Outputs

Return value

Every entry point returns a tuple whose first item is a results dictionary:

out, poll_step, search_step = mads.main(data)
out, poll_step = poll.main(data)
out, search_step = search.main(data)

Key	Content
xmin	Best point found. For D_ and C_ variables, the set values (not the indices).
fmin	Objective value(s) at xmin, as a list (one entry per objective).
hmin	Constraint violation \(h\) at xmin (1e-15, the machine zero, when feasible).
nbb_evals	Number of blackbox evaluations.
niterations	Number of iterations.
nb_success	Number of successful poll and search steps.
psize, psuccess	Final frame size, and frame size of the best frame centre (one value per variable).
msize	Final mesh size of the best frame centre.
HV	Hypervolume of the Pareto front for multi-objective runs, "NA" otherwise.
Secondary fmin, Secondary hmin	(poll.main only) Objective and violation of the best infeasible point.

Note

In this release poll.main reports nbb_evals = 1 and nb_success = 0 whatever the run did. Use niterations or the <name>_all.out file to count evaluations.

The second and third items are the poll (Dirs2n) and search (EfficientExploration) objects at the end of the run, for users who want to inspect the final mesh or sampler.

Files in post_dir

post_dir/
├── OMADS.log                      # always: settings, per-iteration status, run summary
├── <name>_run/
│   ├── <name>_all.out             # save_results=True: one row per evaluation
│   └── coords.json                # save_coordinates=True (poll.main): poll points per iteration
└── <name>_ND/
    └── <name>_Pareto.out          # multi-objective runs

OMADS.log

A plain-text log with one block per iteration (step, success status, best point so far) and the termination reason. Set options["is_verbose"] for more detail, for example the decisions of the active sampling search.

<name>_all.out

A comma-separated file with right-aligned columns. Read it with pandas:

import pandas as pd

df = pd.read_csv("post/rosen_run/rosen_all.out", skipinitialspace=True)
df.columns = df.columns.str.strip()

Column	Content
Runtime (Sec)	Evaluation time.
Iteration, Evaluation #	Iteration and evaluation counters.
Step	Poll_2n or Search.
Delta	Smallest component of the frame size.
Status	FEASIBLE, INFEASIBLE or ERROR.
phi	Penalized objective used for comparisons.
one column per fun_names entry	Objective value(s).
max(c_in), hmax	Barrier threshold at that time.
Penalty_parameter, Multipliers	Penalty parameter and largest multiplier.
one column per var_names entry	The design variables.

coords.json

The poll points of every iteration, with the incumbent, for animations of the poll trajectory:

{
  "1": {
    "coord": [[-3.0, -3.0], [-3.0, -2.0]],
    "iter": 1,
    "x_incumbent": [-2.0, -2.0]
  },
  "2": {
    "coord": [[-2.25, -2.25], [-2.25, -1.5], [-1.75, -2.5], [-1.75, -1.75]],
    "iter": 2,
    "x_incumbent": [-2.0, -2.0]
  }
}

Note

save_coordinates works with poll.main. With mads.main it raises an IndexError at the end of the run in this release.

Multi-objective files

<name>_Pareto.out has the same columns as <name>_all.out. In this release it only keeps the starting point, so extract the front from <name>_all.out instead; see Multi-objective optimization.