Commit | Line | Data |
---|---|---|
e030a6e3 | 1 | #' @include Forecaster.R |
3d69ff21 | 2 | #' |
25b75559 | 3 | #' Neighbors Forecaster |
3d69ff21 | 4 | #' |
25b75559 BA |
5 | #' Predict tomorrow as a weighted combination of "futures of the past" days. |
6 | #' Inherits \code{\link{Forecaster}} | |
546b0cb6 | 7 | #' |
25b75559 | 8 | NeighborsForecaster = R6::R6Class("NeighborsForecaster", |
a66a84b5 | 9 | inherit = Forecaster, |
25b75559 BA |
10 | |
11 | public = list( | |
98e958ca | 12 | predictShape = function(data, today, memory, horizon, ...) |
3d69ff21 BA |
13 | { |
14 | # (re)initialize computed parameters | |
a66a84b5 | 15 | private$.params <- list("weights"=NA, "indices"=NA, "window"=NA) |
3d69ff21 | 16 | |
a5a3a294 BA |
17 | # Do not forecast on days with NAs (TODO: softer condition...) |
18 | if (any(is.na(data$getCenteredSerie(today)))) | |
19 | return (NA) | |
20 | ||
af3b84f4 | 21 | # Determine indices of no-NAs days followed by no-NAs tomorrows |
98e958ca | 22 | fdays = getNoNA2(data, max(today-memory,1), today-1) |
af3b84f4 | 23 | |
f17665c7 BA |
24 | # Get optional args |
25 | simtype = ifelse(hasArg("simtype"), list(...)$simtype, "mix") #or "endo", or "exo" | |
26 | kernel = ifelse(hasArg("kernel"), list(...)$kernel, "Gauss") #or "Epan" | |
27 | if (hasArg(h_window)) | |
a66a84b5 | 28 | { |
98e958ca | 29 | return ( private$.predictShapeAux(data, |
a66a84b5 BA |
30 | fdays, today, horizon, list(...)$h_window, kernel, simtype, TRUE) ) |
31 | } | |
3d69ff21 | 32 | |
f17665c7 BA |
33 | # Indices of similar days for cross-validation; TODO: 45 = magic number |
34 | sdays = getSimilarDaysIndices(today, limit=45, same_season=FALSE) | |
3d69ff21 | 35 | |
5e838b3e BA |
36 | cv_days = intersect(fdays,sdays) |
37 | # Limit to 20 most recent matching days (TODO: 20 == magic number) | |
38 | cv_days = sort(cv_days,decreasing=TRUE)[1:min(20,length(cv_days))] | |
39 | ||
3d69ff21 BA |
40 | # Function to optimize h : h |--> sum of prediction errors on last 45 "similar" days |
41 | errorOnLastNdays = function(h, kernel, simtype) | |
42 | { | |
43 | error = 0 | |
44 | nb_jours = 0 | |
5e838b3e | 45 | for (i in seq_along(cv_days)) |
3d69ff21 | 46 | { |
f17665c7 | 47 | # mix_strategy is never used here (simtype != "mix"), therefore left blank |
98e958ca | 48 | prediction = private$.predictShapeAux(data, |
5e838b3e | 49 | fdays, cv_days[i], horizon, h, kernel, simtype, FALSE) |
f17665c7 | 50 | if (!is.na(prediction[1])) |
3d69ff21 BA |
51 | { |
52 | nb_jours = nb_jours + 1 | |
af3b84f4 | 53 | error = error + |
5e838b3e | 54 | mean((data$getCenteredSerie(cv_days[i]+1)[1:horizon] - prediction)^2) |
3d69ff21 BA |
55 | } |
56 | } | |
57 | return (error / nb_jours) | |
58 | } | |
59 | ||
f17665c7 | 60 | if (simtype != "endo") |
af3b84f4 BA |
61 | { |
62 | h_best_exo = optimize( | |
63 | errorOnLastNdays, c(0,10), kernel=kernel, simtype="exo")$minimum | |
64 | } | |
3d69ff21 | 65 | if (simtype != "exo") |
af3b84f4 BA |
66 | { |
67 | h_best_endo = optimize( | |
68 | errorOnLastNdays, c(0,10), kernel=kernel, simtype="endo")$minimum | |
69 | } | |
3d69ff21 BA |
70 | |
71 | if (simtype == "endo") | |
af3b84f4 | 72 | { |
98e958ca | 73 | return (private$.predictShapeAux(data, |
af3b84f4 BA |
74 | fdays, today, horizon, h_best_endo, kernel, "endo", TRUE)) |
75 | } | |
3d69ff21 | 76 | if (simtype == "exo") |
af3b84f4 | 77 | { |
98e958ca | 78 | return (private$.predictShapeAux(data, |
af3b84f4 BA |
79 | fdays, today, horizon, h_best_exo, kernel, "exo", TRUE)) |
80 | } | |
3d69ff21 BA |
81 | if (simtype == "mix") |
82 | { | |
f17665c7 | 83 | h_best_mix = c(h_best_endo,h_best_exo) |
98e958ca | 84 | return(private$.predictShapeAux(data, |
af3b84f4 | 85 | fdays, today, horizon, h_best_mix, kernel, "mix", TRUE)) |
3d69ff21 | 86 | } |
25b75559 BA |
87 | } |
88 | ), | |
89 | private = list( | |
3d69ff21 | 90 | # Precondition: "today" is full (no NAs) |
98e958ca | 91 | .predictShapeAux = function(data, fdays, today, horizon, h, kernel, simtype, final_call) |
3d69ff21 | 92 | { |
f17665c7 | 93 | fdays = fdays[ fdays < today ] |
3d69ff21 | 94 | # TODO: 3 = magic number |
f17665c7 | 95 | if (length(fdays) < 3) |
3d69ff21 BA |
96 | return (NA) |
97 | ||
98 | if (simtype != "exo") | |
99 | { | |
100 | h_endo = ifelse(simtype=="mix", h[1], h) | |
101 | ||
102 | # Distances from last observed day to days in the past | |
5e838b3e BA |
103 | serieToday = data$getSerie(today) |
104 | distances2 = sapply(fdays, function(i) { | |
105 | delta = serieToday - data$getSerie(i) | |
106 | mean(delta^2) | |
107 | }) | |
3d69ff21 BA |
108 | |
109 | sd_dist = sd(distances2) | |
99f83c9a | 110 | if (sd_dist < .Machine$double.eps) |
546b0cb6 | 111 | { |
fa5b7bfc | 112 | # warning("All computed distances are very close: stdev too small") |
99f83c9a | 113 | sd_dist = 1 #mostly for tests... FIXME: |
546b0cb6 | 114 | } |
3d69ff21 | 115 | simils_endo = |
99f83c9a | 116 | if (kernel=="Gauss") |
3d69ff21 | 117 | exp(-distances2/(sd_dist*h_endo^2)) |
546b0cb6 BA |
118 | else |
119 | { | |
120 | # Epanechnikov | |
3d69ff21 BA |
121 | u = 1 - distances2/(sd_dist*h_endo^2) |
122 | u[abs(u)>1] = 0. | |
123 | u | |
124 | } | |
125 | } | |
126 | ||
127 | if (simtype != "endo") | |
128 | { | |
129 | h_exo = ifelse(simtype=="mix", h[2], h) | |
130 | ||
25b75559 BA |
131 | M = matrix( nrow=1+length(fdays), ncol=1+length(data$getExo(today)) ) |
132 | M[1,] = c( data$getLevel(today), as.double(data$getExo(today)) ) | |
f17665c7 | 133 | for (i in seq_along(fdays)) |
25b75559 | 134 | M[i+1,] = c( data$getLevel(fdays[i]), as.double(data$getExo(fdays[i])) ) |
3d69ff21 BA |
135 | |
136 | sigma = cov(M) #NOTE: robust covariance is way too slow | |
613a986f | 137 | sigma_inv = solve(sigma) #TODO: use pseudo-inverse if needed? |
3d69ff21 BA |
138 | |
139 | # Distances from last observed day to days in the past | |
5e838b3e BA |
140 | distances2 = sapply(seq_along(fdays), function(i) { |
141 | delta = M[1,] - M[i+1,] | |
142 | delta %*% sigma_inv %*% delta | |
143 | }) | |
3d69ff21 BA |
144 | |
145 | sd_dist = sd(distances2) | |
546b0cb6 BA |
146 | if (sd_dist < .Machine$double.eps) |
147 | { | |
fa5b7bfc | 148 | # warning("All computed distances are very close: stdev too small") |
546b0cb6 BA |
149 | sd_dist = 1 #mostly for tests... FIXME: |
150 | } | |
3d69ff21 | 151 | simils_exo = |
f17665c7 | 152 | if (kernel=="Gauss") |
3d69ff21 | 153 | exp(-distances2/(sd_dist*h_exo^2)) |
546b0cb6 BA |
154 | else |
155 | { | |
156 | # Epanechnikov | |
3d69ff21 BA |
157 | u = 1 - distances2/(sd_dist*h_exo^2) |
158 | u[abs(u)>1] = 0. | |
159 | u | |
160 | } | |
161 | } | |
162 | ||
3d69ff21 | 163 | similarities = |
f17665c7 | 164 | if (simtype == "exo") |
3d69ff21 | 165 | simils_exo |
f17665c7 BA |
166 | else if (simtype == "endo") |
167 | simils_endo | |
168 | else #mix | |
169 | simils_endo * simils_exo | |
3d69ff21 BA |
170 | |
171 | prediction = rep(0, horizon) | |
a66a84b5 | 172 | for (i in seq_along(fdays)) |
5c49f6ce | 173 | prediction = prediction + similarities[i] * data$getCenteredSerie(fdays[i]+1)[1:horizon] |
3d69ff21 | 174 | prediction = prediction / sum(similarities, na.rm=TRUE) |
99f83c9a | 175 | |
3d69ff21 BA |
176 | if (final_call) |
177 | { | |
a66a84b5 BA |
178 | private$.params$weights <- similarities |
179 | private$.params$indices <- fdays | |
180 | private$.params$window <- | |
546b0cb6 | 181 | if (simtype=="endo") |
3d69ff21 | 182 | h_endo |
546b0cb6 | 183 | else if (simtype=="exo") |
3d69ff21 | 184 | h_exo |
546b0cb6 | 185 | else #mix |
3d69ff21 | 186 | c(h_endo,h_exo) |
3d69ff21 | 187 | } |
99f83c9a | 188 | |
3d69ff21 BA |
189 | return (prediction) |
190 | } | |
191 | ) | |
192 | ) |