#1287 pareto_optimal, pareto_optimal_eps: bool criteria, and no epsilon after a bool criterion
Since 20.22.1 a criterion may be bool, minimised (false before true, not(x) where true is
better). pareto_optimal_eps reads its arguments by type: a numeric criterion is followed by its
epsilon, a bool criterion is not. Both pages say what the old builds gave (criterion 2 must be
numeric; Cannot find operator for these arguments), that the uint8(x) workaround gives the same
result, how a forgotten epsilon is now caught, and that a uint2 or uint4 criterion works; each
has an example with a NeedsCar criterion.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#1286 pareto_optimal and pareto_optimal_eps: the groups are sorted and swept in parallel
Performance sections: since 20.22.1 the elements are grouped per partition value by a radix sort
and every partition is sorted and swept on its own, on all threads, with the same result; the
NetworkModel_PBL and 40-million-row measurements, and the working memory.
#1281 #1282 pareto_optimal pages: links to existing pages, with the page name second
Six links on the two pages had the display text and the page name the wrong way round, or named a page
that does not exist: argument(s), index number(s), value type(s), value composition, parameter(s) and
values unit(s). The GitHub wiki shows them as links to missing pages, and the geodms.nl converter leaves
them unresolved, so their | turned the definition paragraph and a bullet into a table. They now follow the
[[text|Page]] order the other pages use and point to Argument, Index-numbers, Value-type, Composition,
Parameter and Values-unit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#1282 pareto_optimal_eps on a page of its own, and imp2_epsilon on the impedance_matrix pages
pareto_optimal_eps.md is new: syntax with an epsilon after each criterion, the bucket rule, which element of a
box survives, the subset and coverage properties, that the fastest element can drop out when the first
criterion has a width, the edge rounding, and an example table with three result columns computed on the
20.21.0 build. pareto_optimal.md keeps the exact function, a pointer to the eps page and a performance section.
Relational-functions.md lists both.
Impedance-options.md: the pareto section names its four forms; the imp2_epsilon bullet says what a node keeps,
the bound and the approximation, and that the argument takes the value type and unit of the alternative
impedance (a float64 value on a float32 search is refused, checked on the build); the performance bullet adds
the width as the fourth control; a worked impedance_matrix_od64 example, run verbatim on the build.
impedance_matrix.md: the alternative impedance, OrgZone_max_imp2 and imp2_epsilon in the argument table, a
dated paragraph on imp2_epsilon, see also. impedance_matrix_od64.md: why a pareto result can need uint64.
pareto_optimal_eps is not yet in nav/geodms.md of the converter, like pareto_optimal.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#1282 epsilon-dominance: pareto(imp2_epsilon) and pareto_optimal_eps
Impedance options: the pareto section gains the imp2_epsilon argument, with the bucket rule, the bound it
gives, what stays exact (the first row per pair) and what becomes approximate. pareto_optimal: the _eps form
with an epsilon per criterion, the sweep on buckets, which element of a box survives, the subset and the
coverage property, and an example on the page's table.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
#1281 pareto_optimal: the non-dominated elements per partition by a sort and a sweep
New page pareto_optimal.md: syntax, the weak-dominance rule with first-of-duplicates, the null rule, the
relation to the pareto option of impedance_matrix, since 20.21.0, an example table. Linked from
Relational-functions.md (beside the join_equal_values family it replaces for this purpose), from the see-also of
Join_equal_values.md, and from the pareto section of Impedance-options.md. The page is not in nav/geodms.md of
the converter yet.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>