gym-inventory

The Inventory environment is a single agent domain featuring discrete state and action spaces. Currently, one task is supported:

Simple Inventory Control with Lost Sales

This environment corresponds to the version of the inventory control with lost sales problem described in Example 1.1 in Algorithms for Reinforcement Learning by Csaba Szepesvari (2010).

Future tasks

Future tasks will have more complex environments that take into account:

• Demand-effecting factors such as trend, seasonality, holidays, weather, etc.
• Business-specific factors such as lead times, penalties for carrying inventory, etc.
• Scaling from 1 to millions of SKUs and learning across SKUs.

Installation

cd gym-inventory
pip install -e .