Lab 08: AdX One Day

This lab introduces real-time bidding in advertising exchanges with one-day campaigns.

Game Overview

Type: Real-time bidding advertising game Players: 10 players Rounds: Single day with multiple user arrivals Stages: Single stage with real-time auction simulation

Games

AdX One Day Game

Actions: Bid bundles for market segments
State Space: Campaign information and user arrivals
Key Concept: Real-time bidding and campaign optimization

State Space

Observations

observation = {
    "campaign": {
        "id": 1,
        "market_segment": "MALE_YOUNG_HIGH_INCOME",
        "reach": 500,
        "budget": 500.0
    },
    "day": 1,
    "total_users": 10000
}

Actions

action = {
    "campaign_id": 1,
    "day_limit": 500.0,
    "bid_entries": [
        {
            "market_segment": "MALE_YOUNG_HIGH_INCOME",
            "bid": 2.5,
            "spending_limit": 100.0
        }
    ]
}

Rewards

Campaign performance payoffs:

# Profit calculation based on reach fulfillment
reach_fulfilled = min(total_impressions, campaign.reach)
profit = (reach_fulfilled / campaign.reach) * campaign.budget - total_spent
reward = profit

Game Structure

Stage Type

Single day simulation with multiple user arrivals
Real-time auctions - second-price auctions for each impression
Campaign constraints - budget and reach targets

Learning Opportunities

Bid optimization - set optimal bids for different segments
Budget allocation - distribute budget across market segments
Reach targeting - maximize reach within budget constraints

Testing

Local Testing

from core.engine import Engine
from core.game.AdxOneDayGame import AdxOneDayGame
from core.agents.lab08.random_agent import RandomAgent

my_agent = MyAgent("MyAgent")
opponents = [RandomAgent(f"Random{i}") for i in range(9)]

engine = Engine(AdxOneDayGame(num_agents=10), [my_agent] + opponents, rounds=1)
results = engine.run()

print(f"My score: {results[0]}")
print(f"Average opponent score: {sum(results[1:]) / len(results[1:])}")

Campaign Analysis

def analyze_campaign_performance(self):
    if hasattr(self, 'campaign_history'):
        for campaign in self.campaign_history:
            reach_rate = campaign['impressions'] / campaign['reach']
            budget_utilization = campaign['spent'] / campaign['budget']
            print(f"Campaign {campaign['id']}:")
            print(f"  Reach rate: {reach_rate:.2%}")
            print(f"  Budget utilization: {budget_utilization:.2%}")

Next Steps

Implement an AdX agent using the common patterns
Study real-time bidding to understand optimal strategies
Test different bidding approaches against various opponents
Compete against other students

Focus on understanding real-time bidding and campaign optimization!