gggg03/Dynamic_Pricing_Optimization

P02 — Dynamic Pricing Optimization (RL)

Retail pricing simulation with Q-Learning (discrete prices) and DDPG (continuous prices, PyTorch), plus a live dashboard, competitor tracker, profitability comparison, and sensitivity analysis.

Project structure

Quick start (Windows)

cd "c:\Users\PMLS\Desktop\rl project"
pip install -r requirements.txt
run_project.bat

Or step by step:

pip install numpy flask torch

# Step 1 — test environment
python -c "from env_and_agents import RetailPricingEnv; e=RetailPricingEnv(42); print(e.reset().shape)"

# Full pipeline (Q-Learning + DDPG training, sensitivity, comparison, report)
python train_all.py

# Or run steps individually:
python step2_train_ql.py          # Q-Learning training → ql_agent.npz
# (DDPG trains inside train_all.py — step 3)
python step5_compare.py           # Comparison → comparison.json
python build_dash_data.py         # Aggregate JSON → dash_data.json
python build_dashboard.py         # Build dashboard HTML
python generate_report.py         # Write REPORT.md

# Live dashboard (real env + trained agents — do NOT open as file://)
python dashboard_server.py
# Then open http://127.0.0.1:5050

Assignment mapping

Environment details

• —Discrete actions: 0.8, 0.9, 1.0, 1.1, 1.2 × base price ($100)
• —Continuous (DDPG): multiplier ∈ [0.7, 1.5] via trained Actor network
• —State: inventory, price, competitor price, hour, day, 7-day demand avg, product age %
• —Reward: (revenue − COGS) − holding cost − stockout penalty − volatility penalty
• —Demand: price-elastic vs competitor; peak-hour boost; weekend boost

DDPG architecture

Outputs

After python train_all.py:

Live dashboard strategies

The live simulator (python dashboard_server.py) supports:

• —fixed — always $100
• —rule — heuristic rule agent
• —ql — trained Q-Learning agent (requires ql_agent.npz)
• —ddpg — trained DDPG agent (requires ddpg_best.pth)
• —random — random discrete action