Blueyonder Crystal Ball 2020 Hackathon Challenge
Demonstrate that a distributed Supply Chain Problem can be managed by AI Agents
Hackathon URL | Reinforcement Learning | Deep Deterministic Policy Gradient Algorithm
Components:
The problem statement was divided into three components -:
-
Component 1: Create a supply chain environment to train AI agents that play the supply chain management.
- [pdf] [environment.py] Environment formulation, to set up a working environment for the agent's seamless interaction.
-
Component 2: Create algorithm agents that follow a logic to manage item locations.
- [pdf] [policy.py] Created baseline (s, Q) policy, and used Bayesian Optimization from Facebook Ax Platform to determine the best set of parameters (such as safe stock levels at different warehouses, production level at the factory, etc.) for the supply chain problem.
-
Component 3: Use Machine Learning (Reinforcement Learning) to build a model that makes the same agent decisions.
- [pdf] [DDPG.ipynb] Using Deep Deterministic Policy Gradient Algorithm, we focused on determining the optimal policy, to maximize the return (profit for the supplier).
Notes:
- For DDPG, please run the IPython Notebook. For Bayesian Optimization, run
main.py.