RemixAutoML (version 0.5.0)

AutoLagRollStats: AutoLagRollStats

Description

AutoLagRollStats Builds lags and a large variety of rolling statistics with options to generate them for hierarchical categorical interactions.

Usage

AutoLagRollStats(
  data,
  Targets = NULL,
  HierarchyGroups = NULL,
  IndependentGroups = NULL,
  DateColumn = NULL,
  TimeUnit = "day",
  TimeUnitAgg = "day",
  TimeGroups = "day",
  TimeBetween = NULL,
  RollOnLag1 = TRUE,
  Type = "Lag",
  SimpleImpute = TRUE,
  Lags = c(1:5),
  MA_RollWindows = c(2, 5, 10),
  SD_RollWindows = c(5, 10),
  Skew_RollWindows = c(5, 10),
  Kurt_RollWindows = c(5, 10),
  Quantile_RollWindows = c(10),
  Quantiles_Selected = c("q25", "q75"),
  Debug = FALSE
)

Arguments

data

A data.table you want to run the function on

Targets

A character vector of the column names for the reference column in which you will build your lags and rolling stats

HierarchyGroups

A vector of categorical column names that you want to have generate all lags and rolling stats done for the individual columns and their full set of interactions.

IndependentGroups

A vector of categorical column names that you want to have run independently of each other. This will mean that no interaction will be done.

DateColumn

The column name of your date column used to sort events over time

TimeUnit

List the time aggregation level for the time between events features, such as "hour", "day", "weeks", "months", "quarter", or "year"

TimeUnitAgg

List the time aggregation of your data that you want to use as a base time unit for your features. E.g. "raw" or "day"

TimeGroups

A vector of TimeUnits indicators to specify any time-aggregated GDL features you want to have returned. E.g. c("raw" (no aggregation is done),"hour", "day","week","month","quarter","year")

TimeBetween

Specify a desired name for features created for time between events. Set to NULL if you don't want time between events features created.

RollOnLag1

Set to FALSE to build rolling stats off of target columns directly or set to TRUE to build the rolling stats off of the lag-1 target

Type

List either "Lag" if you want features built on historical values or "Lead" if you want features built on future values

SimpleImpute

Set to TRUE for factor level imputation of "0" and numeric imputation of -1

Lags

A numeric vector of the specific lags you want to have generated. You must include 1 if WindowingLag = 1.

MA_RollWindows

A numeric vector of the specific rolling statistics window sizes you want to utilize in the calculations.

SD_RollWindows

A numeric vector of Standard Deviation rolling statistics window sizes you want to utilize in the calculations.

Skew_RollWindows

A numeric vector of Skewness rolling statistics window sizes you want to utilize in the calculations.

Kurt_RollWindows

A numeric vector of Kurtosis rolling statistics window sizes you want to utilize in the calculations.

Quantile_RollWindows

A numeric vector of Quantile rolling statistics window sizes you want to utilize in the calculations.

Quantiles_Selected

Select from the following c("q5", "q10", "q15", "q20", "q25", "q30", "q35", "q40", "q45", "q50", "q55", "q60"," q65", "q70", "q75", "q80", "q85", "q90", "q95")

Debug

Set to TRUE to get a print of which steps are running

Value

data.table of original data plus created lags, rolling stats, and time between event lags and rolling stats

See Also

Other Feature Engineering: AutoDataPartition(), AutoDiffLagN(), AutoHierarchicalFourier(), AutoInteraction(), AutoLagRollStatsScoring(), AutoTransformationCreate(), AutoTransformationScore(), AutoWord2VecModeler(), AutoWord2VecScoring(), ContinuousTimeDataGenerator(), CreateCalendarVariables(), CreateHolidayVariables(), DT_GDL_Feature_Engineering(), DifferenceDataReverse(), DifferenceData(), DummifyDT(), H2OAutoencoderScoring(), H2OAutoencoder(), ModelDataPrep(), Partial_DT_GDL_Feature_Engineering(), TimeSeriesFill()

Examples

# NOT RUN {
# Create fake Panel Data----
Count <- 1L
for(Level in LETTERS) {
  datatemp <- RemixAutoML::FakeDataGenerator(
    Correlation = 0.75,
    N = 25000L,
    ID = 0L,
    ZIP = 0L,
    FactorCount = 0L,
    AddDate = TRUE,
    Classification = FALSE,
    MultiClass = FALSE)
  datatemp[, Factor1 := eval(Level)]
  if(Count == 1L) {
    data <- data.table::copy(datatemp)
  } else {
    data <- data.table::rbindlist(
      list(data, data.table::copy(datatemp)))
  }
  Count <- Count + 1L
}

# Add scoring records
data <- RemixAutoML::AutoLagRollStats(

  # Data
  data                 = data,
  DateColumn           = "DateTime",
  Targets              = "Adrian",
  HierarchyGroups      = NULL,
  IndependentGroups    = c("Factor1"),
  TimeUnitAgg          = "days",
  TimeGroups           = c("days", "weeks",
                           "months", "quarters"),
  TimeBetween          = NULL,
  TimeUnit             = "days",

  # Services
  RollOnLag1           = TRUE,
  Type                 = "Lag",
  SimpleImpute         = TRUE,

  # Calculated Columns
  Lags                 = list("days" = c(seq(1,5,1)),
                              "weeks" = c(seq(1,3,1)),
                              "months" = c(seq(1,2,1)),
                              "quarters" = c(seq(1,2,1))),
  MA_RollWindows       = list("days" = c(seq(1,5,1)),
                              "weeks" = c(seq(1,3,1)),
                              "months" = c(seq(1,2,1)),
                              "quarters" = c(seq(1,2,1))),
  SD_RollWindows       = NULL,
  Skew_RollWindows     = NULL,
  Kurt_RollWindows     = NULL,
  Quantile_RollWindows = NULL,
  Quantiles_Selected   = NULL,
  Debug                = FALSE)
# }