rmp-ci 0.3.0

Bayesian confidence intervals for RateMyProfessors ratings (inspired by 3Blue1Brown)

rmp-ci

Bayesian confidence intervals for RateMyProfessors ratings.

Inspired by 3Blue1Brown's video on binomial distributions and the beta distribution — why you shouldn't trust online ratings at face value.

The Problem

A professor with 2 ratings averaging 5.0/5 looks better than one with 200 ratings averaging 4.5/5. But the first professor's true quality is far more uncertain. This tool computes a Bayesian credible interval using the Dirichlet-Multinomial model to quantify that uncertainty.

Why Dirichlet?

The 3b1b video uses the Beta distribution, which requires collapsing star ratings into binary (positive/negative). This tool uses the Dirichlet distribution — the multivariate generalization of Beta — which models all 5 star categories simultaneously. This gives you a credible interval directly on the expected star rating (1–5) without any arbitrary cutoff.

• Prior: Dir(1,1,1,1,1) — uniform over all 5 categories
• Posterior: Dir(n₁+1, n₂+1, n₃+1, n₄+1, n₅+1)
• CI computed via Monte Carlo sampling from the posterior

Install

pip install rmp-ci
# or
uv tool install rmp-ci
# or
pipx install rmp-ci

Usage

rmp-ci "Chao Zhang" -s "Georgia Tech"
rmp-ci "Mark Riedl" -s "Georgia Tech"
rmp-ci "Lingming Zhang" -s "UIUC"

# JSON output for piping
rmp-ci "Chao Zhang" -s "Georgia Tech" --json | jq .bayesian

# Custom confidence level
rmp-ci "Chao Zhang" -s "Georgia Tech" -c 0.90

# More Monte Carlo samples for tighter estimate
rmp-ci "Chao Zhang" -s "Georgia Tech" -n 200000

Zero dependencies — pure Python stdlib.

License

MIT