# Requirements
- All the analysis code and reproducible workflow are written in R programming language (version 3.6.1).
- To install the R packages used in the code, first make sure that the prerequisites for installing rstan and brms are met; see: https://github.com/stan-dev/rstan/wiki/RStan-Getting-Started. Then, run the R file `install_packages.R`
# How to produce the figures and tables in the paper?
- First, download the whole directory in your system
- Make sure you have the R package "rmarkdown" installed in your system and you are able to generate the PDF from an .Rmd file
- Knit the reproducible workflow document: `Workflow_interference_modeling.Rmd` (Runtime: 2 minutes)
- The workflow generates all the figures and tables used in the paper (You may try this: empty the ./Plots folder and then run the workflow; it will re-fill the ./Plots folder with the figures generated by the Rmd file)
- The PDF generated after knitting the workflow, `Workflow_interference_modeling.pdf`, presents a consolidated summary of the step-by-step analysis (modeling) done in the paper
# How to reproduce the estimates of number agreement effects (used for plotting Figure 1 in the paper)?
- Run the R file: `./Reproduce-observed-effects/Estimate_number_agreement_effects.R` (Runtime: 2 hours, Requirements: 68 cores, Linux)
- If you do not have sufficient computational resource, run the R file `./Reproduce_observed_effects/Estimate_number_agreement_effects_inseries.R` (Runtime: 34 hours, Requirements: 4 cores, Linux)
# How to reproduce the prediction samples that were used for plotting prior predictions for each model?
Run the following R files (all in `./Reproduce-prediction-samples` folder):
- `./Reproduce-prediction-samples/Retrieval-only.R` (Runtime: 1 hour, Requirement: 1 core)
- `./Reproduce-prediction-samples/Direct_access_non_linear_cues_model.R` (Runtime: 1 hour, Requirement: 1 core)
- `./Reproduce-prediction-samples/Feature-percolation.R` (Runtime: 20 minutes, Requirements: 4 cores, Linux)
- `./Reproduce-prediction-samples/Reading_bias_model.R` (Runtime: 1 hour, Requirement: 4 cores, Linux)
- `./Reproduce-prediction-samples/Lossy-compression.R` (Runtime: 2 hours, Requirements: 20 cores, Linux)
- `./Reproduce-prediction-samples/FPR.R` (Runtime: 1 hour, Requirements: 4 cores, Linux)
- `./Reproduce-prediction-samples/LCR.R` (Runtime: 2 hours, Requirements: 20 cores, Linux)
# How to reproduce the posterior samples that were used for computing the predictive performance of each model?
Run the following R files (all in `./Reproduce-posterior-samples` folder):
- `./Reproduce-posterior-samples/Cue-based-retrieval.R` (Runtime: 2 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/Non-linear-cue-based-retrieval.R` (Runtime: 2 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/Feature-percolation.R` (Runtime: 3 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/Reading-bias-model.R` (Runtime: 1 hour, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/Lossy-compression.R` (Runtime: 140 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/FPR.R` (Runtime: 14 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/LCR.R` (Runtime: 140 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/Feature-percolation-with-markedness.R` (Runtime: 3 hours, Requirements: 34 cores, Linux)
- `./Reproduce-posterior-samples/FPR-with-markedness.R` (Runtime: 14 hours, Requirements: 34 cores, Linux)
Note: The above runtimes are given for estimating model parameters under all seven priors on distortion rate.
