English · Türkçe
Assembly line balancing (SALBP-1) with Hoffmann's (1963) precedence-matrix method. Pure-Python core with no mandatory dependencies; command-line tool and a Streamlit interface available in English, Türkçe, Deutsch, Español, Français.
Live demo: https://hoffmann-line-balancing.streamlit.app/ — no installation needed.
$ python -m hoffmann_lb examples/course_example.csv --cycle 18
Method: hoffmann Cycle time: 18
Station 1: 1, 2 load 18 idle 0
Station 2: 3, 4, 5, 7, 9 load 18 idle 0
Station 3: 6, 8 load 17 idle 1
Station 4: 10, 11, 12 load 17 idle 1
Stations : 4 (equals lower bound)
Lower bound : 4
Total time : 70
Total idle : 2
Efficiency (E) : 97.22 %
Balance delay (D) : 2.78 %
Smoothness (SI) : 1.4142 [ref: max_load]
git clone https://github.com/DailyDana/hoffmann-line-balancing.git
cd hoffmann-line-balancing
python -m venv .venv && source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e . # core + CLI
pip install -e ".[app]" # + Streamlit interface
pip install -e ".[dev]" # + ruff, mypy, pytest, hypothesispip install -e ".[app]"
streamlit run app/streamlit_app.py- Language selector in the sidebar (EN / TR / DE / ES / FR).
- Task table (
st.data_editor): duration and successor list; row count follows "Number of tasks". - Solve → metric cards, station table, CSV/JSON download.
- "Compare with legacy method (greedy)": both solutions side by side.
- Cycle-time sweep: station count / efficiency table and chart over a range of C.
- CSV upload (
,or;delimiter, decimal comma).
python -m hoffmann_lb input.csv --cycle 15
python -m hoffmann_lb input.csv --cycle 15 --lang tr # tr, de, es, fr
python -m hoffmann_lb input.csv --cycle 15 --format json # full result + metrics
python -m hoffmann_lb input.csv --cycle 15 --format csv # station table only
python -m hoffmann_lb input.csv --cycle 15 --method greedy # legacy comparison
python -m hoffmann_lb input.csv --cycle 15 --tie-break most_tasks
python -m hoffmann_lb tr.csv --cycle 15 --delimiter ";" --decimal-comma # Turkish ExcelInput CSV (examples/course_example.csv):
id,duration,successors
1,12,2;3;4;5;6;7;8;9;10;11;12
2,6,3;4;5;6;7;8;9;10;11;12
3,6,7;8;11;12
...
12,7,Column headers may also be Turkish (İş No, Süre, Ardıllar). Successors within a
cell are separated by ; or whitespace.
from hoffmann_lb import PrecedenceGraph, Task, TieBreak, compute_metrics, hoffmann
g = PrecedenceGraph(
[Task(1, 4), Task(2, 3), Task(3, 3)],
edges=[(1, 2), (1, 3)], # (predecessor, successor)
)
result = hoffmann(g, cycle_time=7, tie_break=TieBreak.ORIGINAL)
metrics = compute_metrics(result, total_time=g.total_time)
for s in result.stations:
print(s.index, s.task_ids, s.load, s.idle)
print(metrics.n_stations, metrics.lower_bound, metrics.efficiency_pct)All errors derive from hoffmann_lb.BalancingError (CycleError,
InvalidDurationError, ParseError, GraphError); messages are English and name
the offending task ids.
For each station, every task subset that satisfies precedence is enumerated and
the one with minimum idle time is chosen (Hoffmann 1963). The tie-break rule is
selectable (ORIGINAL, MOST_TASKS, MOST_FOLLOWERS). Details, design decisions
and limitations: docs/ALGORITHM.md.
Metrics: efficiency, balance delay, theoretical lower bound ⌈Σt/C⌉, smoothness
index (Moodie & Young).
The first version of this project was a coursework assignment (MATLAB → Python port
→ Tkinter GUI; see archive/legacy/). A code review
(archive/LEGACY_REVIEW.md, in Turkish) showed that the
implemented loop was not Hoffmann's method but a first-fit heuristic ("take the
first available task whose code is zero"). Without subset enumeration it deviated
from the optimum ([4,4,3,3,3,3], C=10 → 3 stations; Hoffmann finds 2). It also
looped forever on precedence cycles, rejected fractional durations and silently
dropped invalid input.
This repository was written from scratch; the old behaviour is preserved as
greedy_first_fit for comparison.
- 127 tests: unit, end-to-end CLI, Streamlit
AppTestsmoke tests, i18n completeness, and hypothesis property tests on random DAGs. - Expected values for the course example were computed with an independent brute-force oracle and an exact bitmask-DP solver; Hoffmann reaches the exact optimum for every tested C.
- With the
ORIGINALrule, Hoffmann yields exactly the same station sets as the oracle on random graphs up to 8 tasks. ruff,mypy --strict, CI on Python 3.11 / 3.12.
pytest
ruff check . && ruff format --check .
mypy- M1 — core: model, validation, Hoffmann, greedy, metrics (
v0.1.0) - M2 — CSV/JSON, CLI, documentation (
v0.2.0) - M3 — Streamlit interface: table editor, cycle-time sweep, greedy comparison (
v1.0.0) - Multilingual UI and CLI (
v1.1.0)
Possible next steps: exact solver (branch-and-bound / MILP) for optimality checks, precedence-graph visualisation, SALBP-2 (fixed station count, minimise C).
Apache License 2.0 — see LICENSE. Releases up to v1.0.0 were published under MIT (see NOTICE).
English · Türkçe
Hoffmann (1963) öncelik-matrisi yöntemiyle montaj hattı dengeleme (SALBP-1). Saf Python çekirdek, zorunlu bağımlılık yok; komut satırı aracı ve İngilizce, Türkçe, Almanca, İspanyolca, Fransızca arayüzlü Streamlit uygulaması.
Canlı demo: https://hoffmann-line-balancing.streamlit.app/ — kurulum gerekmez.
git clone https://github.com/DailyDana/hoffmann-line-balancing.git
cd hoffmann-line-balancing
python -m venv .venv && source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -e . # yalnızca çekirdek + CLI
pip install -e ".[app]" # + Streamlit arayüzü
pip install -e ".[dev]" # + ruff, mypy, pytest, hypothesisstreamlit run app/streamlit_app.pyKenar çubuğundan dil seçilir (varsayılan İngilizce). İş tablosu, Hesapla, greedy karşılaştırma, çevrim süresi taraması, CSV/JSON indirme ve CSV yükleme mevcuttur.
python -m hoffmann_lb girdi.csv --cycle 15 --lang tr
python -m hoffmann_lb girdi.csv --cycle 15 --lang tr --format json
python -m hoffmann_lb girdi.csv --cycle 15 --lang tr --method greedy
python -m hoffmann_lb tr.csv --cycle 15 --lang tr --delimiter ";" --decimal-comma--lang yalnızca metin çıktısını etkiler; JSON ve CSV dilden bağımsızdır. Girdi
CSV'sinde sütun başlıkları Türkçe olabilir (İş No, Süre, Ardıllar); ardıllar
hücre içinde ; ya da boşlukla ayrılır.
Her istasyon için öncelik kısıtını sağlayan tüm iş alt kümeleri sayılır ve boş süreyi en aza indiren seçilir (Hoffmann 1963). Eşitlik kuralı seçilebilir. Ayrıntı, karar gerekçeleri ve sınırlamalar: docs/ALGORITHM.md (Türkçe).
Ölçütler: verimlilik, denge kaybı, teorik alt sınır ⌈Σt/C⌉, düzgünlük indeksi.
Projenin ilk sürümü bir ders ödeviydi (archive/legacy/). İnceleme
(archive/LEGACY_REVIEW.md) uygulanan döngünün Hoffmann
yöntemi değil first-fit sezgiseli olduğunu gösterdi; optimumdan sapıyor, öncelik
döngüsünde sonsuz döngüye giriyor, ondalık süre kabul etmiyor ve geçersiz girdileri
sessizce yutuyordu. Depo sıfırdan yazıldı; eski davranış greedy_first_fit olarak
karşılaştırma için korundu.
127 test (birim, CLI, Streamlit AppTest, i18n bütünlüğü, hypothesis özellik
testleri). Ders örneğinin beklenen değerleri bağımsız kaba-kuvvet oracle ve kesin
çözücüyle hesaplandı. ruff, mypy --strict, CI (Python 3.11 / 3.12).
Apache License 2.0 — bkz. LICENSE. v1.0.0'a kadar olan sürümler MIT ile yayımlanmıştı (NOTICE).
