In matloff/qeML: Quick and Easy Machine Learning Tools

---

title: "Function List" output: rmarkdown::html_vignette vignette: > %\VignetteIndexEntry{Function List} %\VignetteEngine{knitr::rmarkdown} \usepackage[utf8]{inputenc}

---

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

• ML algorithms

  • qeAdaBoost(): Ada Boosting, wraps Jousboost pkg

  • qeDeepnet(): wraps deepnet pkg

  • qeDT(): decision trees, wraps party pkg

  • qeGBoost(): gradient boosting, wraps gbm pkg

  • qeKNN(): k-Nearest Neighbors, wraps regtools pkg; includes predictor importance settings; allows linear interpolation within a bin

  • qeLASSO(): LASSO and ridge regression, wraps glmment pkg

  • qeLightGBoost(): gradient boosting, wraps lightgbm pkg

  • qeliquidSVM: wraps liquidSVM pkg

  • qeLin(): wraps R's lm()

  • qeLinKNN(): first fits qeLin(), followed by k-NN on the residuals to correct deviations from linearity

  • qeLogit(): wraps R's glm()

  • qeNCVregCV: wraps ncvreg package, linear gen. linear regression regularized via SCAD etc.

  • qeNeural(): wraps keras package, including CNN

  • qePolyLASSO(): LASSO/ridge applied to polynomial regression; wraps glmnet, polyreg pkgs

  • qePolyLin(): polynomial regression on linear models; uses Moore-Penrose inverse if overfitting; wraps polyreg pkg

  • qePolyLog(): polynomial regression on logistic models; wraps polyreg pkg

  • qeRF(): random forests, wraps randomforest pkg

  • qeRFgrf: random forests, wraps grf pkg; allows linear interpolation within a bin

  • qeRpart(): decision trees, wraps Rpart pkg; colorful tree plot

  • qeRFranger(): random forests, wraps ranger pkg

  • qeskRF(): random forests, wraps Python Scilearn pkg

  • qeskSVM(): SVM, wraps Python Scilearn pkg

  • qeSVM(): SVM, wraps e1071 pkg

  • qeSVMliquid(): SVM, wraps liquid SVM pkg

  • qeXGBoost() wraps the xgboost pkg

• feature importance/selection

  • qeFOCI(), qeFOCIrand(): fully nonparametric method for feature selection

  • qeLASSO(): for feature importance, apply coef() to return value

  • qeLeaveOut1Var: fits full model, then with all features but 1, for each feature, reporting difference in predictive power; use with any qeML predictive function

  • qeRareLevels(): investigates whether rare levels of a feature that is an R factor should be included

  • qeRFranger: variable.importance component of return value

• model development

  • doubleD(): computation and plotting for exploring Double Descent

  • plotClassesUMAP(): plot first two UMAP components, color-coding classes

  • plotPairedResiduals(): plot residuals against pairs of features

  • qeCompare(): compare the accuracy various ML methods on a given dataset

  • qeFT(): automated grid hyperparameter search, with Bonferroni-Dunn corrected standard errors

  • qePCA(): find principal components, number specified by user, then fit the resulting model, according to qe* function specified by user

  • qeROC(): ROC computation and plotting, wraps pROC pkg

  • qeUMAP(): same as qePCA() but using UMAP

  • replicMeans(): (from regtools, included in qeML) averages output, e.g. testAcc, over many holdout sets

• application-specific functions (elementary)

  • qeText() text classification

  • qeTS(): time series

• prediction with missing values

  • qeLinMV(), qeLogitMV(), qeKNNMV(), associated predict() generics for use with toweranNA pkg

• utilities, exploratory tools

  • cartesianFactor(): with inputs of R factors of n1, n2... levels, creates a combined "superfactor" of n1n2... levels

  • dataToTopLevels(): applies factorToTopLevels() to all fadtors in the given data frame

  • factorToTopLevels(): removes rare levels from a factor

  • levelCounts(): performs a census of levels for each R factor in the dataset

  • newDFRow(): creates a new case to input to predict()

  • qeParallel(): apply "Software Alchemy" to parallelize qe functions

matloff/qeML documentation built on Dec. 15, 2024, 10:15 a.m.