developmental_trajectory

Open Access Journal Article:

https://www.nature.com/articles/s41562-023-01662-1

Cite as:

Giron, A. P., Ciranka, S., Schulz, E., van den Bos, W., Ruggeri, A., Meder, B., & Wu, C. M. (2023). Developmental changes in exploration resemble stochastic optimization. Nature Human Behaviour. https://doi.org/10.1038/s41562-023-01662-1

Preprint:

https://charleywu.github.io/downloads/giron2023developmental.pdf

Datasets:

• data/behavioralData.csv: behavioral data of all 281 participants
• data/modelFit.csv: parameter estimates of all models for all participants
• data/modelFit_OriginalID: parameter estimates of all models for all participants with original IDs from the Schulz et al and Meder et al (needed for recovery analyses)
• data/paramsYoungestAgegroup.csv: cross-validated parameter estimates of the youngest age group (used as starting points for the optimization algorithms)
• data/smoothKernel.json: all 40 environments, from which a new environment was chosen in each round without replacement. Also used for model and parameter recovery

Scripts:

• dataProcessing.R: import and pre-process behavioral data and parameter estimates (added for reference, since we already include the generated outputs instead of the inputs)
• statisticalTests.R contains wrapper functions for all statistical tests used for the analyses
• behavior_tests.R and behavior_plots.R: analyze and plot participant's behavior in the multi-armed bandit task
  (generates Figure 2)
• crossvalidation.R: optimize parameters of the GP-UCB model and the lesioned models. All models being fit here are defined in models.R.
• learningCurves.R: simulate learning curves
• PXP.ipynb compute and save the protected exceedance probability (pxp) for all models and age groups. Functions to compute the pxp are defined in bms.py and files containing the negative log likelihoods that are imported in the notebook are created in modelResults_tests.R.
• modelResults_tests.R and modelResults_plots.R: analyze and plot model results. Therefore, simulated learning curves from learningCurves.R and pxps from PXP.ipynb are imported.
  (generates Figure 3 and S2)
• reliabilityChecks_tests.R and reliabilityChecks_plots.R: compare participant's performance in the multi-armed bandit task and model results across experiments
  (generates Figure S1)
• simulateModels.R and simulateModels_plots.R: simulate the GP-UCB model with different parameter combinations and plot expected rewards
  (Generates Figure S5)
• hillClimbingAlgorithm.R, hillClimbingAlgorithm_tests.R and hillClimbingAlgorithm_plots.R: run optimization algorithms in the parameter space calculated in simulateModels.R
  (generates Figure 4 and S6)
• Algo-Human-Permute.R: compute the changepoint analysis of human and SHC-fast trajectory

Model and Parameter Recovery

• Recovery/Model_Recovery.Rmd import, analyze and plot model recovery results generated with Recovery/Model_Recovery_Cluster.R, Recovery/Model_Recovery_Cluster_Meder.R and Recovery/Model_Recovery_Cluster_Schulz.R and saved in Recovery/modelRecovery/. Models used for recovery are defined in fit_parallel_cluster.R.
  (generates Figure S3 )
• Recovery/Parameter_Recovery_check.R import, analyze and plot parameter recovery results generated with Recovery/Parameter_Recovery_Cluster.R and saved in Recovery/parameterRecovery/. Models used for recovery are defined in Recovery/fit_parallel_cluster.R.
  (generates Figure S4)