blob: 40758360afb74ff8b93b5ae0b7a6b155259b500a [file] [log] [blame]
/*
* Copyright (C) 2015-2017 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY APPLE INC. AND ITS CONTRIBUTORS ``AS IS''
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL APPLE INC. OR ITS CONTRIBUTORS
* BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
* THE POSSIBILITY OF SUCH DAMAGE.
*/
Pseudo =
{
initialRandomSeed: 49734321,
randomSeed: 49734321,
resetRandomSeed: function()
{
Pseudo.randomSeed = Pseudo.initialRandomSeed;
},
random: function()
{
var randomSeed = Pseudo.randomSeed;
randomSeed = ((randomSeed + 0x7ed55d16) + (randomSeed << 12)) & 0xffffffff;
randomSeed = ((randomSeed ^ 0xc761c23c) ^ (randomSeed >>> 19)) & 0xffffffff;
randomSeed = ((randomSeed + 0x165667b1) + (randomSeed << 5)) & 0xffffffff;
randomSeed = ((randomSeed + 0xd3a2646c) ^ (randomSeed << 9)) & 0xffffffff;
randomSeed = ((randomSeed + 0xfd7046c5) + (randomSeed << 3)) & 0xffffffff;
randomSeed = ((randomSeed ^ 0xb55a4f09) ^ (randomSeed >>> 16)) & 0xffffffff;
Pseudo.randomSeed = randomSeed;
return (randomSeed & 0xfffffff) / 0x10000000;
}
};
Statistics =
{
sampleMean: function(numberOfSamples, sum)
{
if (numberOfSamples < 1)
return 0;
return sum / numberOfSamples;
},
// With sum and sum of squares, we can compute the sample standard deviation in O(1).
// See https://rniwa.com/2012-11-10/sample-standard-deviation-in-terms-of-sum-and-square-sum-of-samples/
unbiasedSampleStandardDeviation: function(numberOfSamples, sum, squareSum)
{
if (numberOfSamples < 2)
return 0;
return Math.sqrt((squareSum - sum * sum / numberOfSamples) / (numberOfSamples - 1));
},
geometricMean: function(values)
{
if (!values.length)
return 0;
var roots = values.map(function(value) { return Math.pow(value, 1 / values.length); })
return roots.reduce(function(a, b) { return a * b; });
},
// Cumulative distribution function
cdf: function(value, mean, standardDeviation)
{
return 0.5 * (1 + Statistics.erf((value - mean) / (Math.sqrt(2 * standardDeviation * standardDeviation))));
},
// Approximation of Gauss error function, Abramowitz and Stegun 7.1.26
erf: function(value)
{
var sign = (value >= 0) ? 1 : -1;
value = Math.abs(value);
var a1 = 0.254829592;
var a2 = -0.284496736;
var a3 = 1.421413741;
var a4 = -1.453152027;
var a5 = 1.061405429;
var p = 0.3275911;
var t = 1.0 / (1.0 + p * value);
var y = 1.0 - (((((a5 * t + a4) * t) + a3) * t + a2) * t + a1) * t * Math.exp(-value * value);
return sign * y;
},
largestDeviationPercentage: function(low, mean, high)
{
return Math.max(Math.abs(low / mean - 1), (high / mean - 1));
}
};
Experiment = Utilities.createClass(
function(includeConcern)
{
if (includeConcern)
this._maxHeap = Heap.createMaxHeap(Experiment.defaults.CONCERN_SIZE);
this.reset();
}, {
reset: function()
{
this._sum = 0;
this._squareSum = 0;
this._numberOfSamples = 0;
if (this._maxHeap)
this._maxHeap.init();
},
get sampleCount()
{
return this._numberOfSamples;
},
sample: function(value)
{
this._sum += value;
this._squareSum += value * value;
if (this._maxHeap)
this._maxHeap.push(value);
++this._numberOfSamples;
},
mean: function()
{
return Statistics.sampleMean(this._numberOfSamples, this._sum);
},
standardDeviation: function()
{
return Statistics.unbiasedSampleStandardDeviation(this._numberOfSamples, this._sum, this._squareSum);
},
cdf: function(value)
{
return Statistics.cdf(value, this.mean(), this.standardDeviation());
},
percentage: function()
{
var mean = this.mean();
return mean ? this.standardDeviation() * 100 / mean : 0;
},
concern: function(percentage)
{
if (!this._maxHeap)
return this.mean();
var size = Math.ceil(this._numberOfSamples * percentage / 100);
var values = this._maxHeap.values(size);
return values.length ? values.reduce(function(a, b) { return a + b; }) / values.length : 0;
},
score: function(percentage)
{
return Statistics.geometricMean([this.mean(), Math.max(this.concern(percentage), 1)]);
}
});
Experiment.defaults =
{
CONCERN: 5,
CONCERN_SIZE: 100,
};
Regression = Utilities.createClass(
function(samples, getComplexity, getFrameLength, startIndex, endIndex, options)
{
var desiredFrameLength = options.desiredFrameLength || 1000/60;
var bestProfile;
if (!options.preferredProfile || options.preferredProfile == Strings.json.profiles.slope) {
var slope = this._calculateRegression(samples, getComplexity, getFrameLength, startIndex, endIndex, {
shouldClip: true,
s1: desiredFrameLength,
t1: 0
});
if (!bestProfile || slope.error < bestProfile.error) {
bestProfile = slope;
this.profile = Strings.json.profiles.slope;
}
}
if (!options.preferredProfile || options.preferredProfile == Strings.json.profiles.flat) {
var flat = this._calculateRegression(samples, getComplexity, getFrameLength, startIndex, endIndex, {
shouldClip: true,
s1: desiredFrameLength,
t1: 0,
t2: 0
});
if (!bestProfile || flat.error < bestProfile.error) {
bestProfile = flat;
this.profile = Strings.json.profiles.flat;
}
}
this.startIndex = Math.min(startIndex, endIndex);
this.endIndex = Math.max(startIndex, endIndex);
this.complexity = bestProfile.complexity;
this.s1 = bestProfile.s1;
this.t1 = bestProfile.t1;
this.s2 = bestProfile.s2;
this.t2 = bestProfile.t2;
this.stdev1 = bestProfile.stdev1;
this.stdev2 = bestProfile.stdev2;
this.n1 = bestProfile.n1;
this.n2 = bestProfile.n2;
this.error = bestProfile.error;
}, {
valueAt: function(complexity)
{
if (this.n1 == 1 || complexity > this.complexity)
return this.s2 + this.t2 * complexity;
return this.s1 + this.t1 * complexity;
},
// A generic two-segment piecewise regression calculator. Based on Kundu/Ubhaya
//
// Minimize sum of (y - y')^2
// where y = s1 + t1*x
// y = s2 + t2*x
// y' = s1 + t1*x' = s2 + t2*x' if x_0 <= x' <= x_n
//
// Allows for fixing s1, t1, s2, t2
//
// x is assumed to be complexity, y is frame length. Can be used for pure complexity-FPS
// analysis or for ramp controllers since complexity monotonically decreases with time.
_calculateRegression: function(samples, getComplexity, getFrameLength, startIndex, endIndex, options)
{
if (startIndex == endIndex) {
// Only one sample point; we can't calculate any regression.
var x = getComplexity(samples, startIndex);
return {
complexity: x,
s1: x,
t1: 0,
s2: x,
t2: 0,
error1: 0,
error2: 0
};
}
// x is expected to increase in complexity
var iterationDirection = endIndex > startIndex ? 1 : -1;
var lowComplexity = getComplexity(samples, startIndex);
var highComplexity = getComplexity(samples, endIndex);
var a1 = 0, b1 = 0, c1 = 0, d1 = 0, h1 = 0, k1 = 0;
var a2 = 0, b2 = 0, c2 = 0, d2 = 0, h2 = 0, k2 = 0;
// Iterate from low to high complexity
for (var i = startIndex; iterationDirection * (endIndex - i) > -1; i += iterationDirection) {
var x = getComplexity(samples, i);
var y = getFrameLength(samples, i);
a2 += 1;
b2 += x;
c2 += x * x;
d2 += y;
h2 += y * x;
k2 += y * y;
}
var s1_best, t1_best, s2_best, t2_best, n1_best, n2_best, error1_best, error2_best, x_best, x_prime;
function setBest(s1, t1, error1, s2, t2, error2, splitIndex, x_prime, x)
{
s1_best = s1;
t1_best = t1;
error1_best = error1;
s2_best = s2;
t2_best = t2;
error2_best = error2;
// Number of samples included in the first segment, inclusive of splitIndex
n1_best = iterationDirection * (splitIndex - startIndex) + 1;
// Number of samples included in the second segment
n2_best = iterationDirection * (endIndex - splitIndex);
if (!options.shouldClip || (x_prime >= lowComplexity && x_prime <= highComplexity))
x_best = x_prime;
else {
// Discontinuous piecewise regression
x_best = x;
}
}
// Iterate from startIndex to endIndex - 1, inclusive
for (var i = startIndex; iterationDirection * (endIndex - i) > 0; i += iterationDirection) {
var x = getComplexity(samples, i);
var y = getFrameLength(samples, i);
var xx = x * x;
var yx = y * x;
var yy = y * y;
// a1, b1, etc. is sum from startIndex to i, inclusive
a1 += 1;
b1 += x;
c1 += xx;
d1 += y;
h1 += yx;
k1 += yy;
// a2, b2, etc. is sum from i+1 to endIndex, inclusive
a2 -= 1;
b2 -= x;
c2 -= xx;
d2 -= y;
h2 -= yx;
k2 -= yy;
var A = c1*d1 - b1*h1;
var B = a1*h1 - b1*d1;
var C = a1*c1 - b1*b1;
var D = c2*d2 - b2*h2;
var E = a2*h2 - b2*d2;
var F = a2*c2 - b2*b2;
var s1 = options.s1 !== undefined ? options.s1 : (A / C);
var t1 = options.t1 !== undefined ? options.t1 : (B / C);
var s2 = options.s2 !== undefined ? options.s2 : (D / F);
var t2 = options.t2 !== undefined ? options.t2 : (E / F);
// Assumes that the two segments meet
var x_prime = (s1 - s2) / (t2 - t1);
var error1 = (k1 + a1*s1*s1 + c1*t1*t1 - 2*d1*s1 - 2*h1*t1 + 2*b1*s1*t1) || Number.MAX_VALUE;
var error2 = (k2 + a2*s2*s2 + c2*t2*t2 - 2*d2*s2 - 2*h2*t2 + 2*b2*s2*t2) || Number.MAX_VALUE;
if (i == startIndex) {
setBest(s1, t1, error1, s2, t2, error2, i, x_prime, x);
continue;
}
if (C == 0 || F == 0)
continue;
// Projected point is not between this and the next sample
if (x_prime > getComplexity(samples, i + iterationDirection) || x_prime < x) {
// Calculate lambda, which divides the weight of this sample between the two lines
// These values remove the influence of this sample
var I = c1 - 2*b1*x + a1*xx;
var H = C - I;
var G = A + B*x - C*y;
var J = D + E*x - F*y;
var K = c2 - 2*b2*x + a2*xx;
var lambda = (G*F + G*K - H*J) / (I*J + G*K);
if (lambda > 0 && lambda < 1) {
var lambda1 = 1 - lambda;
s1 = options.s1 !== undefined ? options.s1 : ((A - lambda1*(-h1*x + d1*xx + c1*y - b1*yx)) / (C - lambda1*I));
t1 = options.t1 !== undefined ? options.t1 : ((B - lambda1*(h1 - d1*x - b1*y + a1*yx)) / (C - lambda1*I));
s2 = options.s2 !== undefined ? options.s2 : ((D + lambda1*(-h2*x + d2*xx + c2*y - b2*yx)) / (F + lambda1*K));
t2 = options.t2 !== undefined ? options.t2 : ((E + lambda1*(h2 - d2*x - b2*y + a2*yx)) / (F + lambda1*K));
x_prime = (s1 - s2) / (t2 - t1);
error1 = ((k1 + a1*s1*s1 + c1*t1*t1 - 2*d1*s1 - 2*h1*t1 + 2*b1*s1*t1) - lambda1 * Math.pow(y - (s1 + t1*x), 2)) || Number.MAX_VALUE;
error2 = ((k2 + a2*s2*s2 + c2*t2*t2 - 2*d2*s2 - 2*h2*t2 + 2*b2*s2*t2) + lambda1 * Math.pow(y - (s2 + t2*x), 2)) || Number.MAX_VALUE;
} else if (t1 != t2)
continue;
}
if (error1 + error2 < error1_best + error2_best)
setBest(s1, t1, error1, s2, t2, error2, i, x_prime, x);
}
return {
complexity: x_best,
s1: s1_best,
t1: t1_best,
stdev1: Math.sqrt(error1_best / n1_best),
s2: s2_best,
t2: t2_best,
stdev2: Math.sqrt(error2_best / n2_best),
error: error1_best + error2_best,
n1: n1_best,
n2: n2_best
};
}
});
Utilities.extendObject(Regression, {
bootstrap: function(samples, iterationCount, processResample, confidencePercentage)
{
var sampleLength = samples.length;
var resample = new Array(sampleLength);
var bootstrapEstimator = new Experiment;
var bootstrapData = new Array(iterationCount);
Pseudo.resetRandomSeed();
for (var i = 0; i < iterationCount; ++i) {
for (var j = 0; j < sampleLength; ++j)
resample[j] = samples[Math.floor(Pseudo.random() * sampleLength)];
var resampleResult = processResample(resample);
bootstrapEstimator.sample(resampleResult);
bootstrapData[i] = resampleResult;
}
bootstrapData.sort(function(a, b) { return a - b; });
return {
confidenceLow: bootstrapData[Math.round((iterationCount - 1) * (1 - confidencePercentage) / 2)],
confidenceHigh: bootstrapData[Math.round((iterationCount - 1) * (1 + confidencePercentage) / 2)],
median: bootstrapData[Math.round(iterationCount / 2)],
mean: bootstrapEstimator.mean(),
data: bootstrapData,
confidencePercentage: confidencePercentage
};
}
});