Outputs#

A run returns its results in memory (RunResult) and writes them to its output folder, <working_dir>/<mdo_name>_post.

Result objects#

RunResult (session.result()):

Field

Meaning

status

created, running, finished, failed or closed.

stop_reason

Why the run ended ("Iteration budget exhausted", "Max inconsitency is below stopping threshold", "No progress after several iterations", "Stopped on request").

iterations

Number of completed outer iterations.

objective

Objective of the master subproblem at the last design.

max_inconsistency

Largest scaled inconsistency between linked copies, \(\|\mathbf{q}\|_\infty\).

feasible

Whether the master subproblem satisfies its constraints (violation \(\le 10^{-12}\)).

variables

Every variable copy, keyed "<name>_<subproblem>".

history

One IterationRecord per outer iteration.

post_dir

The output folder.

RunResult.to_dict() returns the same content as JSON-ready values.

IterationRecord (session.step(), observers): iteration, objective, max_inconsistency, max_penalty_weight, variables_change (Euclidean norm of the change of all copies) and worst_link (index of the link with the largest inconsistency).

All the .out files are comma-separated tables; the post-processing dashboard (The post-processing dashboard) plots any of their columns interactively.

The output folder#

<mdo_name>_post/
├── Coordination_history.out
├── DMDO.log
├── coord_chk.pkl
└── coordination_summary_post/
    ├── SP_1/
    │   ├── SP_1_0.out, SP_1_1.out, ...
    │   └── OMADS.log
    └── SP_2/ ...
Coordination_history.out

One row per outer iteration (comma-separated, columns padded with spaces). Load it with pandas.read_csv(path, skipinitialspace=True) and strip the column names.

DMDO.log

The log of the run: problem setup, the progress lines, stopping messages.

coord_chk.pkl

Checkpoint written at the start of every outer iteration after the first; resume with Session.from_checkpoint (The Session API).

SP_<j>/SP_<j>_<t>.out

Every evaluation of subproblem \(j\) during outer iteration \(t\).

SP_<j>/OMADS.log

Log of the OMADS solver of subproblem \(j\).

Coordination_history.out#

Column

Meaning

Time

Wall-clock time at the end of the outer iteration.

Iteration #

Outer iteration.

Max. inconsistency

\(\|\mathbf{q}\|_\infty\) after the inner loop.

columns 4 and 5

Reserved flags (0).

Objective

Objective of the master subproblem.

Status

Feasible or Infeasible (constraints of the master subproblem); Error if no objective value.

Variables change

Euclidean norm of the change of all copies during the iteration.

max(w)

Largest penalty weight.

Coupling_with_qmax

The link with the largest inconsistency, as <name>_<subproblem> to <name>_<subproblem>.

hmin_<j>

Constraint violation of subproblem \(j\).

<name>_<j>

Value of every variable copy.

SP_<j>_<t>.out#

One row per evaluation of the subproblem: Time, Iteration #, Evaluation #, Max. inconsistency, Penalty (\(\phi\)), Objective (the local objective without the penalty), Coupling Status (feasibility of the local constraints), followed by the design variables and the responses of the subproblem.