cuped

A Python package for the statistical analysis of A/B tests

因果推理&AB实验相关论文小书库

Causalis - State-of-the-art robust causal inference for experiments and observational data in python

Theory and implementatuon of various A/B experiments

Calibration-driven A/B testing toolkit: SRM, CUPED/CUPAC, delta-method ratio metrics, sequential and Bayesian methods, each validated by tests.

AB testing, stattests, poisson bootstrap, cuped, linearization

Monte Carlo audit of CUPED variance reduction and post-treatment bad controls

A/B testing statistics, design, increasing sensitivity, multiple experiments comparison, traffic splitting and full A/B testing pipeline in Python

Синтетические A/B-эксперименты: CUPED, взвешенные оценки, bias и покрытие интервалов.

896 tests / 42 scripts validating Zetyra calculators (GSD, CUPED, Bayesian, SSR incl. Single-Arm, RAR, Master Protocol). Replicates 8 published trials (Salk, DAPA-HF, PACIFIC, MONALEESA-7, I-SPY 2, STAMPEDE, REMAP-CAP, NCT03377023) + Leyrat 2024.

A/B testing engine: frequentist + Bayesian + CUPED variance reduction + peeking-safe sequential testing

A/B testing engine with always-valid sequential inference (mSPRT) — peek continuously without inflating false positives. CUPED variance reduction, SRM detection, guardrails. FastAPI + React.

End-to-end A/B test analysis on real experiments: a 64,000-customer e-mail test taken from analysis plan to rollout, and 22,743 Upworthy headline tests read as a testing programme. Demonstrates pre-registration, sample-ratio checks, CUPED-style adjustment, sizing a test and its holdback, the winner’s curse, and stakeholder communication.

Тыкаюсь в темах интенсива A/B-week от ШАДа

A Python package for the full A/B test lifecycle, from power analysis to sequential monitoring and variance reduction

RetailLab: should a retailer keep sending its marketing e-mail? Prices a 64k-customer randomized experiment at sector margins into Ship / Don't Ship / Keep Testing / Don't Trust.

An end-to-end A/B testing program review: 6 experiments, 2 ships, and the decision-quality framework behind them

End-to-end A/B experimentation toolkit: experiment readouts, MDE & power calculations, CUPED variance reduction, and decision-focused reporting in a Big Tech economist style.

A/B test readout for a UPI cashback: power, SRM root-cause, CUPED, Holm-corrected segments, novelty, guardrails and an ROI rollout decision

Four free calculators for A/B tests and marketing experiments (sample size, minimum detectable effect, peeking, geo holdout) and an interactive CUPED explainer. Each result comes with one plain sentence on what it means and what would invalidate it. Every formula is checked against scipy, statsmodels and numpy on every push.