Skip to main content

Table 4 Computational time and memory requirements for all feature selection methods, compared in terms of number of SNPs

From: Collective feature selection to identify crucial epistatic variants

Method Computational time based on number of SNPs (in seconds) Memory requirements based on number of SNPs
  100 500 50,000 100,000 100 500 50,000 100,000
LASSO 4.65 58.49 NA NA 10gb 10gb NA NA
LASSO with interactions 186.9 1800 NA NA 20gb 20gb NA NA
Elastic Net 31.44 401.11 NA NA 10gb 10gb NA NA
Ranger 151.31 681.83 NA NA 8gb 8gb NA NA
Gradient Boosting 103.22 466.95 NA NA 8gb 8gb NA NA
MDR 0.25 15.03 6102 89,777 1gb 1gb 10gb 30gb
MultiSURF 2.48 5.13 NA NA 18gb 39gb NA NA
MultiSURF + TuRF 0.05 36.72 65.72 4420 8321 18gb 39gb 28gb 28gb
  1. Note that “NA” here stands for where the model could not be tested due to computational infeasibility while keeping all parameters for simulated datasets same
\