This project aims to predict distractibility from video frames.
The data files for this project are too large, so the github repo 
(https://github.com/dgraetz/VideoDistraction) mainly hosts
the report and some example stimuli. 

For full access to the data, please visit the google drive link in the Report. 
It has the full repo for 100 % reproducibility.
https://drive.google.com/drive/folders/1jubBPuDnMHI2Z1b7GBoCmmmd2og4UGZF?usp=sharing

Workflow: 
1) Executing the report should work. It reads in preprocessed data saved as RDS and
   an example eye-tracking file. This part is 100 % reproducible on using the 
   Google Drive repo.
2) Not fully reproducible are are the video embeddings because the stimuli I can 
   host in this repo are limited. The code is available in get_videoembeddings,
   containing code for dividing a video into individual frames 
   (get_videoembeddings/makeframes.R) and the script to get the embeddings 
   (get_videoembeddings/python_img_embeddings.R). All embeddings for the
   current data are saved in get_videoembeddings/results.
3) To run the regression models, you may rely on preprocessed data and use the 
   analysis/models.R script. It produces the RDS files containing the train and test
   data (analysis/results/train_data.RDS; analysis/results/test_data.RDS) and the models
   with and without frame embeddings as predictors (analysis/results/caret_img_content.RDS,
   analysis/results/caret_no_img_content.RDS).
   If you would like to look at the preprocessing script, analysis/prepData.R is
   the script that creates the RDS file for preprocessed data analysis/results/all_subj.RDS.
   
   
Author: Dominik Graetz
Contact: dgrtz@uoregon.edu