starry / backend /libs /three /math /Interpolant.js
k-l-lambda's picture
feat: add Python ML services (CPU mode) with model download
2b7aae2
/**
* Abstract base class of interpolants over parametric samples.
*
* The parameter domain is one dimensional, typically the time or a path
* along a curve defined by the data.
*
* The sample values can have any dimensionality and derived classes may
* apply special interpretations to the data.
*
* This class provides the interval seek in a Template Method, deferring
* the actual interpolation to derived classes.
*
* Time complexity is O(1) for linear access crossing at most two points
* and O(log N) for random access, where N is the number of positions.
*
* References:
*
* http://www.oodesign.com/template-method-pattern.html
*
*/
class Interpolant {
constructor(parameterPositions, sampleValues, sampleSize, resultBuffer) {
this.parameterPositions = parameterPositions;
this._cachedIndex = 0;
this.resultBuffer = resultBuffer !== undefined ? resultBuffer : new sampleValues.constructor(sampleSize);
this.sampleValues = sampleValues;
this.valueSize = sampleSize;
this.settings = null;
this.DefaultSettings_ = {};
}
evaluate(t) {
const pp = this.parameterPositions;
let i1 = this._cachedIndex,
t1 = pp[i1],
t0 = pp[i1 - 1];
validate_interval: {
seek: {
let right;
linear_scan: {
//- See http://jsperf.com/comparison-to-undefined/3
//- slower code:
//-
//- if ( t >= t1 || t1 === undefined ) {
forward_scan: if (!(t < t1)) {
for (let giveUpAt = i1 + 2; ; ) {
if (t1 === undefined) {
if (t < t0) break forward_scan;
// after end
i1 = pp.length;
this._cachedIndex = i1;
return this.afterEnd_(i1 - 1, t, t0);
}
if (i1 === giveUpAt) break; // this loop
t0 = t1;
t1 = pp[++i1];
if (t < t1) {
// we have arrived at the sought interval
break seek;
}
}
// prepare binary search on the right side of the index
right = pp.length;
break linear_scan;
}
//- slower code:
//- if ( t < t0 || t0 === undefined ) {
if (!(t >= t0)) {
// looping?
const t1global = pp[1];
if (t < t1global) {
i1 = 2; // + 1, using the scan for the details
t0 = t1global;
}
// linear reverse scan
for (let giveUpAt = i1 - 2; ; ) {
if (t0 === undefined) {
// before start
this._cachedIndex = 0;
return this.beforeStart_(0, t, t1);
}
if (i1 === giveUpAt) break; // this loop
t1 = t0;
t0 = pp[--i1 - 1];
if (t >= t0) {
// we have arrived at the sought interval
break seek;
}
}
// prepare binary search on the left side of the index
right = i1;
i1 = 0;
break linear_scan;
}
// the interval is valid
break validate_interval;
} // linear scan
// binary search
while (i1 < right) {
const mid = (i1 + right) >>> 1;
if (t < pp[mid]) {
right = mid;
} else {
i1 = mid + 1;
}
}
t1 = pp[i1];
t0 = pp[i1 - 1];
// check boundary cases, again
if (t0 === undefined) {
this._cachedIndex = 0;
return this.beforeStart_(0, t, t1);
}
if (t1 === undefined) {
i1 = pp.length;
this._cachedIndex = i1;
return this.afterEnd_(i1 - 1, t0, t);
}
} // seek
this._cachedIndex = i1;
this.intervalChanged_(i1, t0, t1);
} // validate_interval
return this.interpolate_(i1, t0, t, t1);
}
getSettings_() {
return this.settings || this.DefaultSettings_;
}
copySampleValue_(index) {
// copies a sample value to the result buffer
const result = this.resultBuffer,
values = this.sampleValues,
stride = this.valueSize,
offset = index * stride;
for (let i = 0; i !== stride; ++i) {
result[i] = values[offset + i];
}
return result;
}
// Template methods for derived classes:
interpolate_(/* i1, t0, t, t1 */) {
throw new Error('call to abstract method');
// implementations shall return this.resultBuffer
}
intervalChanged_(/* i1, t0, t1 */) {
// empty
}
}
// ALIAS DEFINITIONS
Interpolant.prototype.beforeStart_ = Interpolant.prototype.copySampleValue_;
Interpolant.prototype.afterEnd_ = Interpolant.prototype.copySampleValue_;
export { Interpolant };