A computational account of threat-related attentional bias
Toby Wise, Jochen Michely, Peter Dayan & Raymond J Dolan
PLoS Computational Biology, 2019
Analyses for this project are contained within Jupyter notebooks (in the
/notebooks directory). Python 2.7 was used for all these analysese, and the code will most likely not run smoothly with Python 3.
These notebooks make use of scripts located in the
/code directory and standard Python packages (e.g. numpy, pandas, matplotlib etc).
In addition, they require a package called DMPy (this package is half-written and doesn't have much functionality beyond that required for this project). Eye-tracking analysis is mostly done using a fork of PyeParse that allows import of iohub eyetracking files.
The majority of the analysis reported in the paper is run in a series of Jupyter notebooks, which are located in the
/notebooks directory. The only exception is the fitting of behavioural models, which was run on a HPC cluster for speed (code for this model fitting is provided in the
There are 5 notebooks, each of which runs a specific section of the analysis pipeline and produces all the figures etc. associated with it.
This notebook prepares raw behavioural data and constructs computatioanl models. It also runs analysese of the results of model fitting.
This notebook prepares eye tracking data and extracts fixations. It then runs analyses exploring effects of model-derived measures on attention.
Attention effects on learning
This notebooks performs analyses looking at how attention affects learning.
This notebook runs regression analyses examining relationships between questionnaire measures of state/trait anxiety and behavioural variables.
This runs some quick parameter recovery checks for the best fitting models.
All data is available on the Open Science Framework here.
Eye tracking data is compressed as
.gz files and so will need to be extracted before use. The analysis code expects to find all the data in a directory called
Processed data is also provided, including fitted models simulated data from these models.
Data is available for 63 subjects - two subjects were excluded prior to analyis as they started the task but did not complete it.