Lrnr_cv_selector: Cross-Validated Selector

In tlverse/sl3: Pipelines for Machine Learning and Super Learning

Cross-Validated Selector

Description

This learner is the cross-validated (CV) selector, and it is intended for use as the metalearner in Lrnr_sl. Lrnr_cv_selector selects the candidate with the best CV predictive performance (i.e., lowest CV risk). Specifically, it aims to optimize the CV risk, and it is defined by a constrained weighted combination: the weights can either be zero or one, and they must sum to one. Lrnr_cv_selector optimizes the CV predictive performance under these constraints by assigning the candidate with the best CV predictive performance a weight of one and all others a weight of zero. Thus, Lrnr_cv_selector and its predictions will be identical to the best-performing candidate learner and its predictions; this is why we say Lrnr_cv_selector "selects" the candidate with the best CV predictive performance.

Format

An R6Class object inheriting from Lrnr_base.

Value

A learner object inheriting from Lrnr_base with methods for training and prediction. For a full list of learner functionality, see the complete documentation of Lrnr_base.

Parameters

• eval_function = loss_squared_error: A function that takes as input a vector of predicted values as its first argument and a vector of observed outcome values as its second argument, and then returns a vector of losses or a numeric risk. See loss_functions and risk_functions for options.

• folds = NULL: Optional origami-structured cross-validation folds from the task for training Lrnr_sl, e.g., task$folds. This argument is only required and utilized when eval_function is not a loss function, since the risk has to be calculated on each validation set separately and then averaged across them in order to estimate the cross-validated risk. This argument is ignored when eval_function is a loss.

See Also

Other Learners: Custom_chain, Lrnr_HarmonicReg, Lrnr_arima, Lrnr_bartMachine, Lrnr_base, Lrnr_bayesglm, Lrnr_caret, Lrnr_cv, Lrnr_dbarts, Lrnr_define_interactions, Lrnr_density_discretize, Lrnr_density_hse, Lrnr_density_semiparametric, Lrnr_earth, Lrnr_expSmooth, Lrnr_gam, Lrnr_ga, Lrnr_gbm, Lrnr_glm_fast, Lrnr_glm_semiparametric, Lrnr_glmnet, Lrnr_glmtree, Lrnr_glm, Lrnr_grfcate, Lrnr_grf, Lrnr_gru_keras, Lrnr_gts, Lrnr_h2o_grid, Lrnr_hal9001, Lrnr_haldensify, Lrnr_hts, Lrnr_independent_binomial, Lrnr_lightgbm, Lrnr_lstm_keras, Lrnr_mean, Lrnr_multiple_ts, Lrnr_multivariate, Lrnr_nnet, Lrnr_nnls, Lrnr_optim, Lrnr_pca, Lrnr_pkg_SuperLearner, Lrnr_polspline, Lrnr_pooled_hazards, Lrnr_randomForest, Lrnr_ranger, Lrnr_revere_task, Lrnr_rpart, Lrnr_rugarch, Lrnr_screener_augment, Lrnr_screener_coefs, Lrnr_screener_correlation, Lrnr_screener_importance, Lrnr_sl, Lrnr_solnp_density, Lrnr_solnp, Lrnr_stratified, Lrnr_subset_covariates, Lrnr_svm, Lrnr_tsDyn, Lrnr_ts_weights, Lrnr_xgboost, Pipeline, Stack, define_h2o_X(), undocumented_learner

Examples

## Not run: 
data(cpp_imputed)
covs <- c("apgar1", "apgar5", "parity", "gagebrth", "mage", "meducyrs")
task <- sl3_Task$new(cpp_imputed, covariates = covs, outcome = "haz")

hal_lrnr <- Lrnr_hal9001$new(
  max_degree = 1, num_knots = c(20, 10), smoothness_orders = 0
)
lasso_lrnr <- Lrnr_glmnet$new()
glm_lrnr <- Lrnr_glm$new()
ranger_lrnr <- Lrnr_ranger$new()
lrnrs <- c(hal_lrnr, lasso_lrnr, glm_lrnr, ranger_lrnr)
names(lrnrs) <- c("hal", "lasso", "glm", "ranger")
lrnr_stack <- make_learner(Stack, lrnrs)
metalrnr_discrete_MSE <- Lrnr_cv_selector$new(loss_squared_error)
discrete_sl <- Lrnr_sl$new(
  learners = lrnr_stack, metalearner = metalrnr_discrete_MSE
)
discrete_sl_fit <- discrete_sl$train(task)
discrete_sl_fit$cv_risk(loss_squared_error)

## End(Not run)

tlverse/sl3 documentation built on Nov. 18, 2024, 12:46 a.m.