This repository contains data and code used to produce the paper "Fitness tracking reveals task-specific associations between memory, mental health, and physical activity" by Jeremy R. Manning, Gina M. Notaro, Esme Chen, and Paxton C. Fitzpatrick.

Additional information on this project can be found here. The exact version of the experiment used to acquire our dataset via Amazon Mechanical Turk (sans credentials in config.txt and credentials.json) may be found here.

This repository is organized as follows:

root
├── code: all analysis code used in the paper
├── data: all data analyzed in the paper
│   └── task: reference files used to analyze the data
├── docker-setup: configuration files for the Docker image
├── paper: all files needed to generate the main paper and supplement
│   ├── figs: PDF files for all figures in the main paper and supplement
│   │   ├── source: source images used to build the final figures
├── ref: a screen cast of the full experiment
│   └── sandbox_screenrecording_062618.mov

Note that compiling the PDFs for the main text and supplement will require you to set up the Contextual Dynamic Laboratory's Bibliography Management Tool.

To run the included code, used for the analyses presented in our paper, please follow the instructions below. Note: we've been inconsistent with our internal naming systems for this project; at different points in the project's development and implementation, we have referred to it as FitBrain, BrainFit, and/or FitWit.

This package requires Docker (v3) to run. Installation instructions may be found here.

1. Clone this repository to your computer git clone https://github.com/ContextLab/brainfit-paper.git
2. Build an image from the provided Dockerfile (from your local repository directory):
   • docker build -t brainfit-analyses .
3. Run a container instance from the image, publishing port 8888 and mounting the repository as a volume to access the notebooks. The Jupyter Notebook server will start automatically

• docker run --rm -it -p 8888:8888 -v $PWD:/mnt brainfit

4. Copy the third link that appears and paste it into a web browser.

Our main analysis code is organized into the following Jupyter (.ipynb) notebooks:

code
├── demographics.ipynb: generate Figure S1 and summarize participant demographics
├── behavioral_data.ipynb: generate Figures 2, 3, S2, S3, S4, and S5
├── fitness_data.ipynb: generate Figures 4, S6, S7, S8, S9, and S10
├── exploratory_analysis_correlations.ipynb: generate Figures 5, S11, S12, S13, and S14, along with the statistical tests associated with the "exploratory correlation analyses" in the main text
└── reverse_correlation_analysis.ipynb: generate Figures 6, S15, S16, S17, S18, S19, and S20

Figure 1 (a graphical summary of the experimental tasks) was generated manually, using screen captures from the experiment and/or manually recreated depictions of key events in the experiment.