (w_scheme), outer budget budget and the master subproblem index_of_master_SP. scaling is a constant or a list with one scale per variable copy.

class DMDO.SubProblem(nv, index, vars, resps, is_main, analysis, coordination, opt, fmin_nop, budget, display, psize, pupdate, solver, freal=None, scipy=None, conf=None)#

One subproblem: its design variables vars and responses resps, its analysis (MDA), its coordination (ADMM), the objective callable opt, and the inner solver with its evaluation budget and settings (scipy for solver="scipy", conf for the OMADS solvers).

Enumerations#

class DMDO.COUPLING_TYPE#

SHARED, FEEDFORWARD, FEEDBACK, UNCOUPLED, CONSTANT, DUMMY.

class DMDO.w_scheme#

Penalty-weight update schemes: NORMAL, MEDIAN, MEAN, MAX, NORM_2, RANK, NO_UPDATE.

class DMDO.PSIZE_UPDATE#

How the OMADS frame size carries over between outer iterations: DEFAULT, SUCCESS, MAX, LAST.