Tea: A High-level Language and Runtime System for Automating Statistical Analysis
summary
Tea is a high-level declarative language plus runtime that compiles specifications into constraint satisfaction to pick and run valid statistical tests automatically. It checks assumptions (normality, group counts), falls back to bootstrapped medians if nothing applies, and ships as an open-source Python package on pip. Evaluation on 12 textbook examples matches expert choices in 9 of 12 cases, selects more conservative tests when assumptions fail, and avoids common novice errors.
AI-written summary · about
chapters (11)
- 0:00Introduction to tea
- 0:21The challenge of statistics
- 2:35The gap in analysis tools
- 4:26Overview of the tea system
- 5:41Using the tea language
- 7:05Declaring program variables
- 8:53How tea works internally
- 10:41Benefits of the approach
- 12:20Evaluation of tea
- 13:12Democratizing data science
- 14:44Q&A session
related talks
46mAI Testing Beyond the Basics: Ensuring Truthful and Reliable Agents
45mUsing AI, statistics, and set theory to know what to test
41mFrom Lab To Life: Practical AI System Evaluation
15mCI in the Era of AI: From Unit Tests to Stochastic Evals
33mEvaluating Your AI Agent
26mAl and Software: The Good, The Bad, The Scary
45mWhere AI Meets Code
31mComputer, Program Theyself!