Real-World Examples

The examples/real_world_scenarios.py script contains copyable templates for common bandit use cases:

Recommendation

Compare recommendation modules using Bernoulli click rewards.

Ad Allocation

Use LinearContextualEnvironment and LinUCBPolicy when the best ad depends on user segment or request features.

Pricing

Simulate revenue-like continuous rewards across candidate price points.

Clinical Trials

Use Bernoulli rewards and Thompson Sampling for binary treatment outcomes.

Proxy or Server Selection

Treat latency savings, success rate, or quality score as rewards for routing decisions.

Non-Stationary Demand

Compare stationary policies with sliding-window policies under abrupt shifts or gradual drift.

Run the examples with:

python examples/real_world_scenarios.py

Each scenario intentionally includes a simple baseline. In your own experiments, keep that habit: a sophisticated policy should beat a random or simple policy before it is trusted.