Description

This notebook is designed for quick, high‑quality submissions in VN2. It prioritizes actionability over demo:

Feature-driven routing

• Computes per‑SKU features: mean demand, zero fraction (intermittency), STL seasonal strength (period=52), and series length.
• Routing rules:

1. Intermittent (zeros high, mean low) → Croston‑SBA
2. Seasonal (high seasonal strength, sufficient length) → TBATS
3. Otherwise → SES

• Per‑SKU demand and uncertainty

1. Estimates μ̂ from short‑horizon forecasts (or series mean for Croston).
2. Estimates σ̂ from residuals (fallback to series std when needed).

• Base‑stock policy and submission

1. Uses protection period P=3 (lead 2 + review 1).
2. Computes orders from demand_stats_modelled and current state built from Initial State (on_hand = End Inventory, on_order = W+1 + W+2).
3. Writes submissions/orders_round1_advanced_modelled.csv with the exact platform index order.

• Clean outputs and inline docs

1. Inline markdown before each step explains the logic; warnings are suppressed to keep outputs readable.

How to use

• Run top‑to‑bottom; the last cell prints the path to the generated CSV.
• Modify routing thresholds (ZERO_FRAC_THR, LOW_MEAN_THR, SEASON_STRENGTH_THR, MIN_LEN_TBATS) to fit your portfolio.
• Repository: http://github.com/senoni-research/vn2inventory.