pkg/R/computeForecast.R

   1 #' Compute forecast
   2 #'
   3 #' Predict time-series curves for the selected days indices (lines in data).
   4 #'
   5 #' @param data Object of type \code{Data}, output of \code{getData()}.
   6 #' @param indices Indices where to forecast (the day after); integers relative to the
   7 #'   beginning of data, or (convertible to) Date objects.
   8 #' @param forecaster Name of the main forecaster; more details: ?F_<forecastername>
   9 #' \itemize{
  10 #'   \item Persistence : use last (similar, next) day
  11 #'   \item Neighbors : weighted tomorrows of similar days
  12 #'   \item Average : average tomorrow of all same day-in-week
  13 #'   \item Zero : just output 0 (benchmarking purpose)
  14 #' }
  15 #' @param pjump Function to predict the jump at the interface between two days;
  16 #'   more details: ?J_<functionname>
  17 #' \itemize{
  18 #'   \item Persistence : use last (similar, next) day
  19 #'   \item Neighbors: re-use the weights from F_Neighbors
  20 #'   \item Zero: just output 0 (no adjustment)
  21 #' }
  22 #' @param memory Data depth (in days) to be used for prediction.
  23 #' @param horizon Number of time steps to predict.
  24 #' @param ncores Number of cores for parallel execution (1 to disable).
  25 #' @param ... Additional parameters for the forecasting models.
  26 #'
  27 #' @return An object of class Forecast
  28 #'
  29 #' @examples
  30 #' ts_data <- system.file("extdata","pm10_mesures_H_loc.csv",package="talweg")
  31 #' exo_data <- system.file("extdata","meteo_extra_noNAs.csv",package="talweg")
  32 #' data <- getData(ts_data, exo_data, input_tz="GMT", working_tz="GMT", predict_at=7)
  33 #' pred <- computeForecast(data, 2200:2230, "Persistence", "Zero",
  34 #'   memory=500, horizon=12, ncores=1)
  35 #' \dontrun{#Sketch for real-time mode:
  36 #' data <- Data$new()
  37 #' forecaster <- MyForecaster$new(myJumpPredictFunc)
  38 #' repeat {
  39 #'   # In the morning 7am+ or afternoon 1pm+:
  40 #'   data$append(
  41 #'     times_from_H+1_yersteday_to_Hnow,
  42 #'     PM10_values_of_last_24h,
  43 #'     exogenous_measures_of_last_24h,
  44 #'     exogenous_predictions_for_next_24h)
  45 #'   pred <- forecaster$predictSerie(data, data$getSize()-1, ...)
  46 #'   #do_something_with_pred
  47 #' }}
  48 #' @export
  49 computeForecast = function(data, indices, forecaster, pjump,
  50     memory=Inf, horizon=data$getStdHorizon(), ncores=3, ...)
  51 {
  52     # (basic) Arguments sanity checks
  53     horizon = as.integer(horizon)[1]
  54     if (horizon<=0 || horizon>length(data$getCenteredSerie(1)))
  55         stop("Horizon too short or too long")
  56     integer_indices = sapply(indices, function(i) dateIndexToInteger(i,data))
  57     if (any(integer_indices<=0 | integer_indices>data$getSize()))
  58         stop("Indices out of range")
  59     if (!is.character(forecaster) || !is.character(pjump))
  60         stop("forecaster (name) and pjump (function) should be of class character")
  61
  62     pred = Forecast$new( sapply(indices, function(i) integerIndexToDate(i,data)) )
  63     forecaster_class_name = getFromNamespace(
  64         paste(forecaster,"Forecaster",sep=""), "talweg")
  65     forecaster = forecaster_class_name$new( #.pjump =
  66         getFromNamespace(paste("get",pjump,"JumpPredict",sep=""), "talweg"))
  67
  68     if (ncores > 1 && requireNamespace("parallel",quietly=TRUE))
  69     {
  70         p <- parallel::mclapply(seq_along(integer_indices), function(i) {
  71             list(
  72                 "forecast" = forecaster$predictSerie(
  73                     data, integer_indices[i], memory, horizon, ...),
  74                 "params"= forecaster$getParameters(),
  75                 "index" = integer_indices[i] )
  76             }, mc.cores=ncores)
  77     }
  78     else
  79     {
  80         p <- lapply(seq_along(integer_indices), function(i) {
  81             list(
  82                 "forecast" = forecaster$predictSerie(
  83                     data, integer_indices[i], memory, horizon, ...),
  84                 "params"= forecaster$getParameters(),
  85                 "index" = integer_indices[i] )
  86             })
  87     }
  88
  89     # TODO: find a way to fill pred in //...
  90     for (i in seq_along(integer_indices))
  91     {
  92         pred$append(
  93             forecast = p[[i]]$forecast,
  94             params = p[[i]]$params,
  95             index_in_data = p[[i]]$index
  96         )
  97     }
  98     pred
  99 }