rx_sql_forunco: forunco function for SQL Server compute context

In yvesmauron/univariate-time-series-forecasting: Intelligent Algorithms to be Integrated in Business Intelligence solutions.

Description

Implements the forunco function for optimal parallelisation with machine learning services of sql server 2017

Usage

rx_sql_forunco(
  connection_string,
  table,
  num_cores = 8,
  h = 2,
  levels = c(95),
  methods = c("auto_ets", "auto_arima", "auto_thetaf"),
  pp_methods = c("boxcox"),
  point_combination = "median",
  pi_combination_upper = "median",
  pi_combination_lower = "median",
  pool_limit = length(methods),
  error_fun = "rmse",
  weight_fun = "inverse",
  val_h = h,
  sov_only = F,
  max_years = 30,
  val_min_years = 4,
  cv_min_years = 5,
  cv_max_samples = 3,
  allow_negatives = F,
  ...
)

Arguments

connection_string	mandatory: Connectionstring to the database
table	mandatory: Table name that is the source for the forecasting objects
num_cores	number of cores to be used, default 8.
h	number of horizons to predict.
levels	Prediction interval levels. Optional, default c(95).
methods	Methods to be combined. Optional, default auto_ets, auto_arima, auto_dotm.
point_combination	Point forecast combination operator. Optional, default meidan.
pi_combination_upper	combination operator of the upper bound of the prediction intervals. Optional, default max.
pi_combination_lower	combination operator for the lower bounds of the prediction intervals. Optional, default min.
pool_limit	number of methods selected from the pool to reach the final forecasts. Optional, default length(methods).
error_fun	error function to determine the validation performance Optional, default rmse.
weight_fun	weight function for calculating the definitive weights for combination. Optional, default inverse
val_h	The horizon for the validation samples. Optional, default h.
sov_only	Flag indicating whether only single origin validation should be considered. Optional, default TRUE.
max_years	Maxmium of years to consider during model fitting. Optional, default 30.
val_min_years	Minimum years required to conduct single origin validation. Optional, default 4.
cv_min_years	Minimum years required to conduct cross-origin validation. Optional, default 5.
cv_max_samples	Maximum samples that should be considered during cross-validation, i.e. 3 indicates the algorithm validates from 3 origins. Optional, default 3.
allow_negatives	Flag indicating whether to allow negative values or not. Optional, default FALSE.
...

Value

forecest data.frame

Examples

## Not run: 
connection_string <- "Server=LTYMA01\\QWERTZ;Database=FORECASTING_DATA;UID=ruser;PWD=ruser;"
table <- paste0("m3.vw_Micro")
result <- rx_sql_forunco(connection_string=connection_string, table=table)
#    .id  TSN       date  type       value
#     P1 1403 1994-03-02 Point  1468.48451
#     P1 1403 1994-03-30 Point  1468.48451
#     P1 1403 1994-04-30 Point  1468.48451
#     P1 1403 1994-05-30 Point  1468.48451
#     P1 1403 1994-06-30 Point  1468.48451

## End(Not run)

yvesmauron/univariate-time-series-forecasting documentation built on March 2, 2020, 12:20 a.m.