RunNetSNE: Run Net-SNE.
In schwikowskilab/rNetSNE: Run Net-SNE on R
Description
Usage
Arguments
Examples
Description
Run Net-SNE in three possible modes: "t-SNE for a t-SNE-like embedding, "Learn" for a training Net-SNE on an embedding and "Project" to project a new dataset on a learnt embedding (with "Learn").
Input: Depends on the mode.
Output: Different text (.txt) files depending on the mode saved in a single folder.
Requirment: C++ implementation of Net-SNE installed and functional.
Usage
1
2
3
4
5
6
7
8
9
10
11
12
RunNetSNE(utilis = c("t-SNE", "Learn", "Project"), out.dims = 2L,
max.iter = 1000L, theta = 0.5, step.method = "adam",
NN.layers = 2L, NN.units = 50L, NN.function = "relu", l2.reg.param = 0,
mom.init = 0.5, mom.final = 0.8, mom.switch.iter = 250L,
early.exag.iter = 250L, learn.rate = 0.02,
local.sample = 20L, batch.frac = 0.1, min.sample.Z = 0.1,
sgd = TRUE, seed = -1, verbose = TRUE,
permute.after.iters = NULL, save.iters.cache = NULL,
path.netSNE.dir = path.netSNE.dir,
path.to.bin.train.file = path.to.bin.file, path.to.bin.test.file = NULL,
path.to.simil.file = NULL, path.output.dir, path.ref.embedding = NULL,
path.to.model.dir = NULL, model.prefix = "model_final")
Arguments
utilis
Character; Vector to choose the use of Net-SNE (default: c("t-SNE", "Learn", "Project") for running the three)
out.dims
Integer; Specifies the output dimensionality (default: 2)
max.iter
Integer; Specifies the number of iterations (default: 1e3)
theta
Double; Bounded by 0 and 1, controls the accuracy-efficiency tradeoff in SPTree for gradient computation; 0 means exact (default: 0.5)
step.method
Character; Specifies the gradient step schedule. Possible values: 'adam', 'mom' (momentum), 'mom_gain' (momentum with gains) or 'fixed' (default: adam)
NN.layers
Integer: Number of layers in the Neural Network (default: 2)
NN.units
Integer: Number of units for each layer in the Neural Network (default: 50)
NN.function
Character: Specifies the activation function of the Neural Network. Possible values: 'sigmoid' or 'relu' (default: relu)
l2.reg.param
Numerical; L2 regularization parameter for introducing sparcity in the Neural Network and avoid train data overfitting (default: 0, i.e. no sparcity)
mom.init
Double; Bounded by 0 and 1, defines the momentum used before n=mom.switch.iter iterations (default: 0.5)
mom.final
Double; Bounded by 0 and 1, defines the momentum used after n=mom.switch.iter iterations (default: 0.8)
mom.switch.iter
Integer; Number of iterations before switching the value of momentum (default: 250)
early.exag.iter
Integer; Number of iterations of early exaggeration (default: 250)
learn.rate
Double; Learning rate used for gradient steps (default: 0.02)
local.sample
Integer: Number of local samples for each data point in the mini-batch (default: 20)
batch.frac
Integer: Fraction of data to sample for mini-batch (default: 0.1, i.e. 10%)
min.sample.Z
Double; Minimum fraction of data to use for approximating the normalization factor Z in the gradient (default: 0.1, i.e. 10%)
sgd
Logical; Set to TRUE to use SGD acceleration. If set to FALSE (effective for small datasets), equivalent to t-SNE with an additional backpropagation step to train a neural network (default: TRUE)
seed
Integer; Equivalent to set.seed (default: -1, i.e. use current time as seed)
verbose
Logical; Should the outputs be printed to the console? (default: TRUE)
permute.after.iters
Integer; Number of iterations after which the ordering of data points is repeatidly permuted for fast mini-batching (default: NULL, i.e. permute.after.iters = max.iter)
save.iters.cache
Integer; Number of iterations after which an intermediary embedding is repeatidly recorded. The final embedding (Y_final.txt) is always saved (default: NULL, i.e. no intermediary embedding is recorded, only the definitive one after max.iter)
path.netSNE.dir
Character; The path to the directory containing the executables created after Net-SNE installation (usually : path/to/netsne-master/bin). Must end with the name of directory with exectuables (here: '/bin')
path.to.bin.train.file
Character; The path to the binary file of the train data matrix obtained with Write_binary_file (same as parameter path.to.bin.file in RunBhtSNE)
path.to.bin.test.file
Character; The path to the binary file of the test data matrix obtained with Write_binary_file
path.to.simil.file
Character; The path to the binary file obtained with Compute_similarities
path.output.dir
Character; The path to save to output directory containing the .txt files
path.ref.embedding
Character; The path to the Y_final.txt file produced by RunBhtSNE (or RunNetSNE in 't-SNE' mode) used as a reference embedding to learn
path.to.model.dir
Character; The path to the directory obtained with RunNetSNE in 'Learn' mode, containing 'model_final*.txt' files used to project new data.
model.prefix
Charatcer; Prefix of the text files used as model for RunNetSNE 'Project' mode (default: 'model_final')
Examples
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
##---- Should be DIRECTLY executable !! ----
##-- ==> Define data, use random,
##-- or do help(data=index) for the standard data sets.
## The function is currently defined as
function (utilis = c("t-SNE", "Learn", "Project"), out.dims = 2L,
max.iter = 1000L, theta = 0.5, step.method = "adam", NN.layers = 2L,
NN.units = 50L, NN.function = "relu", mom.init = 0.5, mom.final = 0.8,
mom.switch.iter = 250L, early.exag.iter = 250L, learn.rate = 0.02,
local.sample = 20L, min.sample.Z = 0.1, l2.reg.param = 0,
sgd = TRUE, seed = -1, verbose = TRUE, batch.frac = 0.1,
random.init = TRUE, permute.after.iters = NULL, save.iters.cache = NULL,
path.netSNE.dir = path.netSNE.dir, path.to.bin.train.file = path.to.bin.file,
path.to.bin.test.file = NULL, path.to.simil.file = NULL,
path.output.dir, path.ref.embedding = NULL, path.to.model.dir = NULL,
model.prefix = "model_final")
{
utilis = tolower(utilis)
for (i in 1:length(utilis)) {
if (!utilis[i] %in% c("t-sne", "tsne", "learn", "train",
"project", "proj", "projection")) {
stop(paste(utilis[i], ": Unknown use of Net-SNE (param: utilis).\nPossible values:\n\t- 't-SNE': t-SNE like embedding\n\t- 'Learn': train Net-SNE on a training dataset to latter project new data on the embedding\n\t- 'Project': project a new dataset on a prior embedding (obtain with utilis = Train)"),
call. = FALSE)
}
}
if (step.method != "adam" & step.method != "mom" & step.method !=
"mom_gain" & step.method != "fixed") {
stop(paste(step.method, ": Unknown gradient step schedule (param: step.method).\nPossible values:'adam', 'mom' (momentum), 'mom_gain' (momentum with gains) or 'fixed'"),
call. = FALSE)
}
if (NN.function != "relu" & NN.function != "sigmoid") {
stop(paste(NN.function, ": Unknown activation function of the neural network (param: NN.function).\nPossible values:'relu'or 'sigmoid'"),
call. = FALSE)
}
path.netSNE.dir = paste(path.netSNE.dir, "RunNetsne", sep = "/")
if ("t-sne" %in% utilis | "tsne" %in% utilis) {
command = paste(path.netSNE.dir, "--input-P", path.to.simil.file,
"--input-X", path.to.bin.train.file, "--out-dir",
path.output.dir, sep = " ")
command = paste(command, "--out-dim", out.dims, "--max-iter",
max.iter, "--rand-seed", seed, "--theta", theta,
"--learn-rate", learn.rate, "--mom-init", mom.init,
"--mom-final", mom.final, "--mom-switch-iter", mom.switch.iter,
"--early-exag-iter", early.exag.iter, "--num-local-sample",
local.sample, "--batch-frac", batch.frac, "--min-sample-Z",
min.sample.Z, "--l2-reg", l2.reg.param, "--step-method",
step.method, "--num-layers", NN.layers, "--num-units",
NN.units, "--act-fn", NN.function, sep = " ")
if (!random.init) {
command = paste(command, "--skip-random-init", sep = " ")
}
if (!is.null(permute.after.iters)) {
command = paste(command, "--perm-iter", permute.after.iters,
sep = " ")
}
if (!is.null(save.iters.cache)) {
command = paste(command, "--cache-iter", save.iters.cache,
sep = " ")
}
if (!sgd) {
command = paste(command, "--no-sgd", sep = " ")
}
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (t-SNE like)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (t-SNE like)\n\n")
}
system(command)
}
if ("learn" %in% utilis | "train" %in% utilis) {
command = paste(path.netSNE.dir, "--input-Y", path.ref.embedding,
"--input-X", path.to.bin.train.file, "--out-dir",
path.output.dir, sep = " ")
command = paste(command, "--out-dim", out.dims, "--max-iter",
max.iter, "--rand-seed", seed, "--theta", theta,
"--learn-rate", learn.rate, "--mom-init", mom.init,
"--mom-final", mom.final, "--mom-switch-iter", mom.switch.iter,
"--early-exag-iter", early.exag.iter, "--num-local-sample",
local.sample, "--batch-frac", batch.frac, "--min-sample-Z",
min.sample.Z, "--l2-reg", l2.reg.param, "--step-method",
step.method, "--num-layers", NN.layers, "--num-units",
NN.units, "--act-fn", NN.function, sep = " ")
if (!random.init) {
command = paste(command, "--skip-random-init", sep = " ")
}
if (!is.null(permute.after.iters)) {
command = paste(command, "--perm-iter", permute.after.iters,
sep = " ")
}
if (!is.null(save.iters.cache)) {
command = paste(command, "--cache-iter", save.iters.cache,
sep = " ")
}
if (!sgd) {
command = paste(command, "--no-sgd", sep = " ")
}
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (learnining embedding)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (learnining embedding)\n\n")
}
system(command)
}
if ("projection" %in% utilis | "project" %in% utilis | "proj" %in%
utilis) {
command = paste(path.netSNE.dir, "--input-X", path.to.bin.test.file,
"--init-model-prefix", paste(path.to.model.dir, model.prefix,
sep = "/"), "--test-model", "--no-target", "--out-dir",
path.output.dir, sep = " ")
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (projection)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (projection)\n\n")
}
system(command)
}
}
schwikowskilab/rNetSNE documentation built on May 4, 2019, 6:40 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Description Usage Arguments Examples
Description
Run Net-SNE in three possible modes: "t-SNE for a t-SNE-like embedding, "Learn" for a training Net-SNE on an embedding and "Project" to project a new dataset on a learnt embedding (with "Learn").
Input: Depends on the mode.
Output: Different text (.txt) files depending on the mode saved in a single folder.
Requirment: C++ implementation of Net-SNE installed and functional.
Usage
1 2 3 4 5 6 7 8 9 10 11 12 | RunNetSNE(utilis = c("t-SNE", "Learn", "Project"), out.dims = 2L,
max.iter = 1000L, theta = 0.5, step.method = "adam",
NN.layers = 2L, NN.units = 50L, NN.function = "relu", l2.reg.param = 0,
mom.init = 0.5, mom.final = 0.8, mom.switch.iter = 250L,
early.exag.iter = 250L, learn.rate = 0.02,
local.sample = 20L, batch.frac = 0.1, min.sample.Z = 0.1,
sgd = TRUE, seed = -1, verbose = TRUE,
permute.after.iters = NULL, save.iters.cache = NULL,
path.netSNE.dir = path.netSNE.dir,
path.to.bin.train.file = path.to.bin.file, path.to.bin.test.file = NULL,
path.to.simil.file = NULL, path.output.dir, path.ref.embedding = NULL,
path.to.model.dir = NULL, model.prefix = "model_final")
|
Arguments
utilis |
Character; Vector to choose the use of Net-SNE (default: c("t-SNE", "Learn", "Project") for running the three) |
out.dims |
Integer; Specifies the output dimensionality (default: 2) |
max.iter |
Integer; Specifies the number of iterations (default: 1e3) |
theta |
Double; Bounded by 0 and 1, controls the accuracy-efficiency tradeoff in SPTree for gradient computation; 0 means exact (default: 0.5) |
step.method |
Character; Specifies the gradient step schedule. Possible values: 'adam', 'mom' (momentum), 'mom_gain' (momentum with gains) or 'fixed' (default: adam) |
NN.layers |
Integer: Number of layers in the Neural Network (default: 2) |
NN.units |
Integer: Number of units for each layer in the Neural Network (default: 50) |
NN.function |
Character: Specifies the activation function of the Neural Network. Possible values: 'sigmoid' or 'relu' (default: relu) |
l2.reg.param |
Numerical; L2 regularization parameter for introducing sparcity in the Neural Network and avoid train data overfitting (default: 0, i.e. no sparcity) |
mom.init |
Double; Bounded by 0 and 1, defines the momentum used before n=mom.switch.iter iterations (default: 0.5) |
mom.final |
Double; Bounded by 0 and 1, defines the momentum used after n=mom.switch.iter iterations (default: 0.8) |
mom.switch.iter |
Integer; Number of iterations before switching the value of momentum (default: 250) |
early.exag.iter |
Integer; Number of iterations of early exaggeration (default: 250) |
learn.rate |
Double; Learning rate used for gradient steps (default: 0.02) |
local.sample |
Integer: Number of local samples for each data point in the mini-batch (default: 20) |
batch.frac |
Integer: Fraction of data to sample for mini-batch (default: 0.1, i.e. 10%) |
min.sample.Z |
Double; Minimum fraction of data to use for approximating the normalization factor Z in the gradient (default: 0.1, i.e. 10%) |
sgd |
Logical; Set to TRUE to use SGD acceleration. If set to FALSE (effective for small datasets), equivalent to t-SNE with an additional backpropagation step to train a neural network (default: TRUE) |
seed |
Integer; Equivalent to set.seed (default: -1, i.e. use current time as seed) |
verbose |
Logical; Should the outputs be printed to the console? (default: TRUE) |
permute.after.iters |
Integer; Number of iterations after which the ordering of data points is repeatidly permuted for fast mini-batching (default: NULL, i.e. permute.after.iters = max.iter) |
save.iters.cache |
Integer; Number of iterations after which an intermediary embedding is repeatidly recorded. The final embedding (Y_final.txt) is always saved (default: NULL, i.e. no intermediary embedding is recorded, only the definitive one after max.iter) |
path.netSNE.dir |
Character; The path to the directory containing the executables created after Net-SNE installation (usually : path/to/netsne-master/bin). Must end with the name of directory with exectuables (here: '/bin') |
path.to.bin.train.file |
Character; The path to the binary file of the train data matrix obtained with Write_binary_file (same as parameter path.to.bin.file in RunBhtSNE) |
path.to.bin.test.file |
Character; The path to the binary file of the test data matrix obtained with Write_binary_file |
path.to.simil.file |
Character; The path to the binary file obtained with Compute_similarities |
path.output.dir |
Character; The path to save to output directory containing the .txt files |
path.ref.embedding |
Character; The path to the Y_final.txt file produced by RunBhtSNE (or RunNetSNE in 't-SNE' mode) used as a reference embedding to learn |
path.to.model.dir |
Character; The path to the directory obtained with RunNetSNE in 'Learn' mode, containing 'model_final*.txt' files used to project new data. |
model.prefix |
Charatcer; Prefix of the text files used as model for RunNetSNE 'Project' mode (default: 'model_final') |
Examples
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 | ##---- Should be DIRECTLY executable !! ----
##-- ==> Define data, use random,
##-- or do help(data=index) for the standard data sets.
## The function is currently defined as
function (utilis = c("t-SNE", "Learn", "Project"), out.dims = 2L,
max.iter = 1000L, theta = 0.5, step.method = "adam", NN.layers = 2L,
NN.units = 50L, NN.function = "relu", mom.init = 0.5, mom.final = 0.8,
mom.switch.iter = 250L, early.exag.iter = 250L, learn.rate = 0.02,
local.sample = 20L, min.sample.Z = 0.1, l2.reg.param = 0,
sgd = TRUE, seed = -1, verbose = TRUE, batch.frac = 0.1,
random.init = TRUE, permute.after.iters = NULL, save.iters.cache = NULL,
path.netSNE.dir = path.netSNE.dir, path.to.bin.train.file = path.to.bin.file,
path.to.bin.test.file = NULL, path.to.simil.file = NULL,
path.output.dir, path.ref.embedding = NULL, path.to.model.dir = NULL,
model.prefix = "model_final")
{
utilis = tolower(utilis)
for (i in 1:length(utilis)) {
if (!utilis[i] %in% c("t-sne", "tsne", "learn", "train",
"project", "proj", "projection")) {
stop(paste(utilis[i], ": Unknown use of Net-SNE (param: utilis).\nPossible values:\n\t- 't-SNE': t-SNE like embedding\n\t- 'Learn': train Net-SNE on a training dataset to latter project new data on the embedding\n\t- 'Project': project a new dataset on a prior embedding (obtain with utilis = Train)"),
call. = FALSE)
}
}
if (step.method != "adam" & step.method != "mom" & step.method !=
"mom_gain" & step.method != "fixed") {
stop(paste(step.method, ": Unknown gradient step schedule (param: step.method).\nPossible values:'adam', 'mom' (momentum), 'mom_gain' (momentum with gains) or 'fixed'"),
call. = FALSE)
}
if (NN.function != "relu" & NN.function != "sigmoid") {
stop(paste(NN.function, ": Unknown activation function of the neural network (param: NN.function).\nPossible values:'relu'or 'sigmoid'"),
call. = FALSE)
}
path.netSNE.dir = paste(path.netSNE.dir, "RunNetsne", sep = "/")
if ("t-sne" %in% utilis | "tsne" %in% utilis) {
command = paste(path.netSNE.dir, "--input-P", path.to.simil.file,
"--input-X", path.to.bin.train.file, "--out-dir",
path.output.dir, sep = " ")
command = paste(command, "--out-dim", out.dims, "--max-iter",
max.iter, "--rand-seed", seed, "--theta", theta,
"--learn-rate", learn.rate, "--mom-init", mom.init,
"--mom-final", mom.final, "--mom-switch-iter", mom.switch.iter,
"--early-exag-iter", early.exag.iter, "--num-local-sample",
local.sample, "--batch-frac", batch.frac, "--min-sample-Z",
min.sample.Z, "--l2-reg", l2.reg.param, "--step-method",
step.method, "--num-layers", NN.layers, "--num-units",
NN.units, "--act-fn", NN.function, sep = " ")
if (!random.init) {
command = paste(command, "--skip-random-init", sep = " ")
}
if (!is.null(permute.after.iters)) {
command = paste(command, "--perm-iter", permute.after.iters,
sep = " ")
}
if (!is.null(save.iters.cache)) {
command = paste(command, "--cache-iter", save.iters.cache,
sep = " ")
}
if (!sgd) {
command = paste(command, "--no-sgd", sep = " ")
}
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (t-SNE like)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (t-SNE like)\n\n")
}
system(command)
}
if ("learn" %in% utilis | "train" %in% utilis) {
command = paste(path.netSNE.dir, "--input-Y", path.ref.embedding,
"--input-X", path.to.bin.train.file, "--out-dir",
path.output.dir, sep = " ")
command = paste(command, "--out-dim", out.dims, "--max-iter",
max.iter, "--rand-seed", seed, "--theta", theta,
"--learn-rate", learn.rate, "--mom-init", mom.init,
"--mom-final", mom.final, "--mom-switch-iter", mom.switch.iter,
"--early-exag-iter", early.exag.iter, "--num-local-sample",
local.sample, "--batch-frac", batch.frac, "--min-sample-Z",
min.sample.Z, "--l2-reg", l2.reg.param, "--step-method",
step.method, "--num-layers", NN.layers, "--num-units",
NN.units, "--act-fn", NN.function, sep = " ")
if (!random.init) {
command = paste(command, "--skip-random-init", sep = " ")
}
if (!is.null(permute.after.iters)) {
command = paste(command, "--perm-iter", permute.after.iters,
sep = " ")
}
if (!is.null(save.iters.cache)) {
command = paste(command, "--cache-iter", save.iters.cache,
sep = " ")
}
if (!sgd) {
command = paste(command, "--no-sgd", sep = " ")
}
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (learnining embedding)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (learnining embedding)\n\n")
}
system(command)
}
if ("projection" %in% utilis | "project" %in% utilis | "proj" %in%
utilis) {
command = paste(path.netSNE.dir, "--input-X", path.to.bin.test.file,
"--init-model-prefix", paste(path.to.model.dir, model.prefix,
sep = "/"), "--test-model", "--no-target", "--out-dir",
path.output.dir, sep = " ")
if (!verbose) {
command = paste(command, "> /dev/null", sep = " ")
}
if ("crayon" %in% installed.packages()[, 1]) {
cat(crayon::bold(crayon::red("\n\nRunning NetSNE ! (projection)\n\n")))
}
else {
cat("\n\nRunning NetSNE ! (projection)\n\n")
}
system(command)
}
}
|
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.