Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

update #1

Merged
merged 17 commits into from
Dec 9, 2019
Merged

update #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
3046919
move script dir into a parameter.
jlmelville Mar 23, 2019
9a8548e
Remove duplicated conversion of X to vector.
jlmelville Mar 23, 2019
2f1ec75
ensure integer values are coerced to numeric.
jlmelville Mar 23, 2019
2ef9155
Use file.path to build path to FIt-SNE exe.
jlmelville Mar 23, 2019
f9de091
Remove unused parameters.
jlmelville Mar 23, 2019
554e846
Use message instead of cat.
jlmelville Mar 23, 2019
4a0d508
Remove un-necessary semi-colons.
jlmelville Mar 23, 2019
ebf687a
Use <- instead of =.
jlmelville Mar 23, 2019
f591b2c
Make whitespace more consistent.
jlmelville Mar 23, 2019
42481ad
#73: Move FAST_TSNE_SCRIPT_DIR back to a global.
jlmelville Mar 24, 2019
1ad1d79
#73: More whitespace changes.
jlmelville Mar 24, 2019
0b214f1
Remove assignment to unused tmp.
jlmelville Mar 24, 2019
177f0a2
No need to print df var.
jlmelville Mar 24, 2019
a05a5d7
Merge pull request #73 from jlmelville/master
linqiaozhi Mar 25, 2019
5774b07
Fixed nthreads parameter
linqiaozhi Mar 26, 2019
42e4617
update the annoy version
dkobak Dec 9, 2019
2b9d908
Update Python example notebook
dkobak Dec 9, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
3,102 changes: 2,836 additions & 266 deletions examples/test.ipynb
View file
Open in desktop

Large diffs are not rendered by default.

214 changes: 105 additions & 109 deletions fast_tsne.R
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,10 @@
# Note: this script should be sourced as: source('<path to file>', chdir=T)

FAST_TSNE_SCRIPT_DIR <<-getwd()

cat(sprintf("FIt-SNE R wrapper loading.\nFIt-SNE root directory was set to %s\n", FAST_TSNE_SCRIPT_DIR))
FAST_TSNE_SCRIPT_DIR <<- getwd()

message("FIt-SNE R wrapper loading.")
message("FIt-SNE root directory was set to ", FAST_TSNE_SCRIPT_DIR)

# Compute FIt-SNE of a dataset
# dims - dimensionality of the embedding. Default 2.
# perplexity - perplexity is used to determine the
Expand Down Expand Up @@ -68,40 +70,40 @@ cat(sprintf("FIt-SNE R wrapper loading.\nFIt-SNE root directory was set to %s\n"
# details. Default is 1.0
#
fftRtsne <- function(X,
dims=2, perplexity=30, theta=0.5,
check_duplicates=TRUE,
max_iter=1000,
dims = 2, perplexity = 30, theta = 0.5,
max_iter = 1000,
fft_not_bh = TRUE,
ann_not_vptree = TRUE,
stop_early_exag_iter=250,
exaggeration_factor=12.0, no_momentum_during_exag=FALSE,
start_late_exag_iter=-1.0,late_exag_coeff=1.0,
mom_switch_iter=250, momentum=.5, final_momentum=.8, learning_rate=200,
n_trees=50, search_k = -1,rand_seed=-1,
nterms=3, intervals_per_integer=1, min_num_intervals=50,
K=-1, sigma=-30, initialization=NULL,
data_path=NULL, result_path=NULL,
load_affinities=NULL,
fast_tsne_path=NULL, nthreads=0, perplexity_list = NULL,
get_costs = FALSE, df = 1.0,... ) {
version_number = '1.1.0'
stop_early_exag_iter = 250,
exaggeration_factor = 12.0, no_momentum_during_exag = FALSE,
start_late_exag_iter = -1.0, late_exag_coeff = 1.0,
mom_switch_iter = 250, momentum = 0.5, final_momentum = 0.8, learning_rate = 200,
n_trees = 50, search_k = -1, rand_seed = -1,
nterms = 3, intervals_per_integer = 1, min_num_intervals = 50,
K = -1, sigma = -30, initialization = NULL,
data_path = NULL, result_path = NULL,
load_affinities = NULL,
fast_tsne_path = NULL, nthreads = 0, perplexity_list = NULL,
get_costs = FALSE, df = 1.0) {

version_number <- '1.1.0'

if (is.null(fast_tsne_path)) {
if(.Platform$OS.type == "unix") {
fast_tsne_path = sprintf('%s/bin/fast_tsne', FAST_TSNE_SCRIPT_DIR )
if (.Platform$OS.type == "unix") {
fast_tsne_path <- file.path(FAST_TSNE_SCRIPT_DIR, "bin", "fast_tsne")
} else {
fast_tsne_path = sprintf('%s/bin/FItSNE.exe', FAST_TSNE_SCRIPT_DIR)
fast_tsne_path <- file.path(FAST_TSNE_SCRIPT_DIR, "bin", "FItSNE.exe")
}
}

if (is.null(data_path)) {
data_path <- tempfile(pattern='fftRtsne_data_', fileext='.dat')
data_path <- tempfile(pattern = 'fftRtsne_data_', fileext = '.dat')
}
if (is.null(result_path)) {
result_path <- tempfile(pattern='fftRtsne_result_', fileext='.dat')
result_path <- tempfile(pattern = 'fftRtsne_result_', fileext = '.dat')
}
if (is.null(fast_tsne_path)) {
fast_tsne_path <- system2('which', 'fast_tsne', stdout=TRUE)
fast_tsne_path <- system2('which', 'fast_tsne', stdout = TRUE)
}
fast_tsne_path <- normalizePath(fast_tsne_path)
if (!file_test('-x', fast_tsne_path)) {
Expand All @@ -110,115 +112,109 @@ fftRtsne <- function(X,

is.wholenumber <- function(x, tol = .Machine$double.eps^0.5) abs(x - round(x)) < tol

if (!is.numeric(theta) || (theta<0.0) || (theta>1.0) ) { stop("Incorrect theta.")}
if (!is.numeric(theta) || (theta < 0.0) || (theta > 1.0) ) { stop("Incorrect theta.")}
if (nrow(X) - 1 < 3 * perplexity) { stop("Perplexity is too large.")}
if (!is.matrix(X)) { stop("Input X is not a matrix")}
if (!(max_iter>0)) { stop("Incorrect number of iterations.")}
if (!is.wholenumber(stop_early_exag_iter) || stop_early_exag_iter<0) { stop("stop_early_exag_iter should be a positive integer")}
if (!(max_iter > 0)) { stop("Incorrect number of iterations.")}
if (!is.wholenumber(stop_early_exag_iter) || stop_early_exag_iter < 0) { stop("stop_early_exag_iter should be a positive integer")}
if (!is.numeric(exaggeration_factor)) { stop("exaggeration_factor should be numeric")}
if (!is.numeric(df)) { stop("df should be numeric")}
if (!is.wholenumber(dims) || dims<=0) { stop("Incorrect dimensionality.")}
if (!is.wholenumber(dims) || dims <= 0) { stop("Incorrect dimensionality.")}
if (search_k == -1) {
if (perplexity>0) {
search_k = n_trees*perplexity*3
} else if (perplexity==0) {
search_k = n_trees*max(perplexity_list)*3
} else {
search_k = n_trees*K
}
if (perplexity > 0) {
search_k <- n_trees * perplexity * 3
} else if (perplexity == 0) {
search_k <- n_trees * max(perplexity_list) * 3
} else {
search_k <- n_trees * K
}
}

if (fft_not_bh){
nbody_algo = 2;
}else{
nbody_algo = 1;
if (fft_not_bh) {
nbody_algo <- 2
} else {
nbody_algo <- 1
}

if (is.null(load_affinities)) {
load_affinities = 0;
load_affinities <- 0
} else {
if (load_affinities == 'load') {
load_affinities = 1;
load_affinities <- 1
} else if (load_affinities == 'save') {
load_affinities = 2;
load_affinities <- 2
} else {
load_affinities = 0;
load_affinities <- 0
}
}

if (ann_not_vptree){
knn_algo = 1;
}else{
knn_algo = 2;
if (ann_not_vptree) {
knn_algo <- 1
} else {
knn_algo <- 2
}
tX = c(t(X))
tX <- as.numeric(t(X))

f <- file(data_path, "wb")
n = nrow(X);
D = ncol(X);
writeBin(as.integer(n), f,size= 4)
writeBin( as.integer(D),f,size= 4)
writeBin( as.numeric(theta), f,size= 8) #theta
writeBin( as.numeric(perplexity), f,size= 8) #theta

if (perplexity == 0) {
writeBin( as.integer(length(perplexity_list)), f, size=4)
writeBin( perplexity_list, f)
}

writeBin( as.integer(dims), f,size=4) #theta
writeBin( as.integer(max_iter),f,size=4)
writeBin( as.integer(stop_early_exag_iter),f,size=4)
writeBin( as.integer(mom_switch_iter),f,size=4)
writeBin( as.numeric(momentum),f,size=8)
writeBin( as.numeric(final_momentum),f,size=8)
writeBin( as.numeric(learning_rate),f,size=8)
writeBin( as.integer(K),f,size=4) #K
writeBin( as.numeric(sigma), f,size=8) #sigma
writeBin( as.integer(nbody_algo), f,size=4) #not barnes hut
writeBin( as.integer(knn_algo), f,size=4)
writeBin( as.numeric(exaggeration_factor), f,size=8) #compexag
writeBin( as.integer(no_momentum_during_exag), f,size=4)
writeBin( as.integer(n_trees), f,size=4)
writeBin( as.integer(search_k), f,size=4)
writeBin( as.integer(start_late_exag_iter), f,size=4)
writeBin( as.numeric(late_exag_coeff), f,size=8)
n <- nrow(X)
D <- ncol(X)
writeBin(as.integer(n), f, size = 4)
writeBin(as.integer(D), f, size = 4)
writeBin(as.numeric(theta), f, size = 8) #theta
writeBin(as.numeric(perplexity), f, size = 8)

if (perplexity == 0) {
writeBin(as.integer(length(perplexity_list)), f, size = 4)
writeBin(perplexity_list, f)
}

writeBin(as.integer(dims), f, size = 4)
writeBin(as.integer(max_iter), f, size = 4)
writeBin(as.integer(stop_early_exag_iter), f, size = 4)
writeBin(as.integer(mom_switch_iter), f, size = 4)
writeBin(as.numeric(momentum), f, size = 8)
writeBin(as.numeric(final_momentum), f, size = 8)
writeBin(as.numeric(learning_rate), f, size = 8)
writeBin(as.integer(K), f, size = 4) #K
writeBin(as.numeric(sigma), f, size = 8) #sigma
writeBin(as.integer(nbody_algo), f, size = 4) #not barnes hut
writeBin(as.integer(knn_algo), f, size = 4)
writeBin(as.numeric(exaggeration_factor), f, size = 8) #compexag
writeBin(as.integer(no_momentum_during_exag), f, size = 4)
writeBin(as.integer(n_trees), f, size = 4)
writeBin(as.integer(search_k), f, size = 4)
writeBin(as.integer(start_late_exag_iter), f, size = 4)
writeBin(as.numeric(late_exag_coeff), f, size = 8)

writeBin( as.integer(nterms), f,size=4)
writeBin( as.numeric(intervals_per_integer), f,size=8)
writeBin( as.integer(min_num_intervals), f,size=4)
tX = c(t(X))
writeBin( tX, f)
writeBin( as.integer(rand_seed), f,size=4)
writeBin(as.numeric(df), f, size=8)
writeBin( as.integer(load_affinities), f,size=4)
if (! is.null(initialization)){ writeBin( c(t(initialization)), f) }
print(df)
writeBin(as.integer(nterms), f, size = 4)
writeBin(as.numeric(intervals_per_integer), f, size = 8)
writeBin(as.integer(min_num_intervals), f, size = 4)
writeBin(tX, f)
writeBin(as.integer(rand_seed), f, size = 4)
writeBin(as.numeric(df), f, size = 8)
writeBin(as.integer(load_affinities), f, size = 4)
if (!is.null(initialization)) { writeBin( c(t(initialization)), f) }
close(f)

flag= system2(command=fast_tsne_path, args=c(version_number,data_path, result_path, nthreads));
flag <- system2(command = fast_tsne_path,
args = c(version_number, data_path, result_path, nthreads))
if (flag != 0) {
stop('tsne call failed');
stop('tsne call failed')
}
f <- file(result_path, "rb")
n <- readBin(f, integer(), n=1, size=4);
d <- readBin(f, integer(), n=1, size=4);
Y <- readBin(f, numeric(), n=n*d);
Y <- t(matrix(Y, nrow=d));
if (get_costs ) {
tmp <- readBin(f, integer(), n=1, size=4);
costs <- readBin(f, numeric(), n=max_iter,size=8);
Yout <- list( Y=Y, costs=costs);
}else {
Yout <- Y;
}
close(f)
file.remove(data_path)
file.remove(result_path)
return(Yout)
n <- readBin(f, integer(), n = 1, size = 4)
d <- readBin(f, integer(), n = 1, size = 4)
Y <- readBin(f, numeric(), n = n * d)
Y <- t(matrix(Y, nrow = d))
if (get_costs) {
readBin(f, integer(), n = 1, size = 4)
costs <- readBin(f, numeric(), n = max_iter, size = 8)
Yout <- list(Y = Y, costs = costs)
} else {
Yout <- Y
}
close(f)
file.remove(data_path)
file.remove(result_path)
Yout
}