Summary of a bag of models

Creates a bag of models object from a fitted ensemble of models. This function synthesizes results from multiple model resamples, assigns weights based on validation scores, and computes weighted coefficients and predictions for ensemble modeling.

Usage

bag_models(
  fitted,
  data,
  score2weight = score2weight_min_invmean,
  metric = fitted$metric,
  weights_col = c("validation_score", "habitat_validation_score")[1],
  score_threshold = 0.7,
  weights_function = NULL
)

Arguments

fitted

[list]
Object containing fitted models, typically output from [oneimpact::fit_net_logit()] or [oneimpact::fit_net_clogit()] with multiple resamples. Must contain elements: $models, $metric, $formula, $method, $standardize, and $samples.

data

[data.frame,tibble]
Complete data set used to fit the models. Used here only to compute summary statistics and covariate scaling information.

score2weight

[function]
Function to convert validation scores to model weights. Should accept arguments: x (model result), weights_col (column name), and score_threshold. Default is score2weight_min_invmean. See [oneimpact::score2weight_min_invmean()] and related functions.

metric

[character]
Name of the evaluation metric to extract from models. Default uses the same metric from the fitted models. Common values: "AUC", "Cindex", "conditionalAUC", "conditionalSomersD".

weights_col

[character(1)="validation_score"]
Column name containing validation scores to be converted to weights. One of "validation_score" or "habitat_validation_score". The latter is used when habitat-specific validation scores are available, such as in step-selection functions in which one wants to disentangle the effects of movement and habitat.

score_threshold

[numeric(1)=0.7]
Minimum validation score threshold. Models with validation score below this threshold are assigned zero weight and excluded from the bag.

weights_function

[function or NULL]
Function to normalize and transform weights into the range 0,1. Default is w_strech_maxmin_squared. See [oneimpact::w_strech_maxmin_squared()] and related functions.

Value

A bag object (also a list) containing:

n

Total number of models in the bag

n_no_errors

Number of successfully fitted models

n_errors

Number of models that failed to fit

n_above_threshold

Number of models with weights above zero

formula

Model formula used for fitting

formula_no_strata

Formula without strata terms

method

Modeling method (e.g., "Lasso", "Ridge", "AdaptiveLasso", etc.)

metric

Evaluation metric used (e.g. "AUC", "Cindex", etc.)

weights

Vector of model weights (normalized)

coef

Matrix of raw coefficients for each model

coef_std

Matrix of standardized coefficients for each model

wcoef

Weighted coefficients

wcoef_std

Weighted standardized coefficients

validation_score

Matrix of validation scores

weighted_validation_score

Weighted validation scores

covariate_mean_sd

Data frame with mean and SD of covariates

data_summary

Summary statistics of variables in the original data

errors

Logical vector indicating which models failed

See also

fit_net_logit(), fit_net_clogit(), predict.bag(), score2weight_min_invmean(), w_strech_maxmin_squared()

Examples

if (FALSE) { # \dontrun{
# Assuming you have fitted models from fit_net_logit or fit_net_clogit
fitted_models <- bag_fit_net_logit(f = case_ ~ x + y, data = mydata, samples = samples_list)
bag <- bag_models(fitted = fitted_models, data = mydata, metric = "AUC")
} # }