Skip to content

Commit 78296e2

Browse files
author
Darius Morawiec
committed
Restructure model data and algorithm handling
1 parent 7d31668 commit 78296e2

File tree

11 files changed

+178
-182
lines changed

11 files changed

+178
-182
lines changed

‎examples/estimator/classifier/MLPClassifier/js/basics.ipynb‎

Lines changed: 62 additions & 54 deletions
Large diffs are not rendered by default.

‎examples/estimator/classifier/MLPClassifier/js/basics.py‎

Lines changed: 55 additions & 47 deletions
Original file line numberDiff line numberDiff line change
@@ -26,81 +26,89 @@
2626
print(output)
2727

2828
"""
29-
// Array.prototype.fill polyfill:
30-
[].fill||(Array.prototype.fill=function(a){for(var b=Object(this),c=parseInt(b.length,10),d=arguments[1],e=parseInt(d,10)||0,f=0>e?Math.max(c+e,0):Math.min(e,c),g=arguments[2],h=void 0===g?c:parseInt(g)||0,i=0>h?Math.max(c+h,0):Math.min(h,c);i>f;f++)b[f]=a;return b});
31-
32-
var Brain = function(layers, weights, bias) {
33-
34-
// Activation function (relu):
35-
var compAct = function(v) {
36-
for (var i = 0, l = v.length; i < l; i++) {
37-
v[i] = Math.max(0, v[i]);
38-
}
39-
return v;
40-
};
41-
42-
// Output function (softmax):
43-
var compOut = function(v) {
44-
var max = Number.NEGATIVE_INFINITY;
45-
for (var i = 0, l = v.length; i < l; i++) {
46-
if (v[i] > max) {
47-
max = v[i];
48-
}
49-
}
50-
for (var i = 0, l = v.length; i < l; i++) {
51-
v[i] = Math.exp(v[i] - max);
52-
}
53-
var sum = 0.0;
54-
for (var i = 0, l = v.length; i < l; i++) {
55-
sum += v[i];
56-
}
57-
for (var i = 0, l = v.length; i < l; i++) {
58-
v[i] /= sum;
29+
var Brain = function(hidden, output, layers, weights, bias) {
30+
31+
this.hidden = hidden.toUpperCase();
32+
this.output = output.toUpperCase();
33+
this.layers = layers;
34+
this.weights = weights;
35+
this.bias = bias;
36+
37+
var compute = function(activation, v) {
38+
switch (activation) {
39+
case 'LOGISTIC':
40+
for (var i = 0, l = v.length; i < l; i++) {
41+
v[i] = 1. / (1. + Math.exp(-v[i]));
42+
}
43+
break;
44+
case 'RELU':
45+
for (var i = 0, l = v.length; i < l; i++) {
46+
v[i] = Math.max(0, v[i]);
47+
}
48+
break;
49+
case 'TANH':
50+
for (var i = 0, l = v.length; i < l; i++) {
51+
v[i] = Math.tanh(v[i]);
52+
}
53+
break;
54+
case 'SOFTMAX':
55+
var max = Number.NEGATIVE_INFINITY;
56+
for (var i = 0, l = v.length; i < l; i++) {
57+
if (v[i] > max) {
58+
max = v[i];
59+
}
60+
}
61+
for (var i = 0, l = v.length; i < l; i++) {
62+
v[i] = Math.exp(v[i] - max);
63+
}
64+
var sum = 0.0;
65+
for (var i = 0, l = v.length; i < l; i++) {
66+
sum += v[i];
67+
}
68+
for (var i = 0, l = v.length; i < l; i++) {
69+
v[i] /= sum;
70+
}
71+
break;
5972
}
6073
return v;
6174
};
6275
6376
this.predict = function(neurons) {
64-
if (neurons.length != 4) return -1;
65-
var network = [neurons].concat(layers);
77+
var network = [neurons].concat(this.layers);
6678
6779
for (var i = 0; i < network.length - 1; i++) {
6880
for (var j = 0; j < network[i + 1].length; j++) {
6981
for (var l = 0; l < network[i].length; l++) {
70-
network[i + 1][j] += network[i][l] * weights[i][l][j];
82+
network[i + 1][j] += network[i][l] * this.weights[i][l][j];
7183
}
72-
network[i + 1][j] += bias[i][j];
84+
network[i + 1][j] += this.bias[i][j];
7385
}
7486
if ((i + 1) < (network.length - 1)) {
75-
network[i + 1] = compAct(network[i + 1]);
87+
network[i + 1] = compute(this.hidden, network[i + 1]);
7688
}
7789
}
78-
network[network.length - 1] = compOut(network[network.length - 1]);
90+
network[network.length - 1] = compute(this.output, network[network.length - 1]);
7991
80-
var classIdx = -1,
81-
classVal = Number.NEGATIVE_INFINITY;
92+
var index = 0;
8293
for (var i = 0, l = network[network.length - 1].length; i < l; i++) {
83-
if (network[network.length - 1][i] > classVal) {
84-
classVal = network[network.length - 1][i];
85-
classIdx = i;
86-
}
94+
index = network[network.length - 1][i] > network[network.length - 1][index] ? i : index;
8795
}
88-
return classIdx;
96+
return index;
8997
};
9098
9199
};
92100
93-
// Model data:
101+
// Parameters:
94102
const layers = [new Array(50).fill(.0), new Array(3).fill(.0)];
95103
const weights = [[[-0.055317816158370905, -0.25162425407767391, -0.33325197861130057, -0.13177626632767314, -0.23549246545141095, -0.27177010710624283, -0.20915665516342635, 0.015057965302911078, -0.72987930075766605, -0.89266106096454045, -0.053869498543610707, -0.22216066222318268, -0.39899502812673082, -0.28681010148595915, -0.31507011153899617, -0.6703997317135284, -0.46627854428112819, -0.55220519291597536, -0.65753556794341805, -0.2012625909486952, 0.33662471612525208, 0.25425126671040588, -0.12438197591047517, 0.22085158121433768, -0.43004111901975961, 0.12047821481685615, -0.27663295488184064, -0.30729210399464546, -0.2201096819240726, 0.19698947979492068, -0.26776467575302032, -0.55963223046623012, -0.66057684440281672, 0.022109851297338449, 0.054364802303907972, -0.30531205478429857, -0.94689187532840058, -1.1472602273053634, -0.32113622918572421, -0.97806246488102366, 0.12604362195657265, 0.52310425825050821, -0.14636842984592813, -0.11855589566328117, -0.2645119437495197, -0.034737117103522681, -0.057695932963591608, -0.13758846013369239, -0.14148094036584954, -0.24664384043086116], [-0.32041711985049837, -0.043058968510450109, -0.19224497780614169, -0.15629983284213439, -0.0056178077580114668, -0.29775374218304873, 0.04941098010393874, -0.261639308519379, -0.19756633578664284, -0.5892530836660409, -0.26510631981125071, -0.28998233921283872, -0.0419201891702871, -0.43290432093263115, -0.30002643148337588, -0.48659301663778098, -0.081521874811590178, -0.27842983575218716, -0.039622604256915966, 0.05770247643254544, 0.71332882051710444, -0.36349619254759175, -0.24047875144770522, 0.53045027219372654, -0.59792191320512289, -0.35631856975018028, 0.28500135453630931, -0.10148787221769563, 0.16720550745832014, 0.25971388860710132, 0.2555334798063561, -0.10856461864559649, -0.41610950088536125, -0.10073289562774027, -0.35234810455711829, 0.13500748765932857, -0.59236329802269883, -0.69284565960163447, 0.10895932947719877, -0.53435965541922248, -0.28749631798822212, 0.95302314181559677, -0.03339139951489787, -0.21465662301256946, -0.061241193383340838, -0.17531266101796375, 0.032502649222591065, 0.049118905364866981, -0.33141470527927364, 0.078095412793859842], [-0.11556829517891457, -0.28839659290844372, 0.25729079162402252, -0.095152035767087209, 0.27235259530926254, 0.082238817438121087, -0.32278089365740004, 0.61089504167538256, -0.53946895727666411, 0.66336833631951808, -0.21843631503665203, -0.34459111930044772, 0.26988257637824437, 0.22285364019620213, -0.28932878616881214, -0.16019244916792288, -0.1518820719052954, -0.17375211722189424, 0.12003588634597376, -0.25048218839808589, -1.1814760275905127, -0.1167992478381189, -0.31445752437419289, -0.71528558652206731, 0.47396757392660221, 0.14299307810727685, 0.035214113037884201, 0.22801710207885825, -0.25054728530241616, -0.55191137079060171, 0.057171971863154362, -0.16787314862568956, -0.35815855192213464, -0.3208968914994626, 0.60992105666976504, -0.22919074588760435, -0.63633627319871755, -0.3709104501371045, 0.24235752387495327, -0.60846958788171723, -0.25014809614706662, -1.0335138637201615, -0.29338371286091719, -0.18073925955574419, -0.3036274299231928, -0.26166657231672047, -0.29878288629035105, 0.14199047853995642, 0.039810711538198082, -0.32495770210408736], [-0.28534610919825071, 0.22769872913725114, 0.045399612503103627, -0.19780148024951577, -0.16511364103110895, 0.16254807942801869, -0.20304390231246003, 0.2098484608353903, 0.05685419958788708, 0.62516653349040463, -0.17343217055411875, -0.022987144452160622, 0.079240581612972066, 0.40417766317484555, -0.22880223204823208, -0.42337014160558745, -0.38940003536068196, -0.17326402738615923, 0.08842740443002832, 0.04590025494515363, -0.54686530359318153, 0.42810346180594139, 0.053162665137136862, -0.3210385389422018, 0.31367095881914692, 0.24999765928913104, 0.11282020075848748, -0.15671789424443483, -0.2891056817821327, -0.28253677782850617, 0.086477013336133002, -0.36053269357622619, 0.086846992712735396, -0.28897125255757344, 0.059892059841316708, 0.19412094475437708, -0.53174460572971327, 0.049633740361929246, 0.016446620815376744, 0.0064188476524308074, -0.26556784617072887, -0.65689592060012736, 0.15670824169328113, 0.25181970948115312, 0.27187307037723824, 0.28797696815559204, -0.36711566037352317, -0.17708922988507345, 0.077851045457767137, 0.29933962842647954]], [[-0.33618760940544834, 0.32081899159033467, -0.083051115109439883], [0.0096383304388562799, 0.2419360820188009, 0.35899310789861316], [0.050274587597035091, 0.086184558536937531, -0.14428921022123034], [0.058431568313521046, 0.16824371439906796, 0.2411152128064154], [0.17164895659196974, 0.13327594625746894, 0.24526414111440331], [-0.11932084344947159, 0.11492655724963714, -0.033057727812256854], [-0.079334977171167304, -0.060016494360769022, -0.066295991899890286], [-0.71486824845646157, 0.1600868064828144, 0.46699990709340361], [-0.083938207334024614, 0.27187161898314255, 0.047703604119507262], [-0.58793523568582362, -0.27887692495317457, 0.91235920748291544], [0.25570703360541275, 0.27175182540995058, 0.10949681033724466], [-0.041009631889342278, -0.23307375710751005, 0.19163291322703432], [-0.15189717246377266, -0.025154028942508011, 0.44780865283409321], [-0.31152787299551726, -0.0056627446166318185, 0.66915144386449055], [0.046335662746525126, -0.023074380835666274, -0.10585719439044175], [-0.070740857338026972, -0.51355079283235028, -0.071290569035165757], [0.31604404174783435, -0.21123109460406667, 0.21568944659130201], [-0.1929211346954548, 0.13803331731090368, 0.48075587735422887], [-0.039151124981216867, -0.35635262270414042, 0.040892690463512522], [-0.18502508791073968, 0.06225274230793363, -0.12632668189370647], [1.3423577687187584, -0.73226914108310404, -0.21773848562508957], [-0.25166254943541461, -0.10212412903126872, -0.11494109237158928], [0.1538190166683838, -0.19636071211303363, -0.16955231982502666], [0.69624421451105989, -0.21575476077610581, -0.22195121116483674], [-0.67316493619822171, -0.2515486267500906, 0.48832090783945642], [-0.60011861521328635, 0.22897217494908029, 0.20400673019934595], [0.029615125302479721, 0.10372152966977911, -0.23919109178136511], [0.16925716226788123, -0.18703765803400574, 0.013022157605494675], [0.1919803411529804, -0.32143163003060055, -0.11818935970200066], [0.80761667938359993, -0.33839244383018496, 0.039549307829734572], [0.24669667496849926, 0.30268177307271782, 0.21964458163057088], [-0.14219315348518174, -0.1147780335070106, 0.32706428493072132], [0.30427505815686307, -0.24247142530193411, 0.350945084796704], [-0.31319460193132892, 0.18184804315717218, 0.15593394728341345], [-0.6898315948318976, 0.15504602360039726, 0.29863974148792216], [-0.082896141260148104, 0.18651145103423833, -0.044376894020555085], [-0.065412112288315574, 0.19446473254155391, 0.25684436618993889], [0.74747044101940729, -1.0212095046332004, 0.33152483608646377], [0.3194605228960728, 0.091923324270713175, 0.3323621136813108], [-0.28265940510607457, -0.29937990221663746, 0.38549704422299402], [-0.36478754627854076, -0.14371799448150677, -0.13593364762469248], [0.94470188168709901, -0.48391730610634592, -0.701298324844149], [0.24382869144366329, 0.12965597232536735, 0.12848807536103687], [0.25456653136274154, -0.29157794873701731, -0.1567087371918561], [0.32956189577408246, -0.19924632055817562, -0.16975463655496348], [-0.16003760303725742, 0.16834508896389047, -0.02895200248099724], [0.79191574248070296, -0.94258942667724954, -0.33557168526432851], [0.20093565667960181, -0.13637901222029258, -0.31788160371143598], [0.062872216044461984, 0.2313758213174964, -0.080066186818751758], [0.16813372423064626, 0.0074972820959486042, 0.027557136131471489]]];
96104
const bias = [[0.30011741283138643, -0.029751221601027597, 0.27707089984418304, 0.09437747263089169, -0.073328190572502505, -0.009339555268726818, 0.069540321946648831, 0.030114067358476229, 0.1926469413394869, 0.023144226681430048, -0.070082924717630057, 0.24099343057814002, -0.25759272493282814, -0.3911290402719596, -0.24328056130217912, -0.13877438337256462, -0.3840609961969641, 0.20571932163518283, 0.12990256746838591, -0.32332067950525173, -0.11080976020947265, -0.1415205938964631, -0.24600210345938872, 0.26790419688164757, -0.29106359813082855, 0.2495076134848766, 0.054676119964720049, 0.25255465627456269, 0.22982296359481452, 0.29374548533376282, -0.026746489455462041, -0.044633956332849833, 0.015346134348148879, -0.1428540988439016, -0.059281796245451462, 0.030997844617798281, -0.48568377250041256, -0.2100376289112072, -0.044215767340361145, 0.018240118057367673, -0.1466920807505617, 0.39966256678781159, 0.051904810189690342, -0.28673053302698287, 0.19195282255033613, 0.074687451363289303, -0.35766933287571323, -0.05320421333257852, 0.11937922437695309, 0.27906785198501721], [0.37658323831187673, 0.45913319499996535, -0.63671974160497524]];
97105
98106
// Estimator:
99-
var brain = new Brain(layers, weights, bias);
107+
var brain = new Brain('relu', 'softmax', layers, weights, bias);
100108
101109
if (typeof process !== 'undefined' && typeof process.argv !== 'undefined') {
102110
if (process.argv.length - 2 === 4) {
103-
const features = process.argv.slice(2);
111+
var features = process.argv.slice(2);
104112
var prediction = brain.predict(features);
105113
console.log(prediction);
106114
}
@@ -119,7 +127,7 @@
119127
// Web Worker (brain.js):
120128
var onmessage = function(e) {
121129
if (e.data.length === 4) {
122-
const features = e.data;
130+
var features = e.data;
123131
var prediction = brain.predict(features);
124132
postMessage(prediction);
125133
}

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/activation_fn.identity.txt‎

Lines changed: 0 additions & 4 deletions
This file was deleted.

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/activation_fn.logistic.txt‎

Lines changed: 0 additions & 7 deletions
This file was deleted.

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/activation_fn.relu.txt‎

Lines changed: 0 additions & 7 deletions
This file was deleted.

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/activation_fn.tanh.txt‎

Lines changed: 0 additions & 7 deletions
This file was deleted.

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/class.txt‎

Lines changed: 48 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1,27 +1,65 @@
1-
// Array.prototype.fill polyfill:
2-
[].fill||(Array.prototype.fill=function(a){{for(var b=Object(this),c=parseInt(b.length,10),d=arguments[1],e=parseInt(d,10)||0,f=0>e?Math.max(c+e,0):Math.min(e,c),g=arguments[2],h=void 0===g?c:parseInt(g)||0,i=0>h?Math.max(c+h,0):Math.min(h,c);i>f;f++)b[f]=a;return b}});
1+
var {class_name} = function(hidden, output, layers, weights, bias) {{
32

4-
var {class_name} = function(layers, weights, bias) {{
3+
this.hidden = hidden.toUpperCase();
4+
this.output = output.toUpperCase();
5+
this.layers = layers;
6+
this.weights = weights;
7+
this.bias = bias;
58

6-
{activation_function}
7-
8-
{output_function}
9+
var compute = function(activation, v) {{
10+
switch (activation) {{
11+
case 'LOGISTIC':
12+
for (var i = 0, l = v.length; i < l; i++) {{
13+
v[i] = 1. / (1. + Math.exp(-v[i]));
14+
}}
15+
break;
16+
case 'RELU':
17+
for (var i = 0, l = v.length; i < l; i++) {{
18+
v[i] = Math.max(0, v[i]);
19+
}}
20+
break;
21+
case 'TANH':
22+
for (var i = 0, l = v.length; i < l; i++) {{
23+
v[i] = Math.tanh(v[i]);
24+
}}
25+
break;
26+
case 'SOFTMAX':
27+
var max = Number.NEGATIVE_INFINITY;
28+
for (var i = 0, l = v.length; i < l; i++) {{
29+
if (v[i] > max) {{
30+
max = v[i];
31+
}}
32+
}}
33+
for (var i = 0, l = v.length; i < l; i++) {{
34+
v[i] = Math.exp(v[i] - max);
35+
}}
36+
var sum = 0.0;
37+
for (var i = 0, l = v.length; i < l; i++) {{
38+
sum += v[i];
39+
}}
40+
for (var i = 0, l = v.length; i < l; i++) {{
41+
v[i] /= sum;
42+
}}
43+
break;
44+
}}
45+
return v;
46+
}};
947

1048
{method}
1149

1250
}};
1351

14-
// Model data:
52+
// Parameters:
1553
const {layers}
1654
const {weights}
1755
const {bias}
1856

1957
// Estimator:
20-
var brain = new {class_name}(layers, weights, bias);
58+
var brain = new {class_name}('{hidden_activation}', '{output_activation}', layers, weights, bias);
2159

2260
if (typeof process !== 'undefined' && typeof process.argv !== 'undefined') {{
2361
if (process.argv.length - 2 === {n_features}) {{
24-
const features = process.argv.slice(2);
62+
var features = process.argv.slice(2);
2563
var prediction = brain.{method_name}(features);
2664
console.log(prediction);
2765
}}
@@ -40,7 +78,7 @@ if (typeof process !== 'undefined' && typeof process.argv !== 'undefined') {{
4078
// Web Worker ({file_name}):
4179
var onmessage = function(e) {{
4280
if (e.data.length === {n_features}) {{
43-
const features = e.data;
81+
var features = e.data;
4482
var prediction = brain.{method_name}(features);
4583
postMessage(prediction);
4684
}}

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/method.binary.txt‎

Lines changed: 5 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -1,19 +1,18 @@
11
this.{method_name} = function(neurons) {{
2-
if (neurons.length != {n_features}) return -1;
3-
var network = [neurons].concat(layers);
2+
var network = [neurons].concat(this.layers);
43

54
for (var i = 0; i < network.length - 1; i++) {{
65
for (var j = 0; j < network[i + 1].length; j++) {{
76
for (var l = 0; l < network[i].length; l++) {{
8-
network[i + 1][j] += network[i][l] * weights[i][l][j];
7+
network[i + 1][j] += network[i][l] * this.weights[i][l][j];
98
}}
10-
network[i + 1][j] += bias[i][j];
9+
network[i + 1][j] += this.bias[i][j];
1110
}}
1211
if ((i + 1) < (network.length - 1)) {{
13-
network[i + 1] = compAct(network[i + 1]);
12+
network[i + 1] = compute(this.hidden, network[i + 1]);
1413
}}
1514
}}
16-
network[network.length - 1] = compOut(network[network.length - 1]);
15+
network[network.length - 1] = compute(this.output, network[network.length - 1]);
1716

1817
if (network[network.length - 1][0] > .5) {{
1918
return 1;

‎sklearn_porter/estimator/classifier/MLPClassifier/templates/js/method.multi.txt‎

Lines changed: 8 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -1,27 +1,22 @@
11
this.{method_name} = function(neurons) {{
2-
if (neurons.length != {n_features}) return -1;
3-
var network = [neurons].concat(layers);
2+
var network = [neurons].concat(this.layers);
43

54
for (var i = 0; i < network.length - 1; i++) {{
65
for (var j = 0; j < network[i + 1].length; j++) {{
76
for (var l = 0; l < network[i].length; l++) {{
8-
network[i + 1][j] += network[i][l] * weights[i][l][j];
7+
network[i + 1][j] += network[i][l] * this.weights[i][l][j];
98
}}
10-
network[i + 1][j] += bias[i][j];
9+
network[i + 1][j] += this.bias[i][j];
1110
}}
1211
if ((i + 1) < (network.length - 1)) {{
13-
network[i + 1] = compAct(network[i + 1]);
12+
network[i + 1] = compute(this.hidden, network[i + 1]);
1413
}}
1514
}}
16-
network[network.length - 1] = compOut(network[network.length - 1]);
15+
network[network.length - 1] = compute(this.output, network[network.length - 1]);
1716

18-
var classIdx = -1,
19-
classVal = Number.NEGATIVE_INFINITY;
17+
var index = 0;
2018
for (var i = 0, l = network[network.length - 1].length; i < l; i++) {{
21-
if (network[network.length - 1][i] > classVal) {{
22-
classVal = network[network.length - 1][i];
23-
classIdx = i;
24-
}}
19+
index = network[network.length - 1][i] > network[network.length - 1][index] ? i : index;
2520
}}
26-
return classIdx;
21+
return index;
2722
}};

0 commit comments

Comments
 (0)