predNew_local: Prediction at new locations based on the fitting results in...

In XiaosuTong/drsstl: Spatial Seasonal Trend Loess analysis routine using divide and recombined

Description

The prediction at new locations are calculated based on the fitting results saved in memory based on the original dataset.

Usage

predNew_local(original, newdata, mlcontrol = spacetime.control())

Arguments

original	The data.frame which contains all fitting results of original dataset. The data.frame is saved in memory, not on HDFS.
newdata	A data.frame includes all locations' longitude, latitude, and elevation, where the prediction is to be calculated.
mlcontrol	Should be a list object generated from spacetime.control function. The list including all necessary smoothing parameters of nonparametric fitting.

Author(s)

Xiaosu Tong

See Also

spacetime.control, mapreduce.control

Examples

## Not run: 
    library(maps)
    library(Spaloess)
    library(datadr)
    new.grid <- expand.grid(
      lon = seq(-126, -67, by = 1),
      lat = seq(25, 49, by = 1)
    )
    instate <- !is.na(map.where("state", new.grid$lon, new.grid$lat))
    new.grid <- new.grid[instate, ]

    elev.fit <- spaloess( elev ~ lon + lat,
      data = station_info,
      degree = 2,
      span = 0.05,
      distance = "Latlong",
      normalize = FALSE,
      napred = FALSE,
      alltree = FALSE,
      family="symmetric",
      control=loess.control(surface = "direct")
    )
    grid.fit <- predloess(
      object = elev.fit,
      newdata = data.frame(
        lon = new.grid$lon,
        lat = new.grid$lat
      )
    )
    new.grid$elev <- grid.fit

    n <- 5000 # just use 5000 stations as example
    set.seed(99)
    first_stations <- sample(unique(tmax_all$station.id), n)
    small_dt <- subset(tmax_all, station.id %in% first_stations)
    small_dt$station.id <- as.character(small_dt$station.id)
    small_dt$month <- as.character(small_dt$month)
    mlcontrol <- spacetime.control(
      vari="tmax", time="date", n=576, n.p=12, stat_n=n, surf = "interpolate",
      s.window="periodic", t.window = 241, degree=2, span=0.75, Edeg=0
    )

    fitted <- drsstl(
      data=small_dt,
      output=NULL,
      model_control=mlcontrol
    )
    rst <- predNew_local(
      original = recombine(fitted, combRbind), newdata = new.grid, mlcontrol = mlcontrol
    )

## End(Not run)

XiaosuTong/drsstl documentation built on May 9, 2019, 11:06 p.m.