This is a repository associated with a project on automatically discovering and describing planning strategies used by people. Our computational method for enabling this process is described in an article published in Behavior Research Methods journal (Skirzynski, Jain, & Lieder, 2023) and available at https://link.springer.com/article/10.3758/s13428-023-02062-z.
Navigate to python folder to learn more about Human-Interpret, a method for that takes data from a process-tracing experiment and returns descriptions of planning strategies that people used in this experiment. Human-Interpret finds strategies by performing probabilistic clustering on the process-tracing data, transforming those clusters into procedural instructions with AI-Interpret (imitation learning method; Skirzynski, Becker, & Lieder, 2022) and DNF2LTL (obtaining procedural descriptions from flowcharts; Becker, Skirzynski, van Opheusden, & Lieder, 2021), and applying Bayesian model selection on multiple iterations of this procedure.
Navigate to the data folder in order to see the data gathered in process-tracing experiments that externalized human planning. There are 5 available experiments:
- "v1.0" - 3-step task with increasing variance rewards (standard Mouselab task); expert reward: 39.97; participants: 180
- "F1" - Same as "v1.0"
- "c1.1" - 3-step constant variance task (rewards in [-10, -5, 5, 10]); expert reward: 9.33; participants: 62
- "T1.1" - 5-step transfer task. Only the test block of this experiment is to be used.; expert reward: 50; participants: 120
- "c2.1_dec" - 3-step task with decreasing variance rewards; expert reward: 30.14; participants: 35