Zyklus/Farm:
- neuer adaptiver Modus: grosse Startpopulation, die je Generation
schrumpft (Start/Schrumpfung/MinPopulation) und bei Konvergenz stoppt
(Geduld Generationen ohne Verbesserung >= Schwelle, max MaxGenerationen)
- Zyklus.laden erkennt adaptives Profil am Feld 'Start' (Modus adaptiv),
sonst klassisch (Abfolge/Anzahl); Farm._LaufStatisch/_LaufAdaptiv
- neue Trunkierungs-Selektion Farm.PopulationBegrenzen
- [Adaptiv]-Profil in zyklus.cfg (dokumentiert)
Entwicklungsgrafik:
- zweites Panel unter dem Verlauf zeigt die Anzahl erzeugter
Verteilungen im Pool je Schritt (gemeinsame Zeitachse)
Konfiguration:
- platz.cfg entfernt; zyklus.cfg/strafen.cfg werden direkt aus
$PLATZ_CFG geladen
- neuer Schalter --zyklus-art (Default Adaptiv) waehlt das GA-Profil
- web/main.py analog: direktes Laden, Profil via PLATZ_ZYKLUSART
Doku (README.md, CLAUDE.md) und Tests entsprechend aktualisiert;
Beispiele mit dem Adaptiv-Default neu berechnet.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sitzplatzverteilung.alsSVG() (libs/Strukturdaten.py) draws tables as
white circles and people as colored circles inside their table's
circle, colored by group membership. Table radius is derived from
the number of seats so all person-circles fit without touching; the
table-to-table grid spacing is derived from the largest table radius
so tables never overlap regardless of their Koordinaten values.
libs/platz.py gains a --tosvg switch that writes
<indir>/Sitzplatzverteilung.svg alongside the existing text/CSV
output.
The three unittest scenarios now each write their rendering to
doc/images/, and doc/docu_tests.md explains the three configurations
and what each test actually verifies (including why the "split
groups land on true neighbor tables" property only holds for a
specifically chosen seed in the circle scenario, not in general).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Covers three concrete seating layouts to exercise the GA's group
placement and splitting logic:
- three tables chained pairwise (2 seats each), three couples ->
no group gets split, optimal score of 0.
- four tables in a line (2 seats each), two groups of four -> each
group ends up split across one contiguous table pair (T1+T2 or
T3+T4), never mixed across the middle seam.
- six tables in a circle (2 seats each), four groups of three ->
each group is split exactly once (2+1), since no table has three
seats; with the fixed seed the split pairs are also true circle
neighbors.
Also documents how to run the suite in README.md, including the
PYTHONPATH workaround needed when a stray global env var shadows
the stdlib "tests" package name with an unrelated project's tests.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Michael Stangl