Hands-on: ML to ABM Integration

Day 3 | Connecting Predictions to Simulation

Open in Google Colab

Click the badge below to open this notebook directly in Colab:

Open In Colab
TipSave Your Work

After opening, go to File > Save a copy in Drive to save your progress.

Learning Objectives

In this hands-on session, you will:

  1. Load your trained ML model from Day 2
  2. Generate demand predictions for future periods
  3. Format predictions as ABM input parameters
  4. Run a simplified supply chain simulation
  5. Analyze simulation outputs

Prerequisites

  • Completed Day 2 notebooks (trained demand forecasting model)
  • Understanding of modeling concepts from today’s lecture

Exercise Overview

Part 1: Generate Predictions (20 min)

  • Load the trained model
  • Create feature data for future periods
  • Generate demand forecasts
  • Visualize predictions

Part 2: Format for ABM (20 min)

  • Structure predictions by facility and time
  • Create agent parameter files
  • Validate data format

Part 3: Simple ABM Introduction (30 min)

  • Understand the ABM structure
  • Define agent behaviors
  • Run baseline simulation
  • Visualize supply chain dynamics

Part 4: Integration (20 min)

  • Connect ML predictions to ABM inputs
  • Compare scenarios:
    • Static demand estimates
    • ML-based demand forecasts
  • Analyze impact on wastage/shortages

The Integration Pipeline

# 1. Load trained model
import joblib
model = joblib.load('demand_model.pkl')

# 2. Generate predictions
future_features = create_future_features(start='2026-03', periods=12)
predictions = model.predict(future_features)

# 3. Format for ABM
agent_params = format_for_abm(predictions, facilities)

# 4. Run ABM with predictions
simulation = SupplyChainABM(agent_params)
results = simulation.run(steps=365)

Key Concepts

Agent Parameters

Each facility agent needs:

  • Expected demand: ML predictions by month
  • Order policy: When and how much to order
  • Capacity constraints: Storage limits
  • Lead time expectations: Delivery delays

Connecting the Pieces

The ML model provides demand forecasts that inform:

  1. Order quantities: How much to request
  2. Order timing: When to place orders
  3. Buffer stock: Safety margins based on forecast uncertainty

After This Session

You will have:

  • Experience with the ML-ABM integration workflow
  • Understanding of how predictions parameterize simulations
  • Foundation for Day 4’s deeper ABM exploration
  • Insight into the WISE project modeling approach