Learn R Programming
RemixAutoML (version 0.4.2)

AutoLimeAid: AutoLimeAid automated lime

Description

AutoLimeAid automated lime explanations and lime model builds.

Usage

AutoLimeAid(
  EvalPredsData = data,
  LimeTrainingData = data,
  LimeBins = 10,
  LimeIterations = 7500,
  LimeNumFeatures = 0,
  LimeModel = NULL,
  LimeModelPath = NULL,
  LimeModelID = NULL,
  MLModel = NULL,
  MLModelPath = NULL,
  MLMetaDataPath = NULL,
  MLModelID = NULL,
  ModelType = "xgboost",
  TargetType = "classification",
  NThreads = parallel::detectCores(),
  MaxMem = "32G",
  FeatureColumnNames = TestModel$ColNames,
  IDcols = NULL,
  FactorLevelsList = TestModel$FactorLevels,
  TargetLevels = NULL,
  OneHot = FALSE,
  ReturnFeatures = TRUE,
  TransformNumeric = FALSE,
  BackTransNumeric = FALSE,
  TargetColumnName = NULL,
  TransformationObject = NULL,
  TransID = NULL,
  TransPath = NULL,
  MDP_Impute = TRUE,
  MDP_CharToFactor = TRUE,
  MDP_RemoveDates = TRUE,
  MDP_MissFactor = "0",
  MDP_MissNum = -1
)

Arguments

EvalPredsData

Data used for interpretation. Should be the same kind of data used on ML_Scoring functions.

LimeTrainingData

Data used to train your ML model

LimeBins

Number of bins to use for bucketing numeric variables

LimeIterations

Number of lime permutations ran to generate interpretation of predicted value

LimeNumFeatures

How many features do you want to be considering for the Lime evaluation? Set to 0 to use all features

LimeModel

Supply a model if you have one available. Otherwise, provide a model path and either it will be pulling in or made and saved there.

LimeModelPath

Supply a path to where your model is located or to be stored.

LimeModelID

Provide a name for your model. If left NULL, a name will be created for you (and a new model).

MLModel

Supply the model object (except for H2O models). Can leave null.

MLModelPath

Supply a path to where your model is located. If this is supplied, the model will be pulled in from file (even if you supply a model)

MLMetaDataPath

Supply a path to where your model metadata is located (might be the same of the MLModelPath). If this is supplied, artifacts about the model will be pulled in from there.

MLModelID

The name of your model as read in the file directory

ModelType

Choose from "xgboost", "h2o", "catboost"

TargetType

For catboost models only. Select from "classification", "regression", "multiclass"

NThreads

Number of CPU threads.

MaxMem

Set the max memory you want to allocate. E.g. "32G"

FeatureColumnNames

The names of the features used in training your ML model (should be returned with the model or saved to file)

IDcols

The ID columns used in either CatBoost or XGBoost

FactorLevelsList

= TestModel$FactorLevels,

TargetLevels

The target levels used in MultiClass models

OneHot

Replicate what you did with the model training

ReturnFeatures

TRUE or FALSE

TransformNumeric

Replicate what you did with the model training

BackTransNumeric

TRUE or FALSE. Replicate what you did with the model training.

TargetColumnName

For the transformations

TransformationObject

TRUE or FALSE. Replicate what you did with the model training.

TransID

Set to the ID used in model training.

TransPath

Same path used in model training.

MDP_Impute

Replicate what you did with the model training.

MDP_CharToFactor

Replicate what you did with the model training.

MDP_RemoveDates

Replicate what you did with the model training.

MDP_MissFactor

Replicate what you did with the model training.

MDP_MissNum

Replicate what you did with the model training.

Value

LimeModelObject and Lime Explanations

See Also

Other Model Evaluation and Interpretation: EvalPlot(), LimeModel(), ParDepCalPlots(), RedYellowGreen(), threshOptim()

Examples

Run this code
# NOT RUN {
# CatBoost data generator
dataGenH2O <- function() {
  Correl <- 0.85
  N <- 10000
  data <- data.table::data.table(Classification = runif(N))
  data[, x1 := qnorm(Classification)]
  data[, x2 := runif(N)]
  data[, Independent_Variable1 := log(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable2 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable3 := exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable4 := exp(exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2))))]
  data[, Independent_Variable5 := sqrt(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable6 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.10]
  data[, Independent_Variable7 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.25]
  data[, Independent_Variable8 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.75]
  data[, Independent_Variable9 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^2]
  data[, Independent_Variable10 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^4]
  data[, Independent_Variable11 := as.factor(
    ifelse(Independent_Variable2 < 0.20,
    "A",ifelse(Independent_Variable2 < 0.40,
    "B",ifelse(Independent_Variable2 < 0.6,
    "C",ifelse(Independent_Variable2 < 0.8,  "D", "E")))))]
  data[, ':=' (x1 = NULL, x2 = NULL)]
  data[, Classification := ifelse(Classification > 0.5, 1, 0)]
  rm(N,Correl)
  return(data)
}
data <- dataGenH2O()
TestModel <- RemixAutoML::AutoCatBoostRegression(
  data,
  TrainOnFull = FALSE,
  ValidationData = NULL,
  TestData = NULL,
  TargetColumnName = "Classification",
  FeatureColNames = c(2:12),
  PrimaryDateColumn = NULL,
  IDcols = NULL,
  MaxModelsInGrid = 3,
  task_type = "GPU",
  eval_metric = "RMSE",
  Trees = 50,
  GridTune = FALSE,
  model_path = "C:/Users/aantico/Documents/Package/GUI_Package",
  metadata_path = NULL,
  ModelID = "Adrian",
  NumOfParDepPlots = 15,
  ReturnModelObjects = TRUE,
  SaveModelObjects = TRUE,
  PassInGrid = NULL)

# CatBoost Build Lime Model and Explanations
LimeOutput <- RemixAutoML::AutoLimeAid(
  EvalPredsData = data[c(1,15)],
  LimeTrainingData = data,
  LimeBins = 10,
  LimeIterations = 7500,
  LimeNumFeatures = 0,
  TargetType = "regression",
  LimeModel = NULL,
  LimeModelPath = "C:/Users/aantico/Documents/Package/GUI_Package",
  LimeModelID = "AdrianLime",
  MLModel = NULL,
  MLModelPath = "C:/Users/aantico/Documents/Package/GUI_Package",
  MLMetaDataPath = NULL,
  MLModelID = "Adrian",
  ModelType = "catboost",
  NThreads = parallel::detectCores(),
  MaxMem = "14G",
  FeatureColumnNames = NULL,
  IDcols = NULL,
  FactorLevelsList = NULL,
  TargetLevels = NULL,
  OneHot = FALSE,
  ReturnFeatures = TRUE,
  TransformNumeric = FALSE,
  BackTransNumeric = FALSE,
  TargetColumnName = NULL,
  TransformationObject = NULL,
  TransID = NULL,
  TransPath = NULL,
  MDP_Impute = TRUE,
  MDP_CharToFactor = TRUE,
  MDP_RemoveDates = TRUE,
  MDP_MissFactor = "0",
  MDP_MissNum = -1)

# Plot lime objects
lime::plot_features(LimeOutput$LimeExplanations)
suppressWarnings(lime::plot_explanations(LimeOutput$LimeExplanations))

# H2O data generator
dataGenH2O <- function() {
  Correl <- 0.85
  N <- 10000
  data <- data.table::data.table(Classification = runif(N))
  data[, x1 := qnorm(Classification)]
  data[, x2 := runif(N)]
  data[, Independent_Variable1 := log(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable2 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable3 := exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable4 := exp(exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2))))]
  data[, Independent_Variable5 := sqrt(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable6 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.10]
  data[, Independent_Variable7 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.25]
  data[, Independent_Variable8 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.75]
  data[, Independent_Variable9 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^2]
  data[, Independent_Variable10 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^4]
  data[, Independent_Variable11 := as.factor(ifelse(Independent_Variable2 < 0.20,
    "A",ifelse(Independent_Variable2 < 0.40,
    "B",ifelse(Independent_Variable2 < 0.6,
    "C",ifelse(Independent_Variable2 < 0.8,  "D", "E")))))]
  data[, ':=' (x1 = NULL, x2 = NULL)]
  data[, Classification := ifelse(Classification > 0.5, 1, 0)]
  rm(N,Correl)
  return(data)
}
data <- dataGenH2O()
TestModel <- RemixAutoML::AutoH2oDRFClassifier(
  data = data,
  TrainOnFull = FALSE,
  ValidationData = NULL,
  TestData = NULL,
  TargetColumnName = "Classification",
  FeatureColNames = setdiff(names(data),"Classification"),
  eval_metric = "auc",
  Trees = 50,
  GridTune = FALSE,
  MaxMem = "32G",
  NThreads = max(1, parallel::detectCores()-2),
  MaxModelsInGrid = 10,
  model_path = "C:/Users/aantico/Desktop/Retention Analytics",
  metadata_path = NULL,
  ModelID = "Adrian",
  NumOfParDepPlots = 10,
  ReturnModelObjects = TRUE,
  SaveModelObjects = TRUE,
  IfSaveModel = "standard",
  H2OShutdown = TRUE)

LimeOutput <- RemixAutoML::AutoLimeAid(
  EvalPredsData = data[c(1,15)],
  LimeTrainingData = data,
  LimeBins = 10,
  LimeIterations = 7500,
  TargetType = "regression",
  LimeNumFeatures = 0,
  LimeModel = NULL,
  LimeModelPath = "C:/Users/aantico/Desktop/Retention Analytics",
  LimeModelID = "AdrianLime",
  MLModel = NULL,
  MLModelPath = "C:/Users/aantico/Desktop/Retention Analytics",
  MLMetaDataPath = NULL,
  MLModelID = "Adrian",
  ModelType = "h2o",
  NThreads = parallel::detectCores(),
  MaxMem = "14G",
  FeatureColumnNames = NULL,
  IDcols = NULL,
  FactorLevelsList = NULL,
  TargetLevels = NULL,
  OneHot = FALSE,
  ReturnFeatures = TRUE,
  TransformNumeric = FALSE,
  BackTransNumeric = FALSE,
  TargetColumnName = NULL,
  TransformationObject = NULL,
  TransID = NULL,
  TransPath = NULL,
  MDP_Impute = TRUE,
  MDP_CharToFactor = TRUE,
  MDP_RemoveDates = TRUE,
  MDP_MissFactor = "0",
  MDP_MissNum = -1)

# Plot lime objects
lime::plot_features(LimeOutput$LimeExplanations)
suppressWarnings(lime::plot_explanations(LimeOutput$LimeExplanations))

# XGBoost create data function
dataGenXGBoost <- function() {
  Correl <- 0.85
  N <- 10000
  data <- data.table::data.table(Classification = runif(N))
  data[, x1 := qnorm(Classification)]
  data[, x2 := runif(N)]
  data[, Independent_Variable1 := log(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable2 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable3 := exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable4 := exp(exp(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2))))]
  data[, Independent_Variable5 := sqrt(pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))]
  data[, Independent_Variable6 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.10]
  data[, Independent_Variable7 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.25]
  data[, Independent_Variable8 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^0.75]
  data[, Independent_Variable9 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^2]
  data[, Independent_Variable10 := (pnorm(Correl * x1 + sqrt(1-Correl^2) * qnorm(x2)))^4]
  data[, Independent_Variable11 := as.factor(ifelse(Independent_Variable2 < 0.20,
    "A",ifelse(Independent_Variable2 < 0.40,
    "B",ifelse(Independent_Variable2 < 0.6,
    "C",ifelse(Independent_Variable2 < 0.8,  "D", "E")))))]
  data[, ':=' (x1 = NULL, x2 = NULL)]
  data[, Classification := ifelse(Classification > 0.5, 1, 0)]
  rm(Correl,N)
  return(data)
}
data <- dataGenXGBoost()
TestModel <- RemixAutoML::AutoXGBoostClassifier(
  data,
  TrainOnFull = FALSE,
  ValidationData = NULL,
  TestData = NULL,
  TargetColumnName = "Classification",
  FeatureColNames = 2:12,
  IDcols = NULL,
  eval_metric = "auc",
  Trees = 50,
  GridTune = FALSE,
  grid_eval_metric = "auc",
  MaxModelsInGrid = 10,
  NThreads = 8,
  TreeMethod = "hist",
  model_path = "C:/Users/aantico/Desktop/Retention Analytics",
  metadata_path = NULL,
  ModelID = "Adrian2",
  NumOfParDepPlots = 3,
  ReturnModelObjects = TRUE,
  ReturnFactorLevels = TRUE,
  SaveModelObjects = TRUE,
  PassInGrid = NULL)

# XGBoost Build Lime and Generate Output
LimeOutput <- RemixAutoML::AutoLimeAid(
  EvalPredsData = data[c(1,15)],
  LimeTrainingData = data,
  LimeBins = 10,
  TargetType = "classification",
  LimeIterations = 7500,
  LimeNumFeatures = 0,
  LimeModel = NULL,
  LimeModelPath = "C:/Users/aantico/Desktop/Retention Analytics",
  LimeModelID = "Adrian2Lime",
  MLModel = NULL,
  MLModelPath = "C:/Users/aantico/Desktop/Retention Analytics",
  MLMetaDataPath = NULL,
  MLModelID = "Adrian2",
  ModelType = "xgboost",
  NThreads = parallel::detectCores(),
  MaxMem = "14G",
  FeatureColumnNames = NULL,
  IDcols = NULL,
  FactorLevelsList = NULL,
  TargetLevels = NULL,
  OneHot = FALSE,
  ReturnFeatures = TRUE,
  TransformNumeric = FALSE,
  BackTransNumeric = FALSE,
  TargetColumnName = NULL,
  TransformationObject = NULL,
  TransID = NULL,
  TransPath = NULL,
  MDP_Impute = TRUE,
  MDP_CharToFactor = TRUE,
  MDP_RemoveDates = TRUE,
  MDP_MissFactor = "0",
  MDP_MissNum = -1)

# Plot lime objects
lime::plot_features(LimeOutput$LimeExplanations)
suppressWarnings(lime::plot_explanations(LimeOutput$LimeExplanations))
# }

Run the code above in your browser using DataLab