load_two_step_data#

load_two_step_data()[source]#

Load the adult two-step task dataset of Smid et al. (2022).

Returns:

pd.DataFrame: A pandas DataFrame with one row per trial and participant.

See also

cpm.applications.reinforcement_learning.HybridMBMF: hybrid model-based / model-free reinforcement learning model of the two-step task.

Notes

The data are from the adult group of Smid et al. (2022), who completed 140 trials of the deterministic two-step task of Kool et al. (2016). The task uses the original coding, so the columns need converting before they can be passed to HybridMBMF (see the two-step task example in the documentation).

The data were originally published by the authors in the Model-based_Model-free_Developmental repository on GitHub.

The dataset includes the following columns:

ppt: the participant number.
trial: the trial number.
block: the block number.
stake: the stake of the trial (1 or 5).
s1, s2: the starting state and the planet reached (1 or 2; 0 marks a missed trial).
choice: the chosen spaceship, numbered 1 to 4 across both starting states (-1 marks a missed trial).
stimuli_left, stimuli_right: which spaceships were displayed on the left and right.
points: the reward, scaled between 0 and 1.
rews_1, rews_2: the rewards both planets would have paid out, scaled between 0 and 1.
rt_1, rt_2: the response times at the first and second stage, in seconds.
timeout_1, timeout_2: whether the participant failed to respond in time at either stage.
missed: whether the trial was missed.
score: the running score.

References

Kool, W., Cushman, F. A., & Gershman, S. J. (2016). When does model-based control pay off? PLoS Computational Biology, 12(8), e1005090.

Smid et al. (2022). Computational and behavioral correlates of developmental changes in model-based/model-free decision-making. Developmental Science, e13380.

Examples

>>> data = load_two_step_data()
>>> print(data.head())