Static Order Up To Level Simulation Starter NB With Polars/Numba. This can easily be made forecast driven!

A practical RL-style tuning notebook that wraps an organizer-like simulator to optimize ordering policies. It supports grid-tuned baseline, per‑Department Optuna tuning, and a feature‑driven policy search, each producing a ready‑to‑submit CSV in the platform’s index order.

The "official" code

Stabilize low‑data SKUs via partial pooling across Division/Department/ProductGroup with NB‑GLM + Empirical Bayes shrinkage, per‑segment variance, and CV‑tuned strength. Generates a submission using a newsvendor base‑stock policy.

An 13-week seasonal moving average combined with a forecast-driven policy with a 4 weeks coverage

A forward-looking graphic view of the inventory position and balance

A production‑ready notebook that routes each SKU to SES, Croston, or TBATS based on data features, estimates μ̂ and σ̂ per SKU, and generates a valid base‑stock submission CSV in the platform’s index order. Repo: http://github.com/senoni-research/vn2inventory.

Learn how to use MLForecast to leverage machine learning models for time series forecasting

Use deep learning for time series forecasting

Implementation of Deep RL for the VN2 Inventory Planning challenge

Format and preprocess your data for time series forecasting

End-to-end EDA for VN2 Order #1 with demand stats, availability analysis, ABC/XYZ segmentation, and a service-level sensitivity sweep; includes artifact exports and a ready-to-run CLI for generating a valid submission. Repo: https://github.com/senoni-research/vn2inventory.

Describes how to interpret the Inventory Snapshot

Handbook-grade tutorial demonstrating KumoRFM's zero-shot demand forecasting using relational graphs (sales → SKUs → products/stores). Forecast-only, no policies.

Use statistical models for time series forecasting and learn hierarchical reconciliation methods.

An ordering system that doesn’t just forecast — it understands relationships and anticipates demand.

This notebook demonstrates the use of StatsForecast and NeuralForecast methods to analyze time series data. It includes the methodology to identify the demand type of each time series, helping to understand patterns such as intermittent, smooth, or seasonal demand.

Can AI agents autonomously create state-of-the-art forecasting models? We're about to find out!

Weighted ensemble from AutoMFLES and Lightgbm by minimising a desired error using scipy optimise to reduce regression to the mean.