RemixAutoML (version 0.4.2)

GenTSAnomVars: GenTSAnomVars is an automated z-score anomaly detection via GLM-like procedure

Description

GenTSAnomVars is an automated z-score anomaly detection via GLM-like procedure. Data is z-scaled and grouped by factors and time periods to determine which points are above and below the control limits in a cumulative time fashion. Then a cumulative rate is created as the final variable. Set KeepAllCols to FALSE to utilize the intermediate features to create rolling stats from them. The anomalies are separated into those that are extreme on the positive end versus those that are on the negative end.

Usage

GenTSAnomVars(
  data,
  ValueCol = "Value",
  GroupVars = NULL,
  DateVar = "DATE",
  HighThreshold = 1.96,
  LowThreshold = -1.96,
  KeepAllCols = TRUE,
  IsDataScaled = FALSE
)

Arguments

data

the source residuals data.table

ValueCol

the numeric column to run anomaly detection over

GroupVars

this is a group by variable

DateVar

this is a time variable for grouping

HighThreshold

this is the threshold on the high end

LowThreshold

this is the threshold on the low end

KeepAllCols

set to TRUE to remove the intermediate features

IsDataScaled

set to TRUE if you already scaled your data

Value

The original data.table with the added columns merged in. When KeepAllCols is set to FALSE, you will get back two columns: AnomHighRate and AnomLowRate - these are the cumulative anomaly rates over time for when you get anomalies from above the thresholds (e.g. 1.96) and below the thresholds.

See Also

Other Unsupervised Learning: AutoKMeans(), H2oIsolationForest(), ResidualOutliers()

Examples

# NOT RUN {
data <- data.table::data.table(
  DateTime = as.Date(Sys.time()),
  Target = stats::filter(
    rnorm(10000, mean = 50, sd = 20),
  filter=rep(1,10),
  circular=TRUE))
data[, temp := seq(1:10000)][, DateTime := DateTime - temp][
  , temp := NULL]
data <- data[order(DateTime)]
x <- data.table::as.data.table(sde::GBM(N=10000)*1000)
data[, predicted := x[-1,]]
data[, Fact1 := sample(letters, size = 10000, replace = TRUE)]
data[, Fact2 := sample(letters, size = 10000, replace = TRUE)]
data[, Fact3 := sample(letters, size = 10000, replace = TRUE)]
stuff <- GenTSAnomVars(
  data,
  ValueCol = "Target",
  GroupVars = c("Fact1","Fact2","Fact3"),
  DateVar = "DateTime",
  HighThreshold = 1.96,
  LowThreshold = -1.96,
  KeepAllCols = TRUE,
  IsDataScaled  = FALSE)
# }