-#TODO: toy dataset, check that indices returned are correct + colors
+context("Check that computeFilaments behaves as expected")
-context("Check that getParamsDirs behaves as expected")
-
-test_that("on input of sufficient size, beta is estimated accurately enough", {
- n = 100000
- d = 2
- K = 2
- Pr = c(0.5, 0.5)
-
- betas_ref = array( c(1,0,0,1 , 1,-2,3,1), dim=c(2,2,2) )
- for (i in 1:(dim(betas_ref)[3]))
+test_that("output is as expected on simulated series",
+{
+ x = seq(0,10,0.1)
+ L = length(x)
+ s1 = cos(x)
+ s2 = sin(x)
+ s3 = c( s1[1:(L%/%2)] , s2[(L%/%2+1):L] )
+ #sum((s1-s2)^2) == 97.59381
+ #sum((s1-s3)^2) == 57.03051
+ #sum((s2-s3)^2) == 40.5633
+ s = list( s1, s2, s3 )
+ n = 150
+ series = list()
+ for (i in seq_len(n))
{
- beta_ref = betas_ref[,,i]
- #all parameters are supposed to be of norm 1: thus, normalize beta_ref
- norm2 = sqrt(colSums(beta_ref^2))
- beta_ref = beta_ref / norm2[col(beta_ref)]
-
- io = generateSampleIO(n, d, K, Pr, beta_ref)
- beta = getParamsDirs(io$X, io$Y, K)
- betas = .labelSwitchingAlign(
- array( c(beta_ref,beta), dim=c(d,K,2) ), compare_to="first", ls_mode="exact")
-
- #Some traces: 0 is not well estimated, but others are OK
- cat("\n\nReference parameter matrix:\n")
- print(beta_ref)
- cat("Estimated parameter matrix:\n")
- print(betas[,,2])
- cat("Difference norm (Matrix norm ||.||_1, max. abs. sum on a column)\n")
- diff_norm = norm(beta_ref - betas[,,2])
- cat(diff_norm,"\n")
-
- #NOTE: 0.5 is loose threshold, but values around 0.3 are expected...
- expect_that( diff_norm, is_less_than(0.5) )
+ index = (i%%3) + 1
+ level = mean(s[[index]])
+ serie = s[[index]] - level + rnorm(L,sd=0.05)
+ # 10 series with NAs for index 2
+ if (index == 2 && i >= 60 && i<= 90)
+ serie[sample(seq_len(L),1)] = NA
+ series[[i]] = list("level"=level,"serie"=serie) #no need for more
}
+ data = new("Data", data=series)
+
+ # index 142 : serie type 2
+ f2 = computeFilaments(data, 142, limit=60, plot=FALSE)
+ # Expected output: 22 series of type 3 (closer), then 50-2-10 series of type 2
+ #
+ #
+ #
+ #
+ #
+ #
+ # Simulate shift at origin when predict_at > 0
+ series[2:(n+1)] = series[1:n]
+ series[[1]] = list("level"=0, "serie"=s[[1]][1:(L%/%2)])
+ # index 143 : serie type 3
+ f3 = computeFilaments(data, 143, limit=70, plot=FALSE)
+ # Expected output: 22 series of type 2 (closer) then 50-2 series of type 3
+ # ATTENTION au shift
+ #
+ #
+ # index 144 : serie type 1
+ f1 = computeFilaments(data, 144, limit=50, plot=FALSE)
+ # Expected output: 2 series of type 3 (closer), then 50-2 series of type 1
+ #
+ expect_that( diff_norm, is_less_than(0.5) )
})
-context("Check that getParamsDirs behaves as expected")
+context("Check that dateIndexToInteger behaves as expected")
-test_that("on input of sufficient size, beta is estimated accurately enough", {
- n = 100000
- d = 2
- K = 2
- Pr = c(0.5, 0.5)
+test_that("integer index matches date in data",
+{
- betas_ref = array( c(1,0,0,1 , 1,-2,3,1), dim=c(2,2,2) )
- for (i in 1:(dim(betas_ref)[3]))
- {
- beta_ref = betas_ref[,,i]
- #all parameters are supposed to be of norm 1: thus, normalize beta_ref
- norm2 = sqrt(colSums(beta_ref^2))
- beta_ref = beta_ref / norm2[col(beta_ref)]
- io = generateSampleIO(n, d, K, Pr, beta_ref)
- beta = getParamsDirs(io$X, io$Y, K)
- betas = .labelSwitchingAlign(
- array( c(beta_ref,beta), dim=c(d,K,2) ), compare_to="first", ls_mode="exact")
- #Some traces: 0 is not well estimated, but others are OK
- cat("\n\nReference parameter matrix:\n")
- print(beta_ref)
- cat("Estimated parameter matrix:\n")
- print(betas[,,2])
- cat("Difference norm (Matrix norm ||.||_1, max. abs. sum on a column)\n")
- diff_norm = norm(beta_ref - betas[,,2])
- cat(diff_norm,"\n")
+#TODO: with and without shift at origin (so series values at least forst ones are required)
- #NOTE: 0.5 is loose threshold, but values around 0.3 are expected...
- expect_that( diff_norm, is_less_than(0.5) )
- }
-})
-dateIndexToInteger = function(index, data)
-{
- index = index[1]
- if (is.numeric(index))
- index = as.integer(index)
- if (is.integer(index))
- return (index)
- if (inherits(index, "Date") || is.character(index))
+
+ n = 1500
+ series = list()
+ for (i in seq_len(n))
{
- tryCatch(dt <- as.POSIXct(index), error=function(e) stop("Unrecognized index format"))
- #TODO: tz arg to difftime ?
- integerIndex <- round( (as.numeric( difftime(dt, data$getTime(1)) ))[1] ) + 1
- if (integerIndex > 0 && integerIndex <= data$getSize())
- {
- #WARNING: if series start at date >0h, result must be shifted
- date1 = as.POSIXlt(data$getTime(1)[1])
- date2 = as.POSIXlt(data$getTime(2)[1])
- shift = (date1$year==date2$year && date1$mon==date2$mon && date1$mday==date2$mday)
- return (integerIndex + ifelse(shift,1,0))
- }
- stop("Date outside data range")
+ index = (i%%3) + 1
+ level = mean(s[[index]])
+ serie = s[[index]] - level + rnorm(L,sd=0.05)
+ # 10 series with NAs for index 2
+ if (index == 2 && i >= 60 && i<= 90)
+ serie[sample(seq_len(L),1)] = NA
+ series[[i]] = list("level"=level,"serie"=serie) #no need for more :: si : time !!!
}
- stop("Unrecognized index format")
-}
-
-#' @title integerIndexToDate
-#'
-#' @description Transform an integer index to date index (relative to data)
-#'
-#' @param index Date (or integer) index
-#' @param data Object of class \code{Data}
-#'
-#' @export
-integerIndexToDate = function(index, data)
+ data = new("Data", data=series)
+ dateIndexToInteger = function(index, data)
+})
-context("Check that getParamsDirs behaves as expected")
+context("Check that integerIndexToDate behaves as expected")
-test_that("on input of sufficient size, beta is estimated accurately enough", {
- n = 100000
- d = 2
- K = 2
- Pr = c(0.5, 0.5)
+test_that("date matches index in data",
+{
+ #TODO: with and without shift at origin (so series values at least forst ones are required)
- betas_ref = array( c(1,0,0,1 , 1,-2,3,1), dim=c(2,2,2) )
- for (i in 1:(dim(betas_ref)[3]))
+ n = 1500
+ series = list()
+ for (i in seq_len(n))
{
- beta_ref = betas_ref[,,i]
- #all parameters are supposed to be of norm 1: thus, normalize beta_ref
- norm2 = sqrt(colSums(beta_ref^2))
- beta_ref = beta_ref / norm2[col(beta_ref)]
-
- io = generateSampleIO(n, d, K, Pr, beta_ref)
- beta = getParamsDirs(io$X, io$Y, K)
- betas = .labelSwitchingAlign(
- array( c(beta_ref,beta), dim=c(d,K,2) ), compare_to="first", ls_mode="exact")
-
- #Some traces: 0 is not well estimated, but others are OK
- cat("\n\nReference parameter matrix:\n")
- print(beta_ref)
- cat("Estimated parameter matrix:\n")
- print(betas[,,2])
- cat("Difference norm (Matrix norm ||.||_1, max. abs. sum on a column)\n")
- diff_norm = norm(beta_ref - betas[,,2])
- cat(diff_norm,"\n")
-
- #NOTE: 0.5 is loose threshold, but values around 0.3 are expected...
- expect_that( diff_norm, is_less_than(0.5) )
+ index = (i%%3) + 1
+ level = mean(s[[index]])
+ serie = s[[index]] - level + rnorm(L,sd=0.05)
+ # 10 series with NAs for index 2
+ if (index == 2 && i >= 60 && i<= 90)
+ serie[sample(seq_len(L),1)] = NA
+ series[[i]] = list("level"=level,"serie"=serie) #no need for more :: si : time !!!
}
+ data = new("Data", data=series)
+
+ integerIndexToDate = function(index, data)
})
projects / talweg.git / commitdiff