ksumHash {FLSSS} | R Documentation |
Build k-sum accelerator
Description
Compute k-sum lookup tables given a set.
Usage
ksumHash(
ksumK,
V,
ksumTableSizeScaler = 30L,
target = NULL,
len = 0L,
approxNinstance = 1000L,
verbose = TRUE,
maxCore = 7L
)
Arguments
ksumK |
See the same argument in |
V |
See the same argument in |
ksumTableSizeScaler |
See the same argument in |
target |
See the same argument in |
len |
See the same argument in |
approxNinstance |
See the same argument in |
verbose |
See the same argument in |
maxCore |
See the same argument in |
Details
k-sums are hashed using Yann Collet's xxHash that is the fastest among all non-cryptographic hash algorithms by 202204. See the benchmark <https://github.com/Cyan4973/xxHash>.
Value
Either an empty list (happens when, e.g. ksumK < 3
), or a list of lists. The first list would be the 3-sum lookup table, and the last would be the ksumK
-sum lookup table.
Examples
set.seed(42)
d = 5L # Set dimension.
N = 30L # Set size.
len = 10L # Subset size.
roundN = 2L # For rounding the numeric values before conversion to strings.
V = matrix(round(runif(N * d, -1e5, 1e5), roundN), nrow = N) # Make superset.
sol = sample(N, len) # Make a solution.
target = round(colSums(V[sol, ]), roundN) # Target subset sum.
optionSave = options()
options(scipen = 999) # Ensure numeric => string conversion does not
# produce strings like 2e-3.
Vstr = matrix(as.character(V), nrow = N) # String version of V.
targetStr = as.character(target)
system.time({
theDecomposed = FLSSS::decomposeArbFLSSS(
len = len, V = Vstr, target = targetStr, approxNinstance = 1000,
maxCore = 2, ksumTable = NULL, ksumK = 4, verbose = TRUE)
})
# Run the objects sequentially.
rst = unlist(lapply(theDecomposed$arbFLSSSobjects, function(x)
{
FLSSS::arbFLSSSobjRun(x, solutionNeed = 1e9, tlimit = 5, verbose = FALSE)
}), recursive = FALSE)
str(rst)
options(optionSave)