Skip to contents

Helper functions for weight calculation in bag models

Source: R/bag_models.R
bag_model_weights.Rd

These functions compute weights for ensemble model combinations based on validation scores. They handle score thresholding and weight normalization differently:

  • score2weight_* functions convert validation scores to weights

  • w_strech_* functions normalize weights after they are computed

Usage

w_strech_maxmin_squared(x)

w_strech_max_squared(x)

score2weight_mean(x, col = "validation_score", score_threshold = 0.7)

score2weight_min_mean(x, col = "validation_score", score_threshold = 0.7)

score2weight_invmean(x, col = "validation_score", score_threshold = 0.7)

score2weight_min_invmean(x, col = "validation_score", score_threshold = 0.7)

Arguments

x

[list or numeric]
For score2weight_* functions: a model result list with a column specified by col parameter. For w_strech_* functions: a numeric vector of weights.

col

[character(1)="validation_score"]
Column name in x containing validation scores.

score_threshold

[numeric(1)=0.7]
Minimum validation score threshold. Models with any validation score below this are assigned zero weight.

Value

Numeric vector or value representing normalized weights or summary statistics.

Details

Score to weight conversion functions:

  • score2weight_mean: Uses mean of all validation scores

  • score2weight_min_mean: Uses mean only if all scores above threshold

  • score2weight_invmean: Uses inverse of mean (harmonic mean-related)

  • score2weight_min_invmean: Uses inverse mean only if all scores above threshold

Weight stretching/normalization functions:

  • w_strech_maxmin_squared: Rescales to 0,1, squares, then normalizes to sum to 1

  • w_strech_max_squared: Divides by max, squares, then normalizes to sum to 1

See also

bag_models(), data_summary()