Commit | Line | Data |
---|---|---|
25b75559 | 1 | #' Neighbors Forecaster |
3d69ff21 | 2 | #' |
c4c329f6 BA |
3 | #' Predict next serie as a weighted combination of "futures of the past" days, |
4 | #' where days in the past are chosen and weighted according to some similarity measures. | |
5 | #' See 'details' section. | |
6 | #' | |
7 | #' TODO: details. | |
8 | #' | |
9 | #' @format R6 class, inherits Forecaster | |
10 | #' @alias F_Neighbors | |
546b0cb6 | 11 | #' |
25b75559 | 12 | NeighborsForecaster = R6::R6Class("NeighborsForecaster", |
a66a84b5 | 13 | inherit = Forecaster, |
25b75559 BA |
14 | |
15 | public = list( | |
98e958ca | 16 | predictShape = function(data, today, memory, horizon, ...) |
3d69ff21 BA |
17 | { |
18 | # (re)initialize computed parameters | |
a66a84b5 | 19 | private$.params <- list("weights"=NA, "indices"=NA, "window"=NA) |
3d69ff21 | 20 | |
a5a3a294 BA |
21 | # Do not forecast on days with NAs (TODO: softer condition...) |
22 | if (any(is.na(data$getCenteredSerie(today)))) | |
23 | return (NA) | |
24 | ||
af3b84f4 | 25 | # Determine indices of no-NAs days followed by no-NAs tomorrows |
98e958ca | 26 | fdays = getNoNA2(data, max(today-memory,1), today-1) |
af3b84f4 | 27 | |
f17665c7 | 28 | # Get optional args |
445e7bbc BA |
29 | local = ifelse(hasArg("local"), list(...)$local, TRUE) #same level + season? |
30 | simtype = ifelse(hasArg("simtype"), list(...)$simtype, "none") #or "endo", or "exo" | |
aa059de7 | 31 | if (hasArg("window")) |
a66a84b5 | 32 | { |
98e958ca | 33 | return ( private$.predictShapeAux(data, |
aa059de7 | 34 | fdays, today, horizon, local, list(...)$window, simtype, TRUE) ) |
a66a84b5 | 35 | } |
3d69ff21 | 36 | |
6774e53d | 37 | # Indices of similar days for cross-validation; TODO: 20 = magic number |
aa059de7 BA |
38 | cv_days = getSimilarDaysIndices(today, data, limit=20, same_season=FALSE, |
39 | days_in=fdays) | |
5e838b3e | 40 | |
445e7bbc | 41 | # Optimize h : h |--> sum of prediction errors on last N "similar" days |
aa059de7 | 42 | errorOnLastNdays = function(window, simtype) |
3d69ff21 BA |
43 | { |
44 | error = 0 | |
45 | nb_jours = 0 | |
5e838b3e | 46 | for (i in seq_along(cv_days)) |
3d69ff21 | 47 | { |
f17665c7 | 48 | # mix_strategy is never used here (simtype != "mix"), therefore left blank |
aa059de7 BA |
49 | prediction = private$.predictShapeAux(data, fdays, cv_days[i], horizon, local, |
50 | window, simtype, FALSE) | |
f17665c7 | 51 | if (!is.na(prediction[1])) |
3d69ff21 BA |
52 | { |
53 | nb_jours = nb_jours + 1 | |
af3b84f4 | 54 | error = error + |
aa059de7 | 55 | mean((data$getSerie(cv_days[i]+1)[1:horizon] - prediction)^2) |
3d69ff21 BA |
56 | } |
57 | } | |
58 | return (error / nb_jours) | |
59 | } | |
60 | ||
445e7bbc | 61 | # TODO: 7 == magic number |
eef54517 | 62 | if (simtype=="endo" || simtype=="mix") |
af3b84f4 | 63 | { |
aa059de7 BA |
64 | best_window_endo = optimize( |
65 | errorOnLastNdays, c(0,7), simtype="endo")$minimum | |
af3b84f4 | 66 | } |
eef54517 | 67 | if (simtype=="exo" || simtype=="mix") |
af3b84f4 | 68 | { |
eef54517 BA |
69 | best_window_exo = optimize( |
70 | errorOnLastNdays, c(0,7), simtype="exo")$minimum | |
3d69ff21 | 71 | } |
eef54517 BA |
72 | |
73 | best_window = | |
74 | if (simtype == "endo") | |
75 | best_window_endo | |
76 | else if (simtype == "exo") | |
77 | best_window_exo | |
78 | else if (simtype == "mix") | |
79 | c(best_window_endo,best_window_exo) | |
80 | else #none: value doesn't matter | |
81 | 1 | |
82 | ||
83 | return(private$.predictShapeAux(data, fdays, today, horizon, local, | |
84 | best_window, simtype, TRUE)) | |
25b75559 BA |
85 | } |
86 | ), | |
87 | private = list( | |
3d69ff21 | 88 | # Precondition: "today" is full (no NAs) |
aa059de7 BA |
89 | .predictShapeAux = function(data, fdays, today, horizon, local, window, simtype, |
90 | final_call) | |
3d69ff21 | 91 | { |
aa059de7 BA |
92 | fdays_cut = fdays[ fdays < today ] |
93 | if (length(fdays_cut) <= 1) | |
3d69ff21 BA |
94 | return (NA) |
95 | ||
aa059de7 BA |
96 | if (local) |
97 | { | |
98 | # Neighbors: days in "same season"; TODO: 60 == magic number... | |
99 | fdays = getSimilarDaysIndices(today, data, limit=60, same_season=TRUE, | |
100 | days_in=fdays_cut) | |
101 | if (length(fdays) <= 1) | |
102 | return (NA) | |
103 | levelToday = data$getLevel(today) | |
104 | distances = sapply(fdays, function(i) abs(data$getLevel(i)-levelToday)) | |
445e7bbc | 105 | #TODO: 2, 10, 3, 12 magic numbers here... |
aa059de7 | 106 | dist_thresh = 2 |
445e7bbc | 107 | min_neighbs = min(10,length(fdays)) |
aa059de7 BA |
108 | repeat |
109 | { | |
110 | same_pollution = (distances <= dist_thresh) | |
111 | nb_neighbs = sum(same_pollution) | |
112 | if (nb_neighbs >= min_neighbs) #will eventually happen | |
113 | break | |
114 | dist_thresh = dist_thresh + 3 | |
115 | } | |
116 | fdays = fdays[same_pollution] | |
445e7bbc | 117 | max_neighbs = 12 |
aa059de7 BA |
118 | if (nb_neighbs > max_neighbs) |
119 | { | |
120 | # Keep only max_neighbs closest neighbors | |
121 | fdays = fdays[ | |
122 | sort(distances[same_pollution],index.return=TRUE)$ix[1:max_neighbs] ] | |
123 | } | |
124 | if (length(fdays) == 1) #the other extreme... | |
125 | { | |
126 | if (final_call) | |
127 | { | |
128 | private$.params$weights <- 1 | |
129 | private$.params$indices <- fdays | |
130 | private$.params$window <- 1 | |
131 | } | |
2057c793 | 132 | return ( data$getSerie(fdays[1])[1:horizon] ) |
aa059de7 BA |
133 | } |
134 | } | |
135 | else | |
136 | fdays = fdays_cut #no conditioning | |
137 | ||
445e7bbc | 138 | if (simtype == "endo" || simtype == "mix") |
3d69ff21 | 139 | { |
aa059de7 BA |
140 | # Compute endogen similarities using given window |
141 | window_endo = ifelse(simtype=="mix", window[1], window) | |
3d69ff21 BA |
142 | |
143 | # Distances from last observed day to days in the past | |
5e838b3e BA |
144 | serieToday = data$getSerie(today) |
145 | distances2 = sapply(fdays, function(i) { | |
146 | delta = serieToday - data$getSerie(i) | |
147 | mean(delta^2) | |
148 | }) | |
3d69ff21 BA |
149 | |
150 | sd_dist = sd(distances2) | |
aa059de7 | 151 | if (sd_dist < .25 * sqrt(.Machine$double.eps)) |
546b0cb6 | 152 | { |
fa5b7bfc | 153 | # warning("All computed distances are very close: stdev too small") |
99f83c9a | 154 | sd_dist = 1 #mostly for tests... FIXME: |
546b0cb6 | 155 | } |
aa059de7 | 156 | simils_endo = exp(-distances2/(sd_dist*window_endo^2)) |
3d69ff21 BA |
157 | } |
158 | ||
445e7bbc | 159 | if (simtype == "exo" || simtype == "mix") |
3d69ff21 | 160 | { |
aa059de7 | 161 | # Compute exogen similarities using given window |
445e7bbc | 162 | window_exo = ifelse(simtype=="mix", window[2], window) |
3d69ff21 | 163 | |
25b75559 BA |
164 | M = matrix( nrow=1+length(fdays), ncol=1+length(data$getExo(today)) ) |
165 | M[1,] = c( data$getLevel(today), as.double(data$getExo(today)) ) | |
f17665c7 | 166 | for (i in seq_along(fdays)) |
25b75559 | 167 | M[i+1,] = c( data$getLevel(fdays[i]), as.double(data$getExo(fdays[i])) ) |
3d69ff21 BA |
168 | |
169 | sigma = cov(M) #NOTE: robust covariance is way too slow | |
ee8b1b4e BA |
170 | # TODO: 10 == magic number; more robust way == det, or always ginv() |
171 | sigma_inv = | |
172 | if (length(fdays) > 10) | |
173 | solve(sigma) | |
174 | else | |
175 | MASS::ginv(sigma) | |
3d69ff21 BA |
176 | |
177 | # Distances from last observed day to days in the past | |
5e838b3e BA |
178 | distances2 = sapply(seq_along(fdays), function(i) { |
179 | delta = M[1,] - M[i+1,] | |
180 | delta %*% sigma_inv %*% delta | |
181 | }) | |
3d69ff21 BA |
182 | |
183 | sd_dist = sd(distances2) | |
ee8b1b4e | 184 | if (sd_dist < .25 * sqrt(.Machine$double.eps)) |
546b0cb6 | 185 | { |
fa5b7bfc | 186 | # warning("All computed distances are very close: stdev too small") |
546b0cb6 BA |
187 | sd_dist = 1 #mostly for tests... FIXME: |
188 | } | |
aa059de7 | 189 | simils_exo = exp(-distances2/(sd_dist*window_exo^2)) |
3d69ff21 BA |
190 | } |
191 | ||
3d69ff21 | 192 | similarities = |
f17665c7 | 193 | if (simtype == "exo") |
3d69ff21 | 194 | simils_exo |
f17665c7 BA |
195 | else if (simtype == "endo") |
196 | simils_endo | |
445e7bbc | 197 | else if (simtype == "mix") |
f17665c7 | 198 | simils_endo * simils_exo |
445e7bbc BA |
199 | else #none |
200 | rep(1, length(fdays)) | |
ea5c7e56 | 201 | similarities = similarities / sum(similarities) |
3d69ff21 BA |
202 | |
203 | prediction = rep(0, horizon) | |
a66a84b5 | 204 | for (i in seq_along(fdays)) |
aa059de7 | 205 | prediction = prediction + similarities[i] * data$getSerie(fdays[i]+1)[1:horizon] |
99f83c9a | 206 | |
3d69ff21 BA |
207 | if (final_call) |
208 | { | |
a66a84b5 BA |
209 | private$.params$weights <- similarities |
210 | private$.params$indices <- fdays | |
211 | private$.params$window <- | |
546b0cb6 | 212 | if (simtype=="endo") |
aa059de7 | 213 | window_endo |
546b0cb6 | 214 | else if (simtype=="exo") |
aa059de7 | 215 | window_exo |
eef54517 | 216 | else if (simtype=="mix") |
aa059de7 | 217 | c(window_endo,window_exo) |
eef54517 BA |
218 | else #none |
219 | 1 | |
3d69ff21 | 220 | } |
99f83c9a | 221 | |
3d69ff21 BA |
222 | return (prediction) |
223 | } | |
224 | ) | |
225 | ) |