Calculate Stable Features

Returns a tibble of all features with a maximum selection probability greater than a minimum threshold.

Usage

get_stable_features(x, thresh, add_features, warn)

Arguments

x

An stab_sel class object OR a matrix containing selection probabilities, i.e. the stabpath_matrix entry of a stab_sel object.

thresh

numeric(1) in [0, 1]. the minimum selection probability threshold. In some instances this value can also be a vector, but is generally a scalar > 0.50 for get_stable_features().

add_features

character(n). A string of additional features to force into the resulting table, irrespective of their threshold. Used mostly in the S3 plot method to see a given stability path of a feature not meeting a threshold cutoff. Must be exact string match.

warn

logical(1). Should warnings be triggered if no stable features were found at the specified threshold OR if the FDR upper bound is undefined at thresholds <= 0.5?

Value

A 2-3 column tibble containing maximum selection probabilities, AUC, and FDR upper bounds (when appropriate), see Details.

Details

A "stable feature" is defined as a feature with a maximum selection probability greater than a supplied threshold. Features that satisfy this criterion are returned along with the maximum selection probability and the upper bound on the false discovery rate. This false discovery rate bound is only defined for thresh > 0.5, it is otherwise undefined.

IMPORTANT! If you pass to the matrix method, there is no permutation analysis performed, i.e. no $EmpFDR column in the returned tibble. This calculation is time consuming and is not always desired. This method offers the user a control mechanism for the output behavior.

See also

stability_selection()

Author

Stu Field

Examples

# logistic regression
n_feat      <- 20
n_samples   <- 100
x           <- matrix(rnorm(n_feat * n_samples), n_samples, n_feat)
colnames(x) <- paste0("feat", "_", head(letters, n_feat))
y           <- sample(1:2, n_samples, replace = TRUE)

stab_sel <- stability_selection(x, y, r_seed = 101)
#> ✓ Using kernel: 'binomial' and 1 core (serial)
#> ✓ Stablity path run time: 1.208s

# Stable features at `thresh =`
get_stable_features(stab_sel, 0.75)
#> $thresh_0.75
#> # A tibble: 18 × 3
#>    feature MaxSelectProb FDR_bound
#>    <chr>           <dbl>     <dbl>
#>  1 feat_i          0.965     0.005
#>  2 feat_b          0.96      0.01 
#>  3 feat_n          0.93      0.015
#>  4 feat_l          0.905     0.02 
#>  5 feat_e          0.9       0.025
#>  6 feat_f          0.9       0.03 
#>  7 feat_p          0.865     0.035
#>  8 feat_c          0.845     0.04 
#>  9 feat_s          0.845     0.045
#> 10 feat_a          0.835     0.05 
#> 11 feat_o          0.835     0.055
#> 12 feat_d          0.83      0.06 
#> 13 feat_r          0.83      0.065
#> 14 feat_h          0.82      0.07 
#> 15 feat_g          0.815     0.075
#> 16 feat_q          0.81      0.08 
#> 17 feat_t          0.81      0.085
#> 18 feat_k          0.8       0.09 
#> 

get_stable_features(stab_sel, c(0.75, 0.9))
#> $thresh_0.75
#> # A tibble: 18 × 3
#>    feature MaxSelectProb FDR_bound
#>    <chr>           <dbl>     <dbl>
#>  1 feat_i          0.965     0.005
#>  2 feat_b          0.96      0.01 
#>  3 feat_n          0.93      0.015
#>  4 feat_l          0.905     0.02 
#>  5 feat_e          0.9       0.025
#>  6 feat_f          0.9       0.03 
#>  7 feat_p          0.865     0.035
#>  8 feat_c          0.845     0.04 
#>  9 feat_s          0.845     0.045
#> 10 feat_a          0.835     0.05 
#> 11 feat_o          0.835     0.055
#> 12 feat_d          0.83      0.06 
#> 13 feat_r          0.83      0.065
#> 14 feat_h          0.82      0.07 
#> 15 feat_g          0.815     0.075
#> 16 feat_q          0.81      0.08 
#> 17 feat_t          0.81      0.085
#> 18 feat_k          0.8       0.09 
#> 
#> $thresh_0.9
#> # A tibble: 6 × 3
#>   feature MaxSelectProb FDR_bound
#>   <chr>           <dbl>     <dbl>
#> 1 feat_i          0.965   0.00312
#> 2 feat_b          0.96    0.00625
#> 3 feat_n          0.93    0.00938
#> 4 feat_l          0.905   0.0125 
#> 5 feat_e          0.9     0.0156 
#> 6 feat_f          0.9     0.0188 
#>