pkg/R/F_Neighbors.R

   1 #' @include Forecaster.R
   2 #'
   3 #' Neighbors Forecaster
   4 #'
   5 #' Predict tomorrow as a weighted combination of "futures of the past" days.
   6 #' Inherits \code{\link{Forecaster}}
   7 #'
   8 NeighborsForecaster = R6::R6Class("NeighborsForecaster",
   9     inherit = Forecaster,
  10
  11     public = list(
  12         predictShape = function(data, today, memory, horizon, ...)
  13         {
  14             # (re)initialize computed parameters
  15             private$.params <- list("weights"=NA, "indices"=NA, "window"=NA)
  16
  17             # Do not forecast on days with NAs (TODO: softer condition...)
  18             if (any(is.na(data$getCenteredSerie(today))))
  19                 return (NA)
  20
  21             # Determine indices of no-NAs days followed by no-NAs tomorrows
  22             fdays = getNoNA2(data, max(today-memory,1), today-1)
  23
  24             # Get optional args
  25             local = ifelse(hasArg("local"), list(...)$local, TRUE) #same level + season?
  26             simtype = ifelse(hasArg("simtype"), list(...)$simtype, "none") #or "endo", or "exo"
  27             if (hasArg("window"))
  28             {
  29                 return ( private$.predictShapeAux(data,
  30                     fdays, today, horizon, local, list(...)$window, simtype, TRUE) )
  31             }
  32
  33             # Indices of similar days for cross-validation; TODO: 20 = magic number
  34             cv_days = getSimilarDaysIndices(today, data, limit=20, same_season=FALSE,
  35                 days_in=fdays)
  36
  37             # Optimize h : h |--> sum of prediction errors on last N "similar" days
  38             errorOnLastNdays = function(window, simtype)
  39             {
  40                 error = 0
  41                 nb_jours = 0
  42                 for (i in seq_along(cv_days))
  43                 {
  44                     # mix_strategy is never used here (simtype != "mix"), therefore left blank
  45                     prediction = private$.predictShapeAux(data, fdays, cv_days[i], horizon, local,
  46                         window, simtype, FALSE)
  47                     if (!is.na(prediction[1]))
  48                     {
  49                         nb_jours = nb_jours + 1
  50                         error = error +
  51                             mean((data$getSerie(cv_days[i]+1)[1:horizon] - prediction)^2)
  52                     }
  53                 }
  54                 return (error / nb_jours)
  55             }
  56
  57             # TODO: 7 == magic number
  58             if (simtype=="endo" || simtype=="mix")
  59             {
  60                 best_window_endo = optimize(
  61                     errorOnLastNdays, c(0,7), simtype="endo")$minimum
  62             }
  63             if (simtype=="exo" || simtype=="mix")
  64             {
  65                 best_window_exo = optimize(
  66                     errorOnLastNdays, c(0,7), simtype="exo")$minimum
  67             }
  68
  69             best_window =
  70                 if (simtype == "endo")
  71                     best_window_endo
  72                 else if (simtype == "exo")
  73                     best_window_exo
  74                 else if (simtype == "mix")
  75                     c(best_window_endo,best_window_exo)
  76                 else #none: value doesn't matter
  77                     1
  78
  79             return(private$.predictShapeAux(data, fdays, today, horizon, local,
  80                 best_window, simtype, TRUE))
  81         }
  82     ),
  83     private = list(
  84         # Precondition: "today" is full (no NAs)
  85         .predictShapeAux = function(data, fdays, today, horizon, local, window, simtype,
  86             final_call)
  87         {
  88             fdays_cut = fdays[ fdays < today ]
  89             if (length(fdays_cut) <= 1)
  90                 return (NA)
  91
  92             if (local)
  93             {
  94                 # Neighbors: days in "same season"; TODO: 60 == magic number...
  95                 fdays = getSimilarDaysIndices(today, data, limit=60, same_season=TRUE,
  96                     days_in=fdays_cut)
  97                 if (length(fdays) <= 1)
  98                     return (NA)
  99                 levelToday = data$getLevel(today)
 100                 distances = sapply(fdays, function(i) abs(data$getLevel(i)-levelToday))
 101                 #TODO: 2, 10, 3, 12 magic numbers here...
 102                 dist_thresh = 2
 103                 min_neighbs = min(10,length(fdays))
 104                 repeat
 105                 {
 106                     same_pollution = (distances <= dist_thresh)
 107                     nb_neighbs = sum(same_pollution)
 108                     if (nb_neighbs >= min_neighbs) #will eventually happen
 109                         break
 110                     dist_thresh = dist_thresh + 3
 111                 }
 112                 fdays = fdays[same_pollution]
 113                 max_neighbs = 12
 114                 if (nb_neighbs > max_neighbs)
 115                 {
 116                     # Keep only max_neighbs closest neighbors
 117                     fdays = fdays[
 118                         sort(distances[same_pollution],index.return=TRUE)$ix[1:max_neighbs] ]
 119                 }
 120                 if (length(fdays) == 1) #the other extreme...
 121                 {
 122                     if (final_call)
 123                     {
 124                         private$.params$weights <- 1
 125                         private$.params$indices <- fdays
 126                         private$.params$window <- 1
 127                     }
 128                     return ( data$getSerie(fdays[1])[1:horizon] ) #what else?!
 129                 }
 130             }
 131             else
 132                 fdays = fdays_cut #no conditioning
 133
 134             if (simtype == "endo" || simtype == "mix")
 135             {
 136                 # Compute endogen similarities using given window
 137                 window_endo = ifelse(simtype=="mix", window[1], window)
 138
 139                 # Distances from last observed day to days in the past
 140                 serieToday = data$getSerie(today)
 141                 distances2 = sapply(fdays, function(i) {
 142                     delta = serieToday - data$getSerie(i)
 143                     mean(delta^2)
 144                 })
 145
 146                 sd_dist = sd(distances2)
 147                 if (sd_dist < .25 * sqrt(.Machine$double.eps))
 148                 {
 149 #                   warning("All computed distances are very close: stdev too small")
 150                     sd_dist = 1 #mostly for tests... FIXME:
 151                 }
 152                 simils_endo = exp(-distances2/(sd_dist*window_endo^2))
 153             }
 154
 155             if (simtype == "exo" || simtype == "mix")
 156             {
 157                 # Compute exogen similarities using given window
 158                 window_exo = ifelse(simtype=="mix", window[2], window)
 159
 160                 M = matrix( nrow=1+length(fdays), ncol=1+length(data$getExo(today)) )
 161                 M[1,] = c( data$getLevel(today), as.double(data$getExo(today)) )
 162                 for (i in seq_along(fdays))
 163                     M[i+1,] = c( data$getLevel(fdays[i]), as.double(data$getExo(fdays[i])) )
 164
 165                 sigma = cov(M) #NOTE: robust covariance is way too slow
 166                 # TODO: 10 == magic number; more robust way == det, or always ginv()
 167                 sigma_inv =
 168                     if (length(fdays) > 10)
 169                         solve(sigma)
 170                     else
 171                         MASS::ginv(sigma)
 172
 173                 # Distances from last observed day to days in the past
 174                 distances2 = sapply(seq_along(fdays), function(i) {
 175                     delta = M[1,] - M[i+1,]
 176                     delta %*% sigma_inv %*% delta
 177                 })
 178
 179                 sd_dist = sd(distances2)
 180                 if (sd_dist < .25 * sqrt(.Machine$double.eps))
 181                 {
 182 #                   warning("All computed distances are very close: stdev too small")
 183                     sd_dist = 1 #mostly for tests... FIXME:
 184                 }
 185                 simils_exo = exp(-distances2/(sd_dist*window_exo^2))
 186             }
 187
 188             similarities =
 189                 if (simtype == "exo")
 190                     simils_exo
 191                 else if (simtype == "endo")
 192                     simils_endo
 193                 else if (simtype == "mix")
 194                     simils_endo * simils_exo
 195                 else #none
 196                     rep(1, length(fdays))
 197             similarities = similarities / sum(similarities)
 198
 199             prediction = rep(0, horizon)
 200             for (i in seq_along(fdays))
 201                 prediction = prediction + similarities[i] * data$getSerie(fdays[i]+1)[1:horizon]
 202
 203             if (final_call)
 204             {
 205                 private$.params$weights <- similarities
 206                 private$.params$indices <- fdays
 207                 private$.params$window <-
 208                     if (simtype=="endo")
 209                         window_endo
 210                     else if (simtype=="exo")
 211                         window_exo
 212                     else if (simtype=="mix")
 213                         c(window_endo,window_exo)
 214                     else #none
 215                         1
 216             }
 217
 218             return (prediction)
 219         }
 220     )
 221 )