'update'
[epclust.git] / epclust / R / computeCoeffs.R
CommitLineData
5c652979
BA
1computeCoeffs = function(data, index, nb_series_per_chunk, wf)
2{
3 coeffs_chunk = NULL
4 if (is.data.frame(data) && index < nrow(data))
5 {
6 #full data matrix
7 coeffs_chunk = curvesToCoeffs(
8 data[index:(min(index+nb_series_per_chunk-1,nrow(data))),], wf)
9 }
10 else if (is.function(data))
11 {
12 #custom user function to retrieve next n curves, probably to read from DB
13 coeffs_chunk = curvesToCoeffs( data(rank=(index-1)+seq_len(nb_series_per_chunk)), wf )
14 }
15 else if (exists(data_con))
16 {
17 #incremental connection ; TODO: more efficient way to parse than using a temp file
18 ascii_lines = readLines(data_con, nb_series_per_chunk)
19 if (length(ascii_lines > 0))
20 {
21 series_chunk_file = ".series_chunk"
22 writeLines(ascii_lines, series_chunk_file)
23 coeffs_chunk = curvesToCoeffs( read.csv(series_chunk_file), wf )
24 unlink(series_chunk_file)
25 }
26 }
27 coeffs_chunk
28}
29
74f571a7 30#NOTE: always keep ID in first column (...? is it good ?!)
5c652979
BA
31curvesToCoeffs = function(series, wf)
32{
33 if (!require(wavelets, quietly=TRUE))
34 stop("Couldn't load wavelets library")
35 L = length(series[1,])
36 D = ceiling( log2(L-1) )
37 nb_sample_points = 2^D
38 #TODO: parallel::parApply() ?!
39 res = apply(series, 1, function(x) {
40 interpolated_curve = spline(1:(L-1), x[2:L], n=nb_sample_points)$y
41 W = wavelets::dwt(interpolated_curve, filter=wf, D)@W
42 nrj_coeffs = rev( sapply( W, function(v) ( sqrt( sum(v^2) ) ) ) )
43 return ( c(x[1], nrj_coeffs) )
44 })
45 return (as.data.frame(res))
46}