Home Original page

GitHub - ErrorLSC/QMS: Lite weighted DRP/MRP system

A pragmatic, data‑driven supply planning toolkit for demand classification, smart lead times, safety stock, ETA prediction, and replenishment (MRP / DRP). Designed for messy ERP data and real‑world processes.

✨ Key Features

  • Demand Classification — steady / intermittent / burst / trended / seasonal, with replacement & stock‑only handling.
  • Forecasting — Croston and other strategies via a registry; writes unified forecast artifacts.
  • Safety Stock — pluggable strategy registry; service‑level driven, robust to outliers.
  • Smart Lead Time — merges vendor transport stats, preferences, and prepare time with layered fallbacks.
  • ETA Transformer — five path splits (confirmed / shipped / tail-batch / batch‑prone / single‑batch), virtual sub‑lines, overdue logic.
  • Transport Mode Intelligence — correct / predict transport mode with group penalties and static caps.
  • Route Selector — candidate (vendor + mode) selection with cost/lead‑time tradeoffs; supports quotes and standard cost.
  • Freight Charge Analyzer — currency normalization, sanity checks, and cost-per‑kg modeling (e.g., regression with non‑negative constraints).
  • MRP / DRP — vectorized standard MRP; output includes algorithm tag & transport mode; dynamic MRP planned.

Built around functional, composable components, with optional OOP facades for pipelines/jobs.

🧠 Core Concepts

Demand Classification

  • Handles replaced items, stock‑only, new items, and main classes: steady / intermittent / burst / trended / seasonal.
  • Winsorization, weighted stats, decomposition for trend/seasonality.

Forecasting

  • Strategy registry dispatches by demand type (e.g., Croston variants for intermittent).
  • Unified outputs (series, monthly aggregates, model_used) stored via ItemForecastRecord.

Safety Stock

  • Pluggable strategies (e.g., service‑level, variability‑aware) under forecast/safety_stock/.

Smart Lead Time

  • Combines TransportPreference, VendorTransportStats, PrepareLTStats, and static caps.
  • Lower/upper bounds = min across stats and static, robust fallbacks, no 999 magic numbers.

ETA Transformer

  • Five flow types; supports virtual sub‑lines, tail‑batch simulation, and overdue re‑assessment (Q60→Q90 fallback).

Transport Mode Intelligence

  • CorrectTransportMode / PredictTransportMode with group‑switch penalties; bans intra‑group flips.

Route Selector

  • Candidate routes assembled from historical preference + vendor stats + quotes.
  • Strategy: lowest total cost (unit price + freight), or shortest lead time. Prepare time integration is planned.

Freight Charge Analyzer

  • Ingests PO_FREIGHT_CHARGE & MultiCurrency; monthly FX; non‑negative regression to split base & per‑kg costs.

MRP / DRP

  • Static MRP implemented; writes algorithm tag and transport mode to MRP_ORDERS.
  • Dynamic MRP planned to incorporate dynamic lead times.

🛠 Usage Examples

Examples assume you’ve configured DB connections and have access to the required tables.

from qms_core.forecast.classifier import DemandClassifier
from qms_core.mrp.calculator import MRPCalculator

item = Item(ITEMNUM="ABC", Warehouse="TOKYO")
DemandClassifier().classify(item)
mrp = MRPCalculator(algorithm="Static")
orders = mrp.run(item)

CLI / Jobs (illustrative)

building

🗃️ Data Inputs (typical)

  • DemandHistoryWeekly, ItemMaster (IIM), Inventory, VendorMaster
  • VendorTransportStats, ItemTransportPreference, ItemPrepareLTStats
  • PO_DeliveryHistoryRaw, PO_FREIGHT_CHARGE, MultiCurrency

🔌 Extensibility

  • Strategy Registries — add new forecasting or safety‑stock strategies without touching callers.
  • Analyzers — drop‑in analyzers (e.g., freight, delivery behavior) with common base class.
  • Pipelines/Jobs — ETL‑style prepare → process → write lifecycle with dry‑run & smart writer support.

📈 Roadmap

  • Dynamic MRP with path‑aware ETA
  • Prepare‑time integration in RouteSelector
  • Cost + lead‑time multi‑objective path selection
  • CI workflow & data‑contract checks

🤝 Contributing

  1. Fork + branch: feat/* or fix/*
  2. Format & lint: ruff, black
  3. Add tests where meaningful
  4. PR template & checklist

🪪 License

TBD (MIT recommended). If you need dual‑license or exceptions for client data, open an issue first.

⚠️ Security

  • Never commit credentials or raw client data.
  • Use .env and secrets managers; rotate keys on any suspicion.
  • If sensitive files were accidentally committed, remove with git rm --cached and rotate secrets; for full history rewrite consider git filter-repo/BFG.

🙋 FAQ

Q: Can I run QMS without a live ERP?
A: Yes, by pointing to local CSV/SQLite mirrors and stubbing extractors.

Q: Does it support partial batches and tail‑batch ETAs?
A: Yes — see ETATransformer’s tail‑batch simulation and overdue heuristics.

🧭 Credits

Designed for messy, real‑world supply chains. Built with love for functional composition and clean, testable code.