This repository contains all the code needed to replicate the results reported in the paper corresponding to The Beauty Survey. All statistical tests were conducted in R and corresponding files can be found in the "R" folder.
All the figures were generated using python and the required scripts can be found in the folder named "python". The file "mainPaperFigures.ipynb'' contains the code needed to generate all the figures, in the order in which they appear in the paper. Multiple helper CSVs were generated while analyzing the data. These should be created automatically upon launching the notebook. If they already exist, they will not be created again. In case there are missing files however, please run the createHelperFiles.py file to generate all the necessary CSVs.
Note: Many of the R tests also use the helper CSV files. Please ensure these are generated before running the tests in R.
The data is available in the corresponding Zenodo repository. Place all the datafiles in the csv_storage folder to re-run the code for the analysis.
The Zenodo repository also contains helper csv files used for the analysis. If you only want to access the collected ratings, scores given by each rater can be found in the fullRatingsFile.csv file. The centralized scores can be found in the allAttributesMedians_df.csv file.
The images used in the study were picked from two publicly available research datasets - the Chicago Faces Database (CFD) and the FACES dataset. The images picked were renamed so that they include information about the person in the picture and to number them serially. The mapping between the names we used and the names in the databases they were sourced from is available as a dictionary in the ImagesPickedToNumbersMapping.pkl file.
Along with the CSV files, all the statistical models generated have also been made available for ease of use. These are placed in csv_storage/models. In case you wish to re-run these models, all the code is available in the R scripts. Please remember to set the flags to the appropriate values to generate the model files. In case you want to recompute the models, please run the appropriate file as described in the following section.
Unlike the figure generation code in python, the R tests are split into multiple subfiles. Below is a short description of the tests available in each file.
mainPaperTests.Rmd: All the statistical tests run on the centralized scores.runModelComparision.Rmd: This file contains the code to generate all the linear models that were created along with the other variants of models that were tested.mainPaperTests_fullData.Rmd: This file contains all the tests run using the per-rater scores. This file mainly generates the results of the estimated marginal means. These are already pre-computed and made available in thecsv_storagefolder.computeNewSpacing_OSM.Rmd: This file contains the code needed to re-run the OSM's and compute the new scales. This file requires thestereordpackage.
All the figures used in the paper were generated using python. The code to generate the figures can be found in the mainPaperFigures.ipynb notebook inside the Python folder. The folder also contains the requirements file for the conda environment. Recreate it by:
conda create --name <env> --file requirements.txt
While extreme care has been taken to ensure all the necessary data and code is available in this repository, errors do tend to creep in. In case you run into an error, please raise an issue on the repository or feel free to reach out to aditya@ellisalicante.org.