context("computeMedoidsIndices")
-test_that("serialization + getDataInFile retrieve original data / from matrix",
+test_that("computeMedoidsIndices behave as expected",
{
- data_bin_file = "/tmp/epclust_test_m.bin"
- unlink(data_bin_file)
+ # Generate a gaussian mixture
+ n = 999
+ L = 7
+ medoids = cbind( rep(0,L), rep(-5,L), rep(5,L) )
+ # short series...
+ series = t( rbind( MASS::mvrnorm(n/3, medoids[,1], diag(L)),
+ MASS::mvrnorm(n/3, medoids[,2], diag(L)),
+ MASS::mvrnorm(n/3, medoids[,3], diag(L)) ) )
- #dataset 200 lignes / 30 columns
- data_ascii = matrix(runif(200*30,-10,10),ncol=30)
- nbytes = 4 #lead to a precision of 1e-7 / 1e-8
- endian = "little"
+ # With high probability, medoids indices should resemble 1,1,1,...,2,2,2,...,3,3,3,...
+ require("bigmemory", quietly=TRUE)
+ mi = epclust:::computeMedoidsIndices(bigmemory::as.big.matrix(medoids)@address, series)
+ mi_ref = rep(1:3, each=n/3)
+ expect_lt( mean(mi != mi_ref), 0.01 )
- #Simulate serialization in one single call
- binarize(data_ascii, data_bin_file, 500, ",", nbytes, endian)
- expect_equal(file.info(data_bin_file)$size, length(data_ascii)*nbytes+8)
- for (indices in list(c(1,3,5), 3:13, c(5,20,50), c(75,130:135), 196:200))
- {
- data_lines = getDataInFile(indices, data_bin_file, nbytes, endian)
- expect_equal(data_lines, data_ascii[indices,], tolerance=1e-6)
- }
- unlink(data_bin_file)
-
- #...in several calls (last call complete, next call NULL)
- for (i in 1:20)
- binarize(data_ascii[((i-1)*10+1):(i*10),], data_bin_file, 20, ",", nbytes, endian)
- expect_equal(file.info(data_bin_file)$size, length(data_ascii)*nbytes+8)
- for (indices in list(c(1,3,5), 3:13, c(5,20,50), c(75,130:135), 196:200))
- {
- data_lines = getDataInFile(indices, data_bin_file, nbytes, endian)
- expect_equal(data_lines, data_ascii[indices,], tolerance=1e-6)
- }
- unlink(data_bin_file)
+ # Now with a random matrix, compare with (trusted) R version
+ series = matrix(runif(n*L, min=-7, max=7), nrow=L)
+ mi = epclust:::computeMedoidsIndices(bigmemory::as.big.matrix(medoids)@address, series)
+ mi_ref = R_computeMedoidsIndices(medoids, series)
+ expect_equal(mi, mi_ref)
})
-
-TODO: test computeMedoids + filter
-# #R-equivalent, requiring a matrix (thus potentially breaking "fit-in-memory" hope)
-# mat_meds = medoids[,]
-# mi = rep(NA,nb_series)
-# for (i in 1:nb_series)
-# mi[i] <- which.min( rowSums( sweep(mat_meds, 2, ref_series[i,], '-')^2 ) )
-# rm(mat_meds); gc()
-