Machines' Specs

Data

• Execution times dataset:
  • Download on Kaggle
  • Or download here (might be slow): exec-times-dataset.csv
• Parameters for the exponentiated Weibull: eweibull-dataset.csv
• Parameters for all distribution families: full-dataset.csv

Code

Inference over the execution times dataset

• Download here: ZIP File
• Run with:

  $ cd /tmp/inference-code/
  $ R
  > source("inference.r");

Simulation study with exponentiated Weibull

The R code is simply:

require(rmutil);
require(nortest);

set.seed(12345);

data = read.csv("eweibull-dataset.csv");
params_df = data[c("param_s", "param_m", "param_f")];

sum_sample = function(n, s, m, f){
    data = rgweibull(n=1000, s=s, m=m, f=f);
    for(i in 2:n){
        data = data + rgweibull(n=1000, s=s, m=m, f=f);
    }
    return(data);
}

counts = NULL;

for(row in 1:nrow(params_df)){
    params = as.numeric(params_df[row,]);

    for(sum_size in seq(10, 10000, by=5)){

        results = NULL;
        for(i in 1:20)
            results = c(results, lillie.test(sum_sample(sum_size, s=params[1], m=params[2], f=params[3]))$p.value);

        count = sum(results > 0.4);
        if(count >= 16){
            print(sum_size);
            counts = c(counts, sum_size);
            break;
        }
    }
}

Where the file "eweibull-dataset.csv" is the one that can be downloaded under section "Data".

To change the p-value threshold, modify the line count = sum(results > 0.4);, in particular, change 0.4 to the desired threshold.

Results are returned into vector counts.