Data and models for "Discovering dynamical models of human behavior"

Ebb and Flow gameplay data and trained model parameters for:

Jaffe PI, Poldrack RA, Schafer RJ, Bissett PG. 2022. Discovering dynamical models of human behavior. bioRxiv. 

Ebb and Flow is a task-switching game offered as a part of the Lumosity cognitive training platform (Lumos Labs, Inc.). The data and model parameters are organized by participant/model in individual archived directories (140 participants; 165 models). Within each model directory, “data_pre_split.pickle” contains the processed Ebb and Flow data used for model training. The holdout data used to evaluate model performance is stored in “test_model_inputs.pt” and “test_other_data.pkl”. The parameters from the trained model are stored in “model_params.pth”. Some intermediate analysis products are contained in the subfolder “model_analysis”.

Metadata for all models can be found in “model_metadata.csv”. The metadata field “switch_cost_type” identifies models that were trained on data with (sc+) or without (sc-) a switch cost (note that models marked “NA” were also trained on data with a switch cost but were not included in the paired comparison of the sc+ and sc- models; see manuscript for details). The other metadata fields are self-explanatory.