Thompson Sampling

Multi-armed bandit solver using Thompson Sampling for ad click-through rate (CTR) optimization.

What It Does

Simulates an online ad-selection scenario where an algorithm must decide which of 10 ads to show in each round, learning from binary rewards (click / no click) to maximise total clicks over 10,000 rounds.

Methodology

Thompson Sampling is a Bayesian approach to the exploration–exploitation trade-off:

1. Model — each ad's unknown CTR is modelled with a Beta distribution: Beta(α, β) where α = successes + 1 and β = failures + 1.
2. Sample — draw a random value from every ad's current Beta distribution.
3. Select — pick the ad whose sample is highest.
4. Update — observe the reward and increment α (click) or β (no click).

Because the Beta distribution naturally narrows as evidence accumulates, the algorithm explores uncertain ads early on and gradually exploits the best performer.

Dataset

Ads_CTR_Optimisation.csv — 10,000 rows × 10 columns. Each cell is a binary reward (1 = click, 0 = no click) from a simulated ad campaign.

Dependencies

Language	Packages
Python 3	matplotlib, pandas
R	base (no extra packages)

How to Run

🐍 Python

pip install -r requirements.txt
python thompson_sampling.py

📊 R

Rscript thompson_sampling.R
# or inside an R session:
source("thompson_sampling.R")

Both scripts print the total reward and display a histogram showing which ad the algorithm converged on.

Tech Stack

🐍 Python 3	Core implementation
📊 R	Alternative implementation
📈 Matplotlib	Visualisation
🐼 pandas	Data loading
📐 Beta Distribution	Bayesian prior/posterior

Known Issues

• Dataset is fully simulated; real-world CTRs would require a live feedback loop.
• No command-line arguments yet — round count and ad count are constants at the top of each script.
• The R script's sys.frame(1)$ofile path resolution only works when the file is source()-d; when run with Rscript it falls back to getwd().

License

MIT