first tests for Neighbors2 after debug; TODO: some missing forecasts
[talweg.git] / pkg / R / F_Neighbors2.R
CommitLineData
5c49f6ce
BA
1#' @include Forecaster.R
2#'
3#' Neighbors2 Forecaster
4#'
5#' Predict tomorrow as a weighted combination of "futures of the past" days.
6#' Inherits \code{\link{Forecaster}}
7#'
8Neighbors2Forecaster = R6::R6Class("Neighbors2Forecaster",
9 inherit = Forecaster,
10
11 public = list(
5e838b3e
BA
12 predictSerie = function(data, today, memory, horizon, ...)
13 {
14 # Parameters (potentially) computed during shape prediction stage
15 predicted_shape = self$predictShape(data, today, memory, horizon, ...)
16# predicted_delta = private$.pjump(data,today,memory,horizon,private$.params,...)
17 # Predicted shape is aligned it on the end of current day + jump
18# predicted_shape+tail(data$getSerie(today),1)-predicted_shape[1]+predicted_delta
19 predicted_shape
20 },
5c49f6ce
BA
21 predictShape = function(data, today, memory, horizon, ...)
22 {
23 # (re)initialize computed parameters
24 private$.params <- list("weights"=NA, "indices"=NA, "window"=NA)
25
26 # Do not forecast on days with NAs (TODO: softer condition...)
27 if (any(is.na(data$getCenteredSerie(today))))
28 return (NA)
29
30 # Determine indices of no-NAs days followed by no-NAs tomorrows
9db234c5 31 fdays = getNoNA2(data, max(today-memory,1), today-1)
5c49f6ce
BA
32
33 # Get optional args
5e838b3e 34 simtype = ifelse(hasArg("simtype"), list(...)$simtype, "mix") #or "endo", or "exo"
5c49f6ce
BA
35 kernel = ifelse(hasArg("kernel"), list(...)$kernel, "Gauss") #or "Epan"
36 if (hasArg(h_window))
37 {
38 return ( private$.predictShapeAux(data,
5e838b3e 39 fdays, today, horizon, list(...)$h_window, kernel, simtype, TRUE) )
5c49f6ce
BA
40 }
41
9db234c5
BA
42 # Indices of similar days for cross-validation; TODO: 45 = magic number
43 sdays = getSimilarDaysIndices(today, limit=45, same_season=FALSE)
44
5e838b3e
BA
45 cv_days = intersect(fdays,sdays)
46 # Limit to 20 most recent matching days (TODO: 20 == magic number)
47 cv_days = sort(cv_days,decreasing=TRUE)[1:min(20,length(cv_days))]
48
5c49f6ce 49 # Function to optimize h : h |--> sum of prediction errors on last 45 "similar" days
5e838b3e 50 errorOnLastNdays = function(h, kernel, simtype)
5c49f6ce
BA
51 {
52 error = 0
53 nb_jours = 0
5e838b3e 54 for (i in seq_along(cv_days))
5c49f6ce
BA
55 {
56 # mix_strategy is never used here (simtype != "mix"), therefore left blank
5e838b3e
BA
57 prediction = private$.predictShapeAux(data,
58 fdays, cv_days[i], horizon, h, kernel, simtype, FALSE)
5c49f6ce
BA
59 if (!is.na(prediction[1]))
60 {
61 nb_jours = nb_jours + 1
62 error = error +
5e838b3e 63 mean((data$getSerie(cv_days[i]+1)[1:horizon] - prediction)^2)
5c49f6ce
BA
64 }
65 }
66 return (error / nb_jours)
67 }
68
5e838b3e
BA
69 if (simtype != "endo")
70 {
71 h_best_exo = optimize(
72 errorOnLastNdays, c(0,10), kernel=kernel, simtype="exo")$minimum
73 }
74 if (simtype != "exo")
75 {
76 h_best_endo = optimize(
77 errorOnLastNdays, c(0,10), kernel=kernel, simtype="endo")$minimum
78 }
79
80 if (simtype == "endo")
81 {
82 return (private$.predictShapeAux(data,
83 fdays, today, horizon, h_best_endo, kernel, "endo", TRUE))
84 }
85 if (simtype == "exo")
86 {
87 return (private$.predictShapeAux(data,
88 fdays, today, horizon, h_best_exo, kernel, "exo", TRUE))
89 }
90 if (simtype == "mix")
91 {
92 h_best_mix = c(h_best_endo,h_best_exo)
93 return(private$.predictShapeAux(data,
94 fdays, today, horizon, h_best_mix, kernel, "mix", TRUE))
95 }
5c49f6ce
BA
96 }
97 ),
98 private = list(
99 # Precondition: "today" is full (no NAs)
5e838b3e 100 .predictShapeAux = function(data, fdays, today, horizon, h, kernel, simtype, final_call)
5c49f6ce
BA
101 {
102 fdays = fdays[ fdays < today ]
103 # TODO: 3 = magic number
104 if (length(fdays) < 3)
105 return (NA)
106
9db234c5
BA
107 # Neighbors: days in "same season"
108 sdays = getSimilarDaysIndices(today, limit=45, same_season=TRUE, data)
109 indices = intersect(fdays,sdays)
110 levelToday = data$getLevel(today)
5e838b3e 111 distances = sapply(indices, function(i) abs(data$getLevel(i)-levelToday))
9db234c5
BA
112 same_pollution = (distances <= 2)
113 if (sum(same_pollution) < 3) #TODO: 3 == magic number
5c49f6ce 114 {
9db234c5
BA
115 same_pollution = (distances <= 5)
116 if (sum(same_pollution) < 3)
117 return (NA)
5c49f6ce 118 }
9db234c5
BA
119 indices = indices[same_pollution]
120
5e838b3e
BA
121 if (simtype != "exo")
122 {
123 h_endo = ifelse(simtype=="mix", h[1], h)
9db234c5 124
5e838b3e
BA
125 # Distances from last observed day to days in the past
126 serieToday = data$getSerie(today)
127 distances2 = sapply(indices, function(i) {
128 delta = serieToday - data$getSerie(i)
129 mean(delta^2)
130 })
5c49f6ce 131
5e838b3e
BA
132 sd_dist = sd(distances2)
133 if (sd_dist < .Machine$double.eps)
134 {
5c49f6ce 135# warning("All computed distances are very close: stdev too small")
5e838b3e
BA
136 sd_dist = 1 #mostly for tests... FIXME:
137 }
138 simils_endo =
139 if (kernel=="Gauss")
140 exp(-distances2/(sd_dist*h_endo^2))
141 else
142 {
143 # Epanechnikov
144 u = 1 - distances2/(sd_dist*h_endo^2)
145 u[abs(u)>1] = 0.
146 u
147 }
5c49f6ce 148 }
5e838b3e
BA
149
150 if (simtype != "endo")
151 {
152 h_exo = ifelse(simtype=="mix", h[2], h)
153
154 M = matrix( nrow=1+length(indices), ncol=1+length(data$getExo(today)) )
155 M[1,] = c( data$getLevel(today), as.double(data$getExo(today)) )
156 for (i in seq_along(indices))
157 M[i+1,] = c( data$getLevel(indices[i]), as.double(data$getExo(indices[i])) )
158
159 sigma = cov(M) #NOTE: robust covariance is way too slow
160# sigma_inv = solve(sigma) #TODO: use pseudo-inverse if needed?
161 sigma_inv = MASS::ginv(sigma)
162#if (final_call) browser()
163 # Distances from last observed day to days in the past
164 distances2 = sapply(seq_along(indices), function(i) {
165 delta = M[1,] - M[i+1,]
166 delta %*% sigma_inv %*% delta
167 })
168
169 sd_dist = sd(distances2)
170 if (sd_dist < .25 * sqrt(.Machine$double.eps))
5c49f6ce 171 {
5e838b3e
BA
172# warning("All computed distances are very close: stdev too small")
173 sd_dist = 1 #mostly for tests... FIXME:
5c49f6ce 174 }
5e838b3e
BA
175 simils_exo =
176 if (kernel=="Gauss")
177 exp(-distances2/(sd_dist*h_exo^2))
178 else
179 {
180 # Epanechnikov
181 u = 1 - distances2/(sd_dist*h_exo^2)
182 u[abs(u)>1] = 0.
183 u
184 }
185 }
5c49f6ce 186
5e838b3e
BA
187 similarities =
188 if (simtype == "exo")
189 simils_exo
190 else if (simtype == "endo")
191 simils_endo
192 else #mix
193 simils_endo * simils_exo
5c49f6ce
BA
194
195 prediction = rep(0, horizon)
9db234c5
BA
196 for (i in seq_along(indices))
197 prediction = prediction + similarities[i] * data$getSerie(indices[i]+1)[1:horizon]
5c49f6ce
BA
198 prediction = prediction / sum(similarities, na.rm=TRUE)
199
200 if (final_call)
201 {
202 private$.params$weights <- similarities
5e838b3e
BA
203 private$.params$indices <- fdays
204 private$.params$window <-
205 if (simtype=="endo")
206 h_endo
207 else if (simtype=="exo")
208 h_exo
209 else #mix
210 c(h_endo,h_exo)
5c49f6ce
BA
211 }
212
213 return (prediction)
214 }
215 )
216)