Pre-Adaptation Tuning

Code for "Prior Sensory Tuning Orients Error Processing During Sensorimotor Adaptation." Implements the three Bayesian state-space models (M1 Coupling, M2 Dissociation, M3 Dissociation + Decomposition) and the parameter / model recovery pipeline.

[Link to paper — TBD]

Setup

Requires Python 3.12 and Poetry.

poetry config virtualenvs.in-project true
poetry env use 3.12
poetry install

Reproduce paper figures

The repository ships fitted posteriors and recovery results — see data/README.md. The figure script runs end-to-end on shipped data:

poetry run python scripts/analysis/fig34.py

Writes fig3.png (β_R) and fig4.png (β_cog) to figures/.

Re-fit from raw data

Raw and trial-level CSVs are not redistributed (see data/README.md for the expected schemas). With those files in place, the full pipeline is:

State-space models (M1 / M2 / M3)

Requires data/adaptation_trials.csv.

# M1 (Coupling)
poetry run python scripts/modeling/fit_m1.py --plateau-group-specific --save-posterior

# M2 (Dissociation) and M3 (Dissociation + Decomposition)
poetry run python scripts/modeling/fit_m2_m3.py --plateau-group-specific --save-posterior

# Parameter recovery and model recovery → Table 2
poetry run python scripts/recovery/parameter_recovery.py
poetry run python scripts/recovery/model_recovery.py

Localisation HLM (Fig. 2)

Heteroscedastic complete-pooling Bayesian HLM in Stan via cmdstanpy. Requires scripts/complete_pooling/_preprocessed/pa_long_trials.csv and a working CmdStan installation (cmdstanpy.install_cmdstan() once).

cd scripts/complete_pooling
poetry run jupyter nbconvert --execute --inplace completepooling_pipe.ipynb
poetry run jupyter nbconvert --execute --inplace completepooling_vis.ipynb

completepooling_pipe.ipynb fits the CP1–CP8 family and writes posteriors to ./results/CP{L}/posterior.nc plus CP_model_comparison_loo.csv. completepooling_vis.ipynb reads CP4 / CP7 / CP8 posteriors and renders the μ and σ panels behind Fig. 2.

Contributors

License

See LICENSE.