AutoH2OCARMA: AutoH2OCARMA

In AdrianAntico/ModelingTools: AutoQuant

AutoH2OCARMA

Description

AutoH2OCARMA Mutlivariate Forecasting with calendar variables, Holiday counts, holiday lags, holiday moving averages, differencing, transformations, interaction-based categorical encoding using target variable and features to generate various time-based aggregated lags, moving averages, moving standard deviations, moving skewness, moving kurtosis, moving quantiles, parallelized interaction-based fourier pairs by grouping variables, and Trend Variables.

Usage

AutoH2OCARMA(
  AlgoType = "drf",
  ExcludeAlgos = "XGBoost",
  data,
  TrainOnFull = FALSE,
  TargetColumnName = "Target",
  PDFOutputPath = NULL,
  SaveDataPath = NULL,
  TimeWeights = NULL,
  NonNegativePred = FALSE,
  RoundPreds = FALSE,
  DateColumnName = "DateTime",
  GroupVariables = NULL,
  HierarchGroups = NULL,
  TimeUnit = "week",
  TimeGroups = c("weeks", "months"),
  FC_Periods = 30,
  PartitionType = "timeseries",
  MaxMem = {
     gc()
    
    paste0(as.character(floor(as.numeric(system("awk '/MemFree/ {print $2}' /proc/meminfo",
    intern = TRUE))/1e+06)), "G")
 },
  NThreads = max(1, parallel::detectCores() - 2),
  Timer = TRUE,
  DebugMode = FALSE,
  TargetTransformation = FALSE,
  Methods = c("YeoJohnson", "BoxCox", "Asinh", "Log", "LogPlus1", "Sqrt", "Asin",
    "Logit"),
  XREGS = NULL,
  Lags = c(1:5),
  MA_Periods = c(1:5),
  SD_Periods = NULL,
  Skew_Periods = NULL,
  Kurt_Periods = NULL,
  Quantile_Periods = NULL,
  Quantiles_Selected = NULL,
  AnomalyDetection = NULL,
  Difference = TRUE,
  FourierTerms = 6,
  CalendarVariables = c("second", "minute", "hour", "wday", "mday", "yday", "week",
    "wom", "isoweek", "month", "quarter", "year"),
  HolidayVariable = c("USPublicHolidays", "EasterGroup", "ChristmasGroup",
    "OtherEcclesticalFeasts"),
  HolidayLookback = NULL,
  HolidayLags = 1,
  HolidayMovingAverages = 1:2,
  TimeTrendVariable = FALSE,
  DataTruncate = FALSE,
  ZeroPadSeries = NULL,
  SplitRatios = c(0.7, 0.2, 0.1),
  EvalMetric = "rmse",
  NumOfParDepPlots = 0L,
  GridTune = FALSE,
  ModelCount = 1,
  NTrees = 1000,
  LearnRate = 0.1,
  LearnRateAnnealing = 1,
  GridStrategy = "Cartesian",
  MaxRunTimeSecs = 60 * 60 * 24,
  StoppingRounds = 10,
  MaxDepth = 20,
  SampleRate = 0.632,
  MTries = -1,
  ColSampleRate = 1,
  ColSampleRatePerTree = 1,
  ColSampleRatePerTreeLevel = 1,
  MinRows = 1,
  NBins = 20,
  NBinsCats = 1024,
  NBinsTopLevel = 1024,
  CategoricalEncoding = "AUTO",
  HistogramType = "AUTO",
  Distribution = "gaussian",
  Link = "identity",
  RandomDistribution = NULL,
  RandomLink = NULL,
  Solver = "AUTO",
  Alpha = NULL,
  Lambda = NULL,
  LambdaSearch = FALSE,
  NLambdas = -1,
  Standardize = TRUE,
  RemoveCollinearColumns = FALSE,
  InterceptInclude = TRUE,
  NonNegativeCoefficients = FALSE,
  RandomColNumbers = NULL,
  InteractionColNumbers = NULL
)

Arguments

AlgoType	Select from "dfr" for RandomForecast, "gbm" for gradient boosting, "glm" for generalized linear model, "automl" for H2O's AutoML algo, and "gam" for H2O's Generalized Additive Model.
ExcludeAlgos	For use when AlgoType = "AutoML". Selections include "DRF","GLM","XGBoost","GBM","DeepLearning" and "Stacke-dEnsemble"
data	Supply your full series data set here
TrainOnFull	Set to TRUE to train on full data
TargetColumnName	List the column name of your target variables column. E.g. "Target"
PDFOutputPath	NULL or a path file to output PDFs to a specified folder
SaveDataPath	NULL Or supply a path. Data saved will be called 'ModelID'_data.csv
TimeWeights	1 or a value between zero and 1. Data will be weighted less and less the more historic it gets, by group
NonNegativePred	TRUE or FALSE
RoundPreds	Rounding predictions to an integer value. TRUE or FALSE. Defaults to FALSE
DateColumnName	List the column name of your date column. E.g. "DateTime"
GroupVariables	Defaults to NULL. Use NULL when you have a single series. Add in GroupVariables when you have a series for every level of a group or multiple groups.
HierarchGroups	Vector of hierachy categorical columns.
TimeUnit	List the time unit your data is aggregated by. E.g. "1min", "5min", "10min", "15min", "30min", "hour", "day", "week", "month", "quarter", "year".
TimeGroups	Select time aggregations for adding various time aggregated GDL features.
FC_Periods	Set the number of periods you want to have forecasts for. E.g. 52 for weekly data to forecast a year ahead
PartitionType	Select "random" for random data partitioning "time" for partitioning by time frames
MaxMem	Set to the maximum amount of memory you want to allow for running this function. Default is "32G".
NThreads	Set to the number of threads you want to dedicate to this function.
Timer	Set to FALSE to turn off the updating print statements for progress
DebugMode	Defaults to FALSE. Set to TRUE to get a print statement of each high level comment in function
TargetTransformation	Run Rodeo::AutoTransformationCreate() to find best transformation for the target variable. Tests YeoJohnson, BoxCox, and Asigh (also Asin and Logit for proportion target variables).
Methods	Choose from "YeoJohnson", "BoxCox", "Asinh", "Log", "LogPlus1", "Sqrt", "Asin", or "Logit". If more than one is selected, the one with the best normalization pearson statistic will be used. Identity is automatically selected and compared.
XREGS	Additional data to use for model development and forecasting. Data needs to be a complete series which means both the historical and forward looking values over the specified forecast window needs to be supplied.
Lags	Select the periods for all lag variables you want to create. E.g. c(1:5,52) or list("day" = c(1:10), "weeks" = c(1:4))
MA_Periods	Select the periods for all moving average variables you want to create. E.g. c(1:5,52) or list("day" = c(2:10), "weeks" = c(2:4))
SD_Periods	Select the periods for all moving standard deviation variables you want to create. E.g. c(1:5,52) or list("day" = c(2:10), "weeks" = c(2:4))
Skew_Periods	Select the periods for all moving skewness variables you want to create. E.g. c(1:5,52) or list("day" = c(2:10), "weeks" = c(2:4))
Kurt_Periods	Select the periods for all moving kurtosis variables you want to create. E.g. c(1:5,52) or list("day" = c(2:10), "weeks" = c(2:4))
Quantile_Periods	Select the periods for all moving quantiles variables you want to create. E.g. c(1:5,52) or list("day" = c(2:10), "weeks" = c(2:4))
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")
AnomalyDetection	NULL for not using the service. Other, provide a list, e.g. AnomalyDetection = list("tstat_high" = 4, tstat_low = -4)
Difference	Puts the I in ARIMA for single series and grouped series.
FourierTerms	Set to the max number of pairs. E.g. 2 means to generate two pairs for by each group level and interations if hierarchy is enabled.
CalendarVariables	NULL, or select from "second", "minute", "hour", "wday", "mday", "yday", "week", "isoweek", "month", "quarter", "year"
HolidayVariable	NULL, or select from "USPublicHolidays", "EasterGroup", "ChristmasGroup", "OtherEcclesticalFeasts"
HolidayLookback	Number of days in range to compute number of holidays from a given date in the data. If NULL, the number of days are computed for you.
HolidayLags	Number of lags to build off of the holiday count variable.
HolidayMovingAverages	Number of moving averages to build off of the holiday count variable.
TimeTrendVariable	Set to TRUE to have a time trend variable added to the model. Time trend is numeric variable indicating the numeric value of each record in the time series (by group). Time trend starts at 1 for the earliest point in time and increments by one for each success time point.
DataTruncate	Set to TRUE to remove records with missing values from the lags and moving average features created
ZeroPadSeries	NULL to do nothing. Otherwise, set to "maxmax", "minmax", "maxmin", "minmin". See TimeSeriesFill for explanations of each type
SplitRatios	E.g c(0.7,0.2,0.1) for train, validation, and test sets
EvalMetric	Select from "RMSE", "MAE", "MAPE", "Poisson", "Quantile", "LogLinQuantile", "Lq", "SMAPE", "R2", "MSLE", "MedianAbsoluteError"
NumOfParDepPlots	Set to zeros if you do not want any returned. Can set to a very large value and it will adjust to the max number of features if it's too high
GridTune	Set to TRUE to run a grid tune
ModelCount	Set the number of models to try in the grid tune
NTrees	Select the number of trees you want to have built to train the model
LearnRate	Default 0.10, models available include gbm
LearnRateAnnealing	Default 1, models available include gbm
GridStrategy	Default "Cartesian", models available include
MaxRunTimeSecs	Default 60*60*24, models available include
StoppingRounds	Default 10, models available include
MaxDepth	Default 20, models available include drf, gbm
SampleRate	Default 0.632, models available include drf, gbm
MTries	Default 1, models available include drf
ColSampleRate	Default 1, model available include gbm
ColSampleRatePerTree	Default 1, models available include drf, gbm
ColSampleRatePerTreeLevel	Default 1, models available include drf, gbm
MinRows	Default 1, models available include drf, gbm
NBins	Default 20, models available include drf, gbm
NBinsCats	Default 1024, models available include drf, gbm
NBinsTopLevel	Default 1024, models available include drf, gbm
CategoricalEncoding	Default "AUTO". Choices include : "AUTO", "Enum", "OneHotInternal", "OneHotExplicit", "Binary", "Eigen", "LabelEncoder", "Sort-ByResponse", "EnumLimited"
HistogramType	Default "AUTO". Select from "AUTO", "UniformAdaptive", "Random", "QuantilesGlobal", "RoundRobin"
Distribution	Model family
Link	Link for model family
RandomDistribution	Default NULL
RandomLink	Default NULL
Solver	Model optimizer
Alpha	Default NULL
Lambda	Default NULL
LambdaSearch	Default FALSE,
NLambdas	Default -1
Standardize	Default TRUE
RemoveCollinearColumns	Default FALSE
InterceptInclude	Default TRUE
NonNegativeCoefficients	Default FALSE
RandomColNumbers	NULL
InteractionColNumbers	NULL

Value

See examples

Author(s)

Adrian Antico

See Also

Other Automated Panel Data Forecasting: AutoCatBoostCARMA(), AutoLightGBMCARMA(), AutoXGBoostCARMA()

Examples

## Not run: 

# Load data
data <- data.table::fread("https://www.dropbox.com/s/2str3ek4f4cheqi/walmart_train.csv?dl=1")

# Ensure series have no missing dates (also remove series with more than 25% missing values)
data <- AutoQuant::TimeSeriesFill(
  data,
  DateColumnName = "Date",
  GroupVariables = c("Store","Dept"),
  TimeUnit = "weeks",
  FillType = "maxmax",
  MaxMissingPercent = 0.25,
  SimpleImpute = TRUE)

# Set negative numbers to 0
data <- data[, Weekly_Sales := data.table::fifelse(Weekly_Sales < 0, 0, Weekly_Sales)]

# Remove IsHoliday column
data[, IsHoliday := NULL]

# Create xregs (this is the include the categorical variables instead of utilizing only the interaction of them)
xregs <- data[, .SD, .SDcols = c("Date", "Store", "Dept")]

# Change data types
data[, ":=" (Store = as.character(Store), Dept = as.character(Dept))]
xregs[, ":=" (Store = as.character(Store), Dept = as.character(Dept))]

# Build forecast
Results <- AutoQuant::AutoH2OCARMA(

  # Data Artifacts
  AlgoType = "drf",
  ExcludeAlgos = NULL,
  data = data,
  TargetColumnName = "Weekly_Sales",
  DateColumnName = "Date",
  HierarchGroups = NULL,
  GroupVariables = c("Dept"),
  TimeUnit = "week",
  TimeGroups = c("weeks","months"),

  # Data Wrangling Features
  SplitRatios = c(1 - 10 / 138, 10 / 138),
  PartitionType = "random",

  # Production args
  FC_Periods = 4L,
  TrainOnFull = FALSE,
  MaxMem = {gc();paste0(as.character(floor(max(32, as.numeric(system("awk '/MemFree/ {print $2}' /proc/meminfo", intern=TRUE)) -32) / 1000000)),"G")},
  NThreads = parallel::detectCores(),
  PDFOutputPath = NULL,
  SaveDataPath = NULL,
  Timer = TRUE,
  DebugMode = TRUE,

  # Target Transformations
  TargetTransformation = FALSE,
  Methods = c("BoxCox", "Asinh", "Asin", "Log",
    "LogPlus1", "Sqrt", "Logit", "YeoJohnson"),
  Difference = FALSE,
  NonNegativePred = FALSE,
  RoundPreds = FALSE,

  # Calendar features
  CalendarVariables = c("week", "wom", "month", "quarter", "year"),
  HolidayVariable = c("USPublicHolidays","EasterGroup",
    "ChristmasGroup","OtherEcclesticalFeasts"),
  HolidayLookback = NULL,
  HolidayLags = 1:7,
  HolidayMovingAverages = 2:7,
  TimeTrendVariable = TRUE,

  # Time series features
  Lags = list("weeks" = c(1:4), "months" = c(1:3)),
  MA_Periods = list("weeks" = c(2:8), "months" = c(6:12)),
  SD_Periods = NULL,
  Skew_Periods = NULL,
  Kurt_Periods = NULL,
  Quantile_Periods = NULL,
  Quantiles_Selected = NULL,

  # Bonus Features
  XREGS = NULL,
  FourierTerms = 2L,
  AnomalyDetection = NULL,
  ZeroPadSeries = NULL,
  DataTruncate = FALSE,

  # ML evaluation args
  EvalMetric = "RMSE",
  NumOfParDepPlots = 0L,

  # ML grid tuning args
  GridTune = FALSE,
  GridStrategy = "Cartesian",
  ModelCount = 5,
  MaxRunTimeSecs = 60*60*24,
  StoppingRounds = 10,

  # ML Args
  NTrees = 1000L,
  MaxDepth = 20,
  SampleRate = 0.632,
  MTries = -1,
  ColSampleRatePerTree = 1,
  ColSampleRatePerTreeLevel  = 1,
  MinRows = 1,
  NBins = 20,
  NBinsCats = 1024,
  NBinsTopLevel = 1024,
  HistogramType = "AUTO",
  CategoricalEncoding = "AUTO",
  RandomColNumbers = NULL,
  InteractionColNumbers = NULL,
  WeightsColumn = NULL,

  # ML args
  Distribution = "gaussian",
  Link = "identity",
  RandomDistribution = NULL,
  RandomLink = NULL,
  Solver = "AUTO",
  Alpha = NULL,
  Lambda = NULL,
  LambdaSearch = FALSE,
  NLambdas = -1,
  Standardize = TRUE,
  RemoveCollinearColumns = FALSE,
  InterceptInclude = TRUE,
  NonNegativeCoefficients = FALSE)

UpdateMetrics <-
  Results$ModelInformation$EvaluationMetrics[
    Metric == "MSE", MetricValue := sqrt(MetricValue)]
print(UpdateMetrics)

# Get final number of trees actually used
Results$Model@model$model_summary$number_of_internal_trees

# Inspect performance
Results$ModelInformation$EvaluationMetricsByGroup[order(-R2_Metric)]
Results$ModelInformation$EvaluationMetricsByGroup[order(MAE_Metric)]
Results$ModelInformation$EvaluationMetricsByGroup[order(MSE_Metric)]
Results$ModelInformation$EvaluationMetricsByGroup[order(MAPE_Metric)]

## End(Not run)

AdrianAntico/ModelingTools documentation built on Feb. 1, 2024, 7:33 a.m.