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