on the way to R6 class + remove truncated days (simplifications)

[talweg.git] / pkg / R / F_Neighbors.R

#' @include Forecaster.R
#'
#' @title Neighbors Forecaster
#'
#' @description Predict tomorrow as a weighted combination of "futures of the past" days.
#'   Inherits \code{\link{Forecaster}}
NeighborsForecaster = setRefClass(
	Class = "NeighborsForecaster",
	contains = "Forecaster",

	methods = list(
		initialize = function(...)
		{
			callSuper(...)
		},
		predictShape = function(today, memory, horizon, ...)
		{
			# (re)initialize computed parameters
			params <<- list("weights"=NA, "indices"=NA, "window"=NA)

			# Get optional args
			simtype = ifelse(hasArg("simtype"), list(...)$simtype, "mix") #or "endo", or "exo"
			kernel = ifelse(hasArg("kernel"), list(...)$kernel, "Gauss") #or "Epan"
			if (hasArg(h_window))
				return (.predictShapeAux(fdays,today,horizon,list(...)$h_window,kernel,simtype,TRUE))

			# HACK for test reports: complete some days with a few NAs, for nicer graphics
			nas_in_serie = is.na(data$getSerie(today))
			if (any(nas_in_serie))
			{
				if (sum(nas_in_serie) >= length(nas_in_serie) / 2)
					return (NA)
				for (i in seq_along(nas_in_serie))
				{
					if (nas_in_serie[i])
					{
						#look left
						left = i-1
						while (left>=1 && nas_in_serie[left])
							left = left-1
						#look right
						right = i+1
						while (right<=length(nas_in_serie) && nas_in_serie[right])
							right = right+1
						#HACK: modify by-reference Data object...
						data$data[[today]]$serie[i] <<-
							if (left==0) data$data[[today]]$serie[right]
							else if (right==0) data$data[[today]]$serie[left]
							else (data$data[[today]]$serie[left] + data$data[[today]]$serie[right]) / 2.
					}
				}
			}

			# Determine indices of no-NAs days followed by no-NAs tomorrows
			first_day = max(today - memory, 1)
			fdays = (first_day:(today-1))[ sapply(first_day:(today-1), function(i) {
				!any(is.na(data$getSerie(i)) | is.na(data$getSerie(i+1)))
			}) ]

			# Indices of similar days for cross-validation; TODO: 45 = magic number
			sdays = getSimilarDaysIndices(today, limit=45, same_season=FALSE)

			# Function to optimize h : h |--> sum of prediction errors on last 45 "similar" days
			errorOnLastNdays = function(h, kernel, simtype)
			{
				error = 0
				nb_jours = 0
				for (i in intersect(fdays,sdays))
				{
					# mix_strategy is never used here (simtype != "mix"), therefore left blank
					prediction = .predictShapeAux(fdays, i, horizon, h, kernel, simtype, FALSE)
					if (!is.na(prediction[1]))
					{
						nb_jours = nb_jours + 1
						error = error + mean((data$getCenteredSerie(i+1)[1:horizon] - prediction)^2)
					}
				}
				return (error / nb_jours)
			}

			if (simtype != "endo")
				h_best_exo = optimize(errorOnLastNdays, c(0,10), kernel=kernel, simtype="exo")$minimum
			if (simtype != "exo")
				h_best_endo = optimize(errorOnLastNdays, c(0,10), kernel=kernel, simtype="endo")$minimum

			if (simtype == "endo")
				return (.predictShapeAux(fdays, today, horizon, h_best_endo, kernel, "endo", TRUE))
			if (simtype == "exo")
				return (.predictShapeAux(fdays, today, horizon, h_best_exo,  kernel, "exo",  TRUE))
			if (simtype == "mix")
			{
				h_best_mix = c(h_best_endo,h_best_exo)
				return (.predictShapeAux(fdays, today, horizon, h_best_mix,  kernel, "mix",  TRUE))
			}
		},
		# Precondition: "today" is full (no NAs)
		.predictShapeAux = function(fdays, today, horizon, h, kernel, simtype, final_call)
		{
			dat = data$data #HACK: faster this way...

			fdays = fdays[ fdays < today ]
			# TODO: 3 = magic number
			if (length(fdays) < 3)
				return (NA)

			if (simtype != "exo")
			{
				h_endo = ifelse(simtype=="mix", h[1], h)

				# Distances from last observed day to days in the past
				distances2 = rep(NA, length(fdays))
				for (i in seq_along(fdays))
				{
					delta = dat[[today]]$serie - dat[[ fdays[i] ]]$serie
					# Require at least half of non-NA common values to compute the distance
					if (sum(is.na(delta)) <= 0) #length(delta)/2)
						distances2[i] = mean(delta^2) #, na.rm=TRUE)
				}

				sd_dist = sd(distances2)
				if (sd_dist < .Machine$double.eps)
					sd_dist = 1 #mostly for tests... FIXME:
				simils_endo =
					if (kernel=="Gauss")
						exp(-distances2/(sd_dist*h_endo^2))
					else { #Epanechnikov
						u = 1 - distances2/(sd_dist*h_endo^2)
						u[abs(u)>1] = 0.
						u
					}
			}

			if (simtype != "endo")
			{
				h_exo = ifelse(simtype=="mix", h[2], h)

				M = matrix( nrow=1+length(fdays), ncol=1+length(dat[[today]]$exo) )
				M[1,] = c( dat[[today]]$level, as.double(dat[[today]]$exo) )
				for (i in seq_along(fdays))
					M[i+1,] = c( dat[[ fdays[i] ]]$level, as.double(dat[[ fdays[i] ]]$exo) )

				sigma = cov(M) #NOTE: robust covariance is way too slow
				sigma_inv = solve(sigma) #TODO: use pseudo-inverse if needed?

				# Distances from last observed day to days in the past
				distances2 = rep(NA, nrow(M)-1)
				for (i in 2:nrow(M))
				{
					delta = M[1,] - M[i,]
					distances2[i-1] = delta %*% sigma_inv %*% delta
				}

				sd_dist = sd(distances2)
				simils_exo =
					if (kernel=="Gauss")
						exp(-distances2/(sd_dist*h_exo^2))
					else { #Epanechnikov
						u = 1 - distances2/(sd_dist*h_exo^2)
						u[abs(u)>1] = 0.
						u
					}
			}

			similarities =
				if (simtype == "exo")
					simils_exo
				else if (simtype == "endo")
					simils_endo
				else #mix
					simils_endo * simils_exo

			prediction = rep(0, horizon)
			for (i in seq_along(fdays_indices))
				prediction = prediction + similarities[i] * dat[[ fdays_indices[i]+1 ]]$serie[1:horizon]
			prediction = prediction / sum(similarities, na.rm=TRUE)

			if (final_call)
			{
				params$weights <<- similarities
				params$indices <<- fdays_indices
				params$window <<-
					if (simtype=="endo") {
						h_endo
					} else if (simtype=="exo") {
						h_exo
					} else { #mix
						c(h_endo,h_exo)
					}
			}

			return (prediction)
		}
	)
)