Quantify Domain Shift Between Source and Target Platforms

Computes a per-feature domain-shift summary using the Kolmogorov-Smirnov test and Cohen's d effect size.

Usage

compute_domain_shift(source_data, target_data)

Arguments

source_data

Numeric matrix with source samples in rows and features in columns.

target_data

Numeric matrix with target samples in rows and features in columns.

Value

A `data.table` with columns `feature`, `ks_stat`, `ks_p`, and `effect_size_d`.

Examples

set.seed(42)
src <- matrix(rnorm(100 * 5), 100, 5)
tgt <- matrix(rnorm(50 * 5, mean = 2), 50, 5)

compute_domain_shift(src, tgt)
#>      feature ks_stat         ks_p effect_size_d
#>       <char>   <num>        <num>         <num>
#> 1: feature_5    0.78 4.853397e-18     -2.039276
#> 2: feature_1    0.74 2.795028e-16     -1.982399
#> 3: feature_3    0.72 1.958000e-15     -1.964085
#> 4: feature_2    0.70 1.300400e-14     -2.076781
#> 5: feature_4    0.68 8.188018e-14     -2.021904