Running an ABC-SMC procedure on a cluster (Slurm)

(Caution, this procedure has not been tested)

To run the ABC-SMC procedure on a cluster, the package has to be installed on the cluster first, and you need two files : (1) a R script containing the code use to run the inference, and (2) a bash script use to launch the master job on the cluster.

1. abc-smc_on_slurm_cluster.R :

library(BRREWABC)

# Model definition
compute_dist <- function(x, ss_obs) {
  ss_sim <- c(x[["alpha"]] + x[["beta"]] + rnorm(1, 0, 0.1),
              x[["alpha"]] * x[["beta"]] + rnorm(1, 0, 0.1))
  dist <- sum((ss_sim - ss_obs)^2)
  return(c(dist))
}

model_list <- list("m1" = compute_dist)

# Define prior distribution
prior_dist <- list("m1" = list(c("alpha", "unif", 0, 4),
                               c("beta", "unif", 0, 1)))

# Create a reference trajectory
sum_stat_obs <- c(2.0, 0.75)

# Run abc smc procedure
res <- abcsmc(model_list = model_list,
              prior_dist = prior_dist,
              ss_obs = sum_stat_obs,
              max_number_of_gen = 30,
              nb_acc_prtcl_per_gen = 5000,
              new_threshold_quantile = 0.75,
              experiment_folderpath = "smploncluster",
              max_concurrent_jobs = 100,
              verbose = TRUE,
              on_cluster = TRUE,
              cluster_type = "slurm")

# Plot results
all_accepted_particles <- res$particles
all_thresholds <- res$thresholds
plot_abcsmc_res(data = all_accepted_particles, prior = prior_dist,
                filename = "smploncluster/res/figs/smploncluster_pairplot_all.png", colorpal = "Blues")
plot_densityridges(data = all_accepted_particles, prior = prior_dist,
                   filename = "smploncluster/res/figs/smploncluster_densityridges.png", colorpal = "Blues")
plot_thresholds(data = all_thresholds, nb_threshold = 1,
                filename = "smploncluster/res/figs/smploncluster_thresholds.png", colorpal = "Blues")

2. launch_abc-smc_on_slurm_cluster.sh :

#!/bin/bash
#SBATCH --job-name abcsmc_prlll

Rscript abc-smc_on_slurm_cluster.R

Then launch the master job on the slurm cluster: sbatch launch_abc-smc_on_slurm_cluster.sh.

Pairplot of all iterations

Threshold evolution over iterations

Density estimates for alpha

Density estimates for beta