In samuelreuther/SRxgboost: Helper and meta-functions for modelling with XGBOOST

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

Installation

# devtools::install_github("samuelreuther/SRxgboost")
library(SRxgboost)

TODOs

SRxgboost_run:
Fehler bei test-02_SRxgboost_run:
- mclass_no_folds_softprob
- mclass_no_folds_softprob_wprec
- mclass_no_folds_softmax
remove parameters for Feature Selection:
- feat_sel
- Feat_Sel
- Feat_Sel_Vars
- Selected_Features = FALSE
add columns in summaryCV for:
- train1f/test1f
- train/CV
- trainfinal/testfinal (calculate test-results if y_test is provided)
- fixed metrics AUC/accuracy/R2/RMSE/MAPE
plot shapley contribution: more expressive graphic for multilabel classification
Shap values für XGBOOST-Modelle: Funktion um für Testdata zu kalkulieren
latin hypercube sampling of hyper-parameter:
https://statcompute.wordpress.com/2019/07/06/latin-hypercube-sampling-in-hyper-parameter-optimization/
Enforcing Monotonic Constraints:
- https://xgboost.readthedocs.io/en/latest/tutorials/monotonic.html
- params_constrained = params.copy()
- params_constrained['monotone_constraints'] = "(1,-1)"
- (1,-1,0): increasing/decreasing constraint, no constraint on the third
- model_with_constraints = xgb.train(params_constrained, d_train,
  num_boost_round = 1000, evals = evallist)
SRxgboost_prepare_data:
Factor Encoder OOF hinzufügen:
- http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-munging/target-encoding.html
- https://towardsdatascience.com/getting-deeper-into-categorical-encodings-for-machine-learning-2312acd347c8
SRxgboost_select_variables:
generate 3 random variables per category
take mean of random variable per category as threshold per variable category
Beispiel für "Starter-Code für Machine Learning-Modell mit XGBOOST inkl. Modell Explainability" erstellen
Surrogate Model berechnen