How arnaudbarth/autonomous-dynamic-pricing works
This project simulates ticket sales for concerts and trains reinforcement-learning agents to choose prices over a sale horizon, comparing them with a fixed-price baseline. Its workflow is reference-data preparation, buyer and market simulation, a Gymnasium decision interface, agent training or inference, then paired evaluation across shared scenarios. The README also documents ChartMetric artist-data retrieval and OFS demographic calibration. Source excerpts show key simulator and agent links; some data-pipeline and Q-table training details are documented but not sampled, so those connections are limited to documented or indexed evidence.
Main entry points: run_comparison.py Comparison runner · gym_wrapper.py Gym environment · registry.py Model registry · train_dqn.py DQN training
Components 20
Data calibration
- ChartMetric client
env/chartmetric.py - Artist data fetch
scripts/fetch_chartmetric.py - Reference profiles
env/refdata.py - Calibrated datasets
data/upsert
Ticket simulation
- Virtual buyers
env/agents.py - Ticket market
env/market.py - Gym environment
env/gym_wrapper.py
Pricing agents
- Fixed-price agent
pricing_agent/agent_fixed.py - Q-table agent
pricing_agent/agent_qtable.py - DQN agent
pricing_agent/agent_dqn.py - Price action mapping
pricing_agent/base.py - Model registry
pricing_agent/registry.py - Trained models
models/pricingAgents - DQN primitives
pricing_agent/dqn.py
Training evaluation
- Q-table training
scripts/train_qtable.py - DQN training
scripts/train_dqn.py - Comparison runner
scripts/run_comparison.py - Paired evaluation
pricing_agent/evaluate.py
Other components
- Event organizer
- ChartMetric API
Connections 25
- Event organizer → Comparison runner: runs comparisons
README.md - Artist data fetch → ChartMetric client: uses client
scripts/fetch_chartmetric.py - ChartMetric client → ChartMetric API: fetches data
- Artist data fetch → Calibrated datasets: builds artist data
README.md - Calibrated datasets → Reference profiles: calibrates profiles
README.md - Reference profiles → Ticket market: provides profiles
env/market.py - Reference profiles → Virtual buyers: provides parameters
env/agents.py - Ticket market → Virtual buyers: solicits buyers
env/market.py - Gym environment → Ticket market: steps market
env/gym_wrapper.py - Fixed-price agent → Price action mapping: maps prices
pricing_agent/agent_fixed.py - Q-table agent → Gym environment: discretizes states
pricing_agent/agent_qtable.py - Q-table agent → Model registry: loads model
pricing_agent/agent_qtable.py - DQN agent → Model registry: loads model
pricing_agent/agent_dqn.py - Model registry → Trained models: resolves versions
pricing_agent/registry.py - Q-table training → Gym environment: trains in
scripts/train_qtable.py - Q-table training → Model registry: saves model
scripts/train_qtable.py - DQN training → Gym environment: trains in
scripts/train_dqn.py - DQN training → DQN primitives: uses primitives
scripts/train_dqn.py - DQN training → Model registry: saves checkpoints
scripts/train_dqn.py - Comparison runner → Gym environment: creates environment
scripts/run_comparison.py - Comparison runner → Fixed-price agent: compares baseline
scripts/run_comparison.py - Comparison runner → Q-table agent: compares policy
scripts/run_comparison.py - Comparison runner → DQN agent: compares policy
scripts/run_comparison.py - Comparison runner → Paired evaluation: runs evaluation
scripts/run_comparison.py - Paired evaluation → Gym environment: runs episodes
pricing_agent/evaluate.py