StackGenVis: Alignment of Data, Algorithms, and Models for Stacking Ensemble Learning Using Performance Metrics
https://doi.org/10.1109/TVCG.2020.3030352
You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
490 lines
16 KiB
490 lines
16 KiB
/*
|
|
* quantize.js Copyright 2008 Nick Rabinowitz
|
|
* Ported to node.js by Olivier Lesnicki
|
|
* Licensed under the MIT license: http://www.opensource.org/licenses/mit-license.php
|
|
*/
|
|
|
|
// fill out a couple protovis dependencies
|
|
/*
|
|
* Block below copied from Protovis: http://mbostock.github.com/protovis/
|
|
* Copyright 2010 Stanford Visualization Group
|
|
* Licensed under the BSD License: http://www.opensource.org/licenses/bsd-license.php
|
|
*/
|
|
if (!pv) {
|
|
var pv = {
|
|
map: function(array, f) {
|
|
var o = {};
|
|
return f ? array.map(function(d, i) {
|
|
o.index = i;
|
|
return f.call(o, d);
|
|
}) : array.slice();
|
|
},
|
|
naturalOrder: function(a, b) {
|
|
return (a < b) ? -1 : ((a > b) ? 1 : 0);
|
|
},
|
|
sum: function(array, f) {
|
|
var o = {};
|
|
return array.reduce(f ? function(p, d, i) {
|
|
o.index = i;
|
|
return p + f.call(o, d);
|
|
} : function(p, d) {
|
|
return p + d;
|
|
}, 0);
|
|
},
|
|
max: function(array, f) {
|
|
return Math.max.apply(null, f ? pv.map(array, f) : array);
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Basic Javascript port of the MMCQ (modified median cut quantization)
|
|
* algorithm from the Leptonica library (http://www.leptonica.com/).
|
|
* Returns a color map you can use to map original pixels to the reduced
|
|
* palette. Still a work in progress.
|
|
*
|
|
* @author Nick Rabinowitz
|
|
* @example
|
|
|
|
// array of pixels as [R,G,B] arrays
|
|
var myPixels = [[190,197,190], [202,204,200], [207,214,210], [211,214,211], [205,207,207]
|
|
// etc
|
|
];
|
|
var maxColors = 4;
|
|
|
|
var cmap = MMCQ.quantize(myPixels, maxColors);
|
|
var newPalette = cmap.palette();
|
|
var newPixels = myPixels.map(function(p) {
|
|
return cmap.map(p);
|
|
});
|
|
|
|
*/
|
|
var MMCQ = (function() {
|
|
// private constants
|
|
var sigbits = 5,
|
|
rshift = 8 - sigbits,
|
|
maxIterations = 1000,
|
|
fractByPopulations = 0.75;
|
|
|
|
// get reduced-space color index for a pixel
|
|
|
|
function getColorIndex(r, g, b) {
|
|
return (r << (2 * sigbits)) + (g << sigbits) + b;
|
|
}
|
|
|
|
// Simple priority queue
|
|
|
|
function PQueue(comparator) {
|
|
var contents = [],
|
|
sorted = false;
|
|
|
|
function sort() {
|
|
contents.sort(comparator);
|
|
sorted = true;
|
|
}
|
|
|
|
return {
|
|
push: function(o) {
|
|
contents.push(o);
|
|
sorted = false;
|
|
},
|
|
peek: function(index) {
|
|
if (!sorted) sort();
|
|
if (index === undefined) index = contents.length - 1;
|
|
return contents[index];
|
|
},
|
|
pop: function() {
|
|
if (!sorted) sort();
|
|
return contents.pop();
|
|
},
|
|
size: function() {
|
|
return contents.length;
|
|
},
|
|
map: function(f) {
|
|
return contents.map(f);
|
|
},
|
|
debug: function() {
|
|
if (!sorted) sort();
|
|
return contents;
|
|
}
|
|
};
|
|
}
|
|
|
|
// 3d color space box
|
|
|
|
function VBox(r1, r2, g1, g2, b1, b2, histo) {
|
|
var vbox = this;
|
|
vbox.r1 = r1;
|
|
vbox.r2 = r2;
|
|
vbox.g1 = g1;
|
|
vbox.g2 = g2;
|
|
vbox.b1 = b1;
|
|
vbox.b2 = b2;
|
|
vbox.histo = histo;
|
|
}
|
|
VBox.prototype = {
|
|
volume: function(force) {
|
|
var vbox = this;
|
|
if (!vbox._volume || force) {
|
|
vbox._volume = ((vbox.r2 - vbox.r1 + 1) * (vbox.g2 - vbox.g1 + 1) * (vbox.b2 - vbox.b1 + 1));
|
|
}
|
|
return vbox._volume;
|
|
},
|
|
count: function(force) {
|
|
var vbox = this,
|
|
histo = vbox.histo;
|
|
if (!vbox._count_set || force) {
|
|
var npix = 0,
|
|
i, j, k, index;
|
|
for (i = vbox.r1; i <= vbox.r2; i++) {
|
|
for (j = vbox.g1; j <= vbox.g2; j++) {
|
|
for (k = vbox.b1; k <= vbox.b2; k++) {
|
|
index = getColorIndex(i, j, k);
|
|
npix += (histo[index] || 0);
|
|
}
|
|
}
|
|
}
|
|
vbox._count = npix;
|
|
vbox._count_set = true;
|
|
}
|
|
return vbox._count;
|
|
},
|
|
copy: function() {
|
|
var vbox = this;
|
|
return new VBox(vbox.r1, vbox.r2, vbox.g1, vbox.g2, vbox.b1, vbox.b2, vbox.histo);
|
|
},
|
|
avg: function(force) {
|
|
var vbox = this,
|
|
histo = vbox.histo;
|
|
if (!vbox._avg || force) {
|
|
var ntot = 0,
|
|
mult = 1 << (8 - sigbits),
|
|
rsum = 0,
|
|
gsum = 0,
|
|
bsum = 0,
|
|
hval,
|
|
i, j, k, histoindex;
|
|
for (i = vbox.r1; i <= vbox.r2; i++) {
|
|
for (j = vbox.g1; j <= vbox.g2; j++) {
|
|
for (k = vbox.b1; k <= vbox.b2; k++) {
|
|
histoindex = getColorIndex(i, j, k);
|
|
hval = histo[histoindex] || 0;
|
|
ntot += hval;
|
|
rsum += (hval * (i + 0.5) * mult);
|
|
gsum += (hval * (j + 0.5) * mult);
|
|
bsum += (hval * (k + 0.5) * mult);
|
|
}
|
|
}
|
|
}
|
|
if (ntot) {
|
|
vbox._avg = [~~(rsum / ntot), ~~ (gsum / ntot), ~~ (bsum / ntot)];
|
|
} else {
|
|
//console.log('empty box');
|
|
vbox._avg = [~~(mult * (vbox.r1 + vbox.r2 + 1) / 2), ~~ (mult * (vbox.g1 + vbox.g2 + 1) / 2), ~~ (mult * (vbox.b1 + vbox.b2 + 1) / 2)];
|
|
}
|
|
}
|
|
return vbox._avg;
|
|
},
|
|
contains: function(pixel) {
|
|
var vbox = this,
|
|
rval = pixel[0] >> rshift;
|
|
gval = pixel[1] >> rshift;
|
|
bval = pixel[2] >> rshift;
|
|
return (rval >= vbox.r1 && rval <= vbox.r2 &&
|
|
gval >= vbox.g1 && gval <= vbox.g2 &&
|
|
bval >= vbox.b1 && bval <= vbox.b2);
|
|
}
|
|
};
|
|
|
|
// Color map
|
|
|
|
function CMap() {
|
|
this.vboxes = new PQueue(function(a, b) {
|
|
return pv.naturalOrder(
|
|
a.vbox.count() * a.vbox.volume(),
|
|
b.vbox.count() * b.vbox.volume()
|
|
)
|
|
});;
|
|
}
|
|
CMap.prototype = {
|
|
push: function(vbox) {
|
|
this.vboxes.push({
|
|
vbox: vbox,
|
|
color: vbox.avg()
|
|
});
|
|
},
|
|
palette: function() {
|
|
return this.vboxes.map(function(vb) {
|
|
return vb.color
|
|
});
|
|
},
|
|
size: function() {
|
|
return this.vboxes.size();
|
|
},
|
|
map: function(color) {
|
|
var vboxes = this.vboxes;
|
|
for (var i = 0; i < vboxes.size(); i++) {
|
|
if (vboxes.peek(i).vbox.contains(color)) {
|
|
return vboxes.peek(i).color;
|
|
}
|
|
}
|
|
return this.nearest(color);
|
|
},
|
|
nearest: function(color) {
|
|
var vboxes = this.vboxes,
|
|
d1, d2, pColor;
|
|
for (var i = 0; i < vboxes.size(); i++) {
|
|
d2 = Math.sqrt(
|
|
Math.pow(color[0] - vboxes.peek(i).color[0], 2) +
|
|
Math.pow(color[1] - vboxes.peek(i).color[1], 2) +
|
|
Math.pow(color[2] - vboxes.peek(i).color[2], 2)
|
|
);
|
|
if (d2 < d1 || d1 === undefined) {
|
|
d1 = d2;
|
|
pColor = vboxes.peek(i).color;
|
|
}
|
|
}
|
|
return pColor;
|
|
},
|
|
forcebw: function() {
|
|
// XXX: won't work yet
|
|
var vboxes = this.vboxes;
|
|
vboxes.sort(function(a, b) {
|
|
return pv.naturalOrder(pv.sum(a.color), pv.sum(b.color))
|
|
});
|
|
|
|
// force darkest color to black if everything < 5
|
|
var lowest = vboxes[0].color;
|
|
if (lowest[0] < 5 && lowest[1] < 5 && lowest[2] < 5)
|
|
vboxes[0].color = [0, 0, 0];
|
|
|
|
// force lightest color to white if everything > 251
|
|
var idx = vboxes.length - 1,
|
|
highest = vboxes[idx].color;
|
|
if (highest[0] > 251 && highest[1] > 251 && highest[2] > 251)
|
|
vboxes[idx].color = [255, 255, 255];
|
|
}
|
|
};
|
|
|
|
// histo (1-d array, giving the number of pixels in
|
|
// each quantized region of color space), or null on error
|
|
|
|
function getHisto(pixels) {
|
|
var histosize = 1 << (3 * sigbits),
|
|
histo = new Array(histosize),
|
|
index, rval, gval, bval;
|
|
pixels.forEach(function(pixel) {
|
|
rval = pixel[0] >> rshift;
|
|
gval = pixel[1] >> rshift;
|
|
bval = pixel[2] >> rshift;
|
|
index = getColorIndex(rval, gval, bval);
|
|
histo[index] = (histo[index] || 0) + 1;
|
|
});
|
|
return histo;
|
|
}
|
|
|
|
function vboxFromPixels(pixels, histo) {
|
|
var rmin = 1000000,
|
|
rmax = 0,
|
|
gmin = 1000000,
|
|
gmax = 0,
|
|
bmin = 1000000,
|
|
bmax = 0,
|
|
rval, gval, bval;
|
|
// find min/max
|
|
pixels.forEach(function(pixel) {
|
|
rval = pixel[0] >> rshift;
|
|
gval = pixel[1] >> rshift;
|
|
bval = pixel[2] >> rshift;
|
|
if (rval < rmin) rmin = rval;
|
|
else if (rval > rmax) rmax = rval;
|
|
if (gval < gmin) gmin = gval;
|
|
else if (gval > gmax) gmax = gval;
|
|
if (bval < bmin) bmin = bval;
|
|
else if (bval > bmax) bmax = bval;
|
|
});
|
|
return new VBox(rmin, rmax, gmin, gmax, bmin, bmax, histo);
|
|
}
|
|
|
|
function medianCutApply(histo, vbox) {
|
|
if (!vbox.count()) return;
|
|
|
|
var rw = vbox.r2 - vbox.r1 + 1,
|
|
gw = vbox.g2 - vbox.g1 + 1,
|
|
bw = vbox.b2 - vbox.b1 + 1,
|
|
maxw = pv.max([rw, gw, bw]);
|
|
// only one pixel, no split
|
|
if (vbox.count() == 1) {
|
|
return [vbox.copy()]
|
|
}
|
|
/* Find the partial sum arrays along the selected axis. */
|
|
var total = 0,
|
|
partialsum = [],
|
|
lookaheadsum = [],
|
|
i, j, k, sum, index;
|
|
if (maxw == rw) {
|
|
for (i = vbox.r1; i <= vbox.r2; i++) {
|
|
sum = 0;
|
|
for (j = vbox.g1; j <= vbox.g2; j++) {
|
|
for (k = vbox.b1; k <= vbox.b2; k++) {
|
|
index = getColorIndex(i, j, k);
|
|
sum += (histo[index] || 0);
|
|
}
|
|
}
|
|
total += sum;
|
|
partialsum[i] = total;
|
|
}
|
|
} else if (maxw == gw) {
|
|
for (i = vbox.g1; i <= vbox.g2; i++) {
|
|
sum = 0;
|
|
for (j = vbox.r1; j <= vbox.r2; j++) {
|
|
for (k = vbox.b1; k <= vbox.b2; k++) {
|
|
index = getColorIndex(j, i, k);
|
|
sum += (histo[index] || 0);
|
|
}
|
|
}
|
|
total += sum;
|
|
partialsum[i] = total;
|
|
}
|
|
} else { /* maxw == bw */
|
|
for (i = vbox.b1; i <= vbox.b2; i++) {
|
|
sum = 0;
|
|
for (j = vbox.r1; j <= vbox.r2; j++) {
|
|
for (k = vbox.g1; k <= vbox.g2; k++) {
|
|
index = getColorIndex(j, k, i);
|
|
sum += (histo[index] || 0);
|
|
}
|
|
}
|
|
total += sum;
|
|
partialsum[i] = total;
|
|
}
|
|
}
|
|
partialsum.forEach(function(d, i) {
|
|
lookaheadsum[i] = total - d
|
|
});
|
|
|
|
function doCut(color) {
|
|
var dim1 = color + '1',
|
|
dim2 = color + '2',
|
|
left, right, vbox1, vbox2, d2, count2 = 0;
|
|
for (i = vbox[dim1]; i <= vbox[dim2]; i++) {
|
|
if (partialsum[i] > total / 2) {
|
|
vbox1 = vbox.copy();
|
|
vbox2 = vbox.copy();
|
|
left = i - vbox[dim1];
|
|
right = vbox[dim2] - i;
|
|
if (left <= right)
|
|
d2 = Math.min(vbox[dim2] - 1, ~~ (i + right / 2));
|
|
else d2 = Math.max(vbox[dim1], ~~ (i - 1 - left / 2));
|
|
// avoid 0-count boxes
|
|
while (!partialsum[d2]) d2++;
|
|
count2 = lookaheadsum[d2];
|
|
while (!count2 && partialsum[d2 - 1]) count2 = lookaheadsum[--d2];
|
|
// set dimensions
|
|
vbox1[dim2] = d2;
|
|
vbox2[dim1] = vbox1[dim2] + 1;
|
|
// console.log('vbox counts:', vbox.count(), vbox1.count(), vbox2.count());
|
|
return [vbox1, vbox2];
|
|
}
|
|
}
|
|
|
|
}
|
|
// determine the cut planes
|
|
return maxw == rw ? doCut('r') :
|
|
maxw == gw ? doCut('g') :
|
|
doCut('b');
|
|
}
|
|
|
|
function quantize(pixels, maxcolors) {
|
|
// short-circuit
|
|
if (!pixels.length || maxcolors < 2 || maxcolors > 256) {
|
|
// console.log('wrong number of maxcolors');
|
|
return false;
|
|
}
|
|
|
|
// XXX: check color content and convert to grayscale if insufficient
|
|
|
|
var histo = getHisto(pixels),
|
|
histosize = 1 << (3 * sigbits);
|
|
|
|
// check that we aren't below maxcolors already
|
|
var nColors = 0;
|
|
histo.forEach(function() {
|
|
nColors++
|
|
});
|
|
if (nColors <= maxcolors) {
|
|
// XXX: generate the new colors from the histo and return
|
|
}
|
|
|
|
// get the beginning vbox from the colors
|
|
var vbox = vboxFromPixels(pixels, histo),
|
|
pq = new PQueue(function(a, b) {
|
|
return pv.naturalOrder(a.count(), b.count())
|
|
});
|
|
pq.push(vbox);
|
|
|
|
// inner function to do the iteration
|
|
|
|
function iter(lh, target) {
|
|
var ncolors = 1,
|
|
niters = 0,
|
|
vbox;
|
|
while (niters < maxIterations) {
|
|
vbox = lh.pop();
|
|
if (!vbox.count()) { /* just put it back */
|
|
lh.push(vbox);
|
|
niters++;
|
|
continue;
|
|
}
|
|
// do the cut
|
|
var vboxes = medianCutApply(histo, vbox),
|
|
vbox1 = vboxes[0],
|
|
vbox2 = vboxes[1];
|
|
|
|
if (!vbox1) {
|
|
// console.log("vbox1 not defined; shouldn't happen!");
|
|
return;
|
|
}
|
|
lh.push(vbox1);
|
|
if (vbox2) { /* vbox2 can be null */
|
|
lh.push(vbox2);
|
|
ncolors++;
|
|
}
|
|
if (ncolors >= target) return;
|
|
if (niters++ > maxIterations) {
|
|
// console.log("infinite loop; perhaps too few pixels!");
|
|
return;
|
|
}
|
|
}
|
|
}
|
|
|
|
// first set of colors, sorted by population
|
|
iter(pq, fractByPopulations * maxcolors);
|
|
// console.log(pq.size(), pq.debug().length, pq.debug().slice());
|
|
|
|
// Re-sort by the product of pixel occupancy times the size in color space.
|
|
var pq2 = new PQueue(function(a, b) {
|
|
return pv.naturalOrder(a.count() * a.volume(), b.count() * b.volume())
|
|
});
|
|
while (pq.size()) {
|
|
pq2.push(pq.pop());
|
|
}
|
|
|
|
// next set - generate the median cuts using the (npix * vol) sorting.
|
|
iter(pq2, maxcolors - pq2.size());
|
|
|
|
// calculate the actual colors
|
|
var cmap = new CMap();
|
|
while (pq2.size()) {
|
|
cmap.push(pq2.pop());
|
|
}
|
|
|
|
return cmap;
|
|
}
|
|
|
|
return {
|
|
quantize: quantize
|
|
}
|
|
})();
|
|
|
|
module.exports = MMCQ.quantize
|
|
|