Sat 13 Jan 2018 14:00 - 14:30 at Crocker - Session 2 Chair(s): William E. Byrd

We introduce SweetPea, a SAT-sampler aided language that addresses problems scientists face when creating randomized experimental designs. SweetPea provides a high-level interface to declaratively describe an experimental design, and a low-level synthesizer to generate unbiased sequences of trials given satisfiable constraints. Our prototype provides a rich set of primitives that closely match natural descriptions of experimental designs, allowing researchers to concisely describe what analysis they wish to perform. To ensure statistically significant results, every possible sequence of trials that satisfies the design must have an equal likelihood of being chosen for the experiment. SweetPea samples sequences of trials by compiling experimental designs into Boolean logic, which are then passed to a SAT-sampler. The SAT-sampler Unigen provides statistical guarantees that the solutions it finds are approximately uniformly probable in the space of all valid solutions.

Abstract (obt18-paper13.pdf)361KiB