written my understanding of Neighbors2; TODO: sameSeasonSimilarIndices

[talweg.git] / pkg / R / F_Neighbors2.R

#' @include Forecaster.R
#'
#' Neighbors2 Forecaster
#'
#' Predict tomorrow as a weighted combination of "futures of the past" days.
#' Inherits \code{\link{Forecaster}}
#'
Neighbors2Forecaster = R6::R6Class("Neighbors2Forecaster",
	inherit = Forecaster,

	public = list(
		predictShape = function(data, today, memory, horizon, ...)
		{
			# (re)initialize computed parameters
			private$.params <- list("weights"=NA, "indices"=NA, "window"=NA)

			# Do not forecast on days with NAs (TODO: softer condition...)
			if (any(is.na(data$getCenteredSerie(today))))
				return (NA)

			# Determine indices of no-NAs days followed by no-NAs tomorrows
			fdays = getNoNA2(data, max(today-memory,1), today-1)

			# Get optional args
			kernel = ifelse(hasArg("kernel"), list(...)$kernel, "Gauss") #or "Epan"
			if (hasArg(h_window))
			{
				return ( private$.predictShapeAux(data,
					fdays, today, horizon, list(...)$h_window, kernel, TRUE) )
			}

			# 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)
			{
				error = 0
				nb_jours = 0
				for (day in intersect(fdays,sdays))
				{
					# mix_strategy is never used here (simtype != "mix"), therefore left blank
					prediction = private$.predictShapeAux(data,fdays,day,horizon,h,kernel,FALSE)
					if (!is.na(prediction[1]))
					{
						nb_jours = nb_jours + 1
						error = error +
							mean((data$getSerie(i+1)[1:horizon] - prediction)^2)
					}
				}
				return (error / nb_jours)
			}

			# h :: only for endo in this variation
			h_best = optimize(errorOnLastNdays, c(0,7), kernel=kernel)$minimum
			return (private$.predictShapeAux(data,fdays,today,horizon,h_best,kernel,TRUE))
		}
	),
	private = list(
		# Precondition: "today" is full (no NAs)
		.predictShapeAux = function(data, fdays, today, horizon, h, kernel, final_call)
		{
			fdays = fdays[ fdays < today ]
			# TODO: 3 = magic number
			if (length(fdays) < 3)
				return (NA)

			# Neighbors: days in "same season"
			sdays = getSimilarDaysIndices(today, limit=45, same_season=TRUE, data)
			indices = intersect(fdays,sdays)
			levelToday = data$getLevel(today)
			distances = sapply(seq_along(indices), function(i) abs(data$getLevel(i)-levelToday))
			same_pollution = (distances <= 2)
			if (sum(same_pollution) < 3) #TODO: 3 == magic number
			{
				same_pollution = (distances <= 5)
				if (sum(same_pollution) < 3)
					return (NA)
			}
			indices = indices[same_pollution]

			# Now OK: indices same season, same pollution level
			# ...........


			# ENDO:: Distances from last observed day to days in the past
			serieToday = data$getSerie(today)
			distances2 = sapply(indices, function(i) {
				delta = serieToday - data$getSerie(i)
				distances2[i] = mean(delta^2)
			})

			sd_dist = sd(distances2)
			if (sd_dist < .Machine$double.eps)
			{
#					warning("All computed distances are very close: stdev too small")
				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
				}

#			# EXOGENS: distances computations are enough
#			# TODO: search among similar concentrations....... at this stage ?!
#			M = matrix( nrow=1+length(fdays), ncol=1+length(data$getExo(today)) )
#			M[1,] = c( data$getLevel(today), as.double(data$getExo(today)) )
#			for (i in seq_along(fdays))
#				M[i+1,] = c( data$getLevel(fdays[i]), as.double(data$getExo(fdays[i])) )
#
#			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
#			}

			similarities = simils_endo

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

			if (final_call)
			{
				private$.params$weights <- similarities
				private$.params$indices <- indices
				private$.params$window <- h
			}

			return (prediction)
		}
	)
)