berkeley-stat159 / project-theta

UC Berkeley Stat 159/259 Fall 2015

Project-Theta

Group members: Siyao Chang (changsiyao), Boying Gong (boyinggong), Benjamin Hsieh (BenjaminHsieh), Brian Qiu (brianqiu), Jiang Zhu (pigriver123)

Topic: [The Neural Basis of Loss Aversion in Decision Making Under Risk] (https://openfmri.org/dataset/ds000005/)

The original Science paper found here was written by Sabrina M. Tom, Craig R. Fox, Christopher Trepel, and Russell A. Poldrack at UCLA. The main aim is to identify regions of brain related to loss aversion in decision making (in this case a gambling task of 50/50 chance of gaining/lossing money) and investigate the behavorial and neural correlates of loss aversion. In our project we aim to recreate a subset of the paper results with the goal of of checking model assumptions, creating analysis pipelines of fMRI research, and comparing results with Tom's (et al) paper. Please see paper report in project-theta/paper for further discussion.

Instructions

To generate paper please navigate to paper directory and make all. Follow the instructions below to reproduce/update analyses figures.

Steps to run all analyses:

Note: These steps, in particular make all, can take dozens of hours due to the size and nature of the analyses. We do not recommend running it but instead follow the instructions below Alternate Steps for specific desired results.

• make resultfolder will generate the results folder project-theta/results to save all analysis output and figures generated from running scripts. Please note that this folder will be git ignored (exist on local repository). If one decides to make all or run any analyses in scripts,
  run this command beforehand.
• make all will generate everything related to the analyses. It will pull in and validate data inside the data directory, create images saved inside the results directory, move relevant images to the paper/figures subdirectory, and generate the final report report.pdf inside the paper directory. The process can take a VERY long time as the data is over 16 GB and the analyses take quite a while as well. To save time one can view the final report without running this command by going to the paper directory and running make all from there since the needed figures are already preloaded there.

Alternate steps for analyses:

• First, make data to download, unzip, and validate data, saving datasets to the data directory. Note: Run time around 30 minutes.
• Second, make resultfolder to initialize analyses output directory.
• Navigate to code/scripts and run the specific desired analysis.
• Navigate to code directory and make paperfig to update paper figures
• Navigate to paper directory and make all to generate paper.

Other Utility Commands:

• make clean will delete extraneous files created by our scripts
• make coverage will check the coverage of our data, code directories
• make test will run the tests in our code/utils/tests and code/data/tests directory
• make verbose is the verbose version of 'make test'

Directories

• code contains the functions and scripts for analysis and graphing
• data holds the relevant datasets that are pulled in and validated either through make commands there or make all here.
• paper contains the final report, report.pdf, once the appropriate make commands are run, either here or in paper. It also has the relevant figures for the paper.
• slides contain our presentation slides used in class (Stat 159/259).
• requirements.txt is the a text file listing the required package versions.
• tools contains the material related to travis ci.

Acknowledgments

Many thanks to Jarrod Millman, Matthew Brett, J-B Poline, and Ross Barnowski for their advice and efforts that made this project possible.