MI-PCR Comparison

Here I describe the content of this repository and how to replicate the simulation study.

How to replicate results

To replicate the study, you first need to make sure you have installed all the packages used. You can use the ./input/prep_machine.R script to install them. You should pay special attention to the version of mice you are using. This study uses the special version of this package that is stored in the input/ folder. The repository mice.sim.pcr stores the code for this version. In the following guide, it is assumed that the machine on which the simulation is run already has all packages installed.

Running the simulation on Lisa

Lisa Cluster is a cluster computer system managed by SURFsara, a cooperative association of Dutch educational and research institutions. Researchers at most dutch universities can request access to this cluster computer.

• Open the initialization script init.R and check that:
  • you have all the required packages installed;
  • the parameter parms$run_type is set to 1 (this will deploy the conditions for the final run of the simulation as opposed to the trial and convergence checks run).
  • the fixed parameters and experimental factor levels are set to the desired values.
• Run on a personal computer the script lisa_step0_time_est.R to check how long it takes to perform a single run across all the conditions with the chosen simulation study set up. This will create an R object called wall_time.
• Open the script lisa_js_normal.sh and replace the wall time in the header with the value of wall_time.
• Decide the number of repetitions in the preparatory script lisa_step1_write_repList.R For example: goal_reps <- 500 ncores <- 15 # Lisa nodes have 16 cores available (16-1 used) narray <- ceiling(goal_reps/ncores) # number of arrays/nodes to use Once you have specified these values, run the script on your computer. This will create a stopos_lines text file that will define the repetition index. This file will be located in the input folder.
• Go to the lisa/ and create a folder with a meaningful name for the run (I usually use the current date). Then, copy here the code and input folders and create an empty output folder.
• Authenticate on Lisa
• Check all the packages called by the init.R script are available and install what is not.
• From your terminal, upload the folder containing the project scp -r path/to/local/project user@lisa.surfsara.nl:project For example: scp -r lisa/20211116 ******@lisa.surfsara.nl:mipca_compare
• Go back to the lisa terminal and load the following modules module load 2020 module load R/4.0.2-intel-2020a module load Stopos/0.93-GCC-9.3.0 (or their most recent version at the time you are running this)
• go to the code folder on the lisa cloned project cd mipca_compare/code/
• run the prepping script by . lisa_prep.sh ../input/stopos_lines
• submit the jobs by sbatch -a 1-34 lisa_js_normal.sh Note that 1-34 defines the dimensionality of the array of jobs. For a short partitioning, only 2 arrays are allowed. For other partitioning, more arrays are allowed. 34 is the result of ceiling(goal_reps/ncores) for the chosen parameters in this example.
• When the array of jobs is done, you can pull the results to your machine by scp -r user@lisa.surfsara.nl:mipca_compare/output/folder path/to/local/project/output/folder
• The script lisa_step4_results.R goes through the Lisa result folder, unzips tar.gz packages and puts results together.
• Finally, the script combine_results.R computes bias, CIC, and all the outcome measures. It also puts together the RDS objects that can be plotted with the functions stored in ./code/plots/

Running the simulation on a PC / Mac

You can also replicate the simulation on a personal computer by following these steps:

• Open the initialization script init.R and check that:
  • you have all the required packages installed
  • the fixed parameters and experimental factor levels are set to the desired values
  • the parameter parms$run_type is set to 1 (this will deploy the conditions for the final run of the simulation as opposed to the trial and convergence checks run)
• Open the script pc_step1_sim.R
• Define the number of desired repetitions by changing the parameter reps
• Define the number of clusters for parallelization by changing the parameter clus
• Run the script pc_step1_sim.R
• Run the script pc_step2_unzip.R to unzip the results and create a single .rds file
• Finally, the script combine_results.R computes bias, CIC, and all the outcome measures. It also puts together the RDS objects that can be plotted with the functions stored in ./code/plots/

Convergence checks

To perform convergence checks:

• Open the initialization script init.R and check that:
  • you have all the required packages installed
  • the fixed parameters and experimental factor levels are set to the desired values
  • the parameter parms$run_type is set to 3 (this will deploy the conditions for convergence check
• Open the script pc_step1_sim.R
• Define the number of desired repetitions by changing the parameter reps
• Define the number of clusters for parallelization by changing the parameter clus
• Run the script pc_step1_sim.R
• Open and run the script checks/check_convergence.R to obtain convergence plots for desired conditions and repetitions

Output files

Notes on what the output files are: - 20220607_175414.tar.gz contains convergence checks - output/lisa/8469421/ contains results reported in 1st submission BRM paper; - output/lisa/9950505/ contains results version of BRM paper results with correct MI-OP method; - output/lisa/???????/ contains HD results requested by of BRM paper revision - 20230303_154402.tar.gz contains results for the non-graphical decision rules checks

Repository content

Here is a brief description of the folders: - code: the main software to run the study - checks folder contains some script to check a few details of the procedure are producing the expected results - experiments folder contains some initial trial scripts - functions folder with the main project specific functions - helper folder with functions to address file management and other small internal tasks - plots folder containing some plotting scripts and pdfs - subroutines folder with the generic functions to run the simulation study - input folder where all the input files (e.g., data) should be stored - output folder where the results of the simulation study will be stored - test folder containing unit testing files