This commit is contained in:
229
src/utils/sampling.ts
Normal file
229
src/utils/sampling.ts
Normal file
@@ -0,0 +1,229 @@
|
||||
/**
|
||||
* 数据抽样算法
|
||||
* 用于在保持视觉保真度的同时减少渲染点数
|
||||
*/
|
||||
|
||||
import type { WaveformPoint } from '../types'
|
||||
|
||||
/**
|
||||
* Largest Triangle Three Buckets (LTTB) 抽样算法
|
||||
*
|
||||
* 这是一种高效的降采样算法,能够在减少数据点的同时保持波形的视觉特征。
|
||||
* 算法通过计算三角形面积来选择最具代表性的点。
|
||||
*
|
||||
* 参考文献: Sveinn Steinarsson. 2013.
|
||||
* "Downsampling Time Series for Visual Representation"
|
||||
*
|
||||
* @param data 原始数据点数组
|
||||
* @param threshold 目标点数(必须 >= 3)
|
||||
* @returns 抽样后的数据点数组
|
||||
*
|
||||
* @example
|
||||
* const original = Array.from({ length: 10000 }, (_, i) => ({ x: i, y: Math.sin(i / 100) }))
|
||||
* const sampled = downsampleLTTB(original, 500) // 从 10000 点降至 500 点
|
||||
*/
|
||||
export function downsampleLTTB(data: WaveformPoint[], threshold: number): WaveformPoint[] {
|
||||
// 边界检查
|
||||
if (!Array.isArray(data) || data.length === 0) {
|
||||
return []
|
||||
}
|
||||
|
||||
const dataLength = data.length
|
||||
|
||||
// 如果数据点数少于或等于阈值,直接返回
|
||||
if (threshold >= dataLength || threshold <= 2) {
|
||||
return data
|
||||
}
|
||||
|
||||
// 确保阈值至少为 3
|
||||
const sampledLength = Math.max(3, Math.floor(threshold))
|
||||
const sampled: WaveformPoint[] = new Array(sampledLength)
|
||||
|
||||
// 始终保留第一个和最后一个点
|
||||
sampled[0] = data[0]!
|
||||
sampled[sampledLength - 1] = data[dataLength - 1]!
|
||||
|
||||
// 计算每个桶的大小(除了第一个和最后一个点)
|
||||
const bucketSize = (dataLength - 2) / (sampledLength - 2)
|
||||
|
||||
// 用于计算三角形面积的辅助变量
|
||||
let sampledIndex = 1
|
||||
|
||||
for (let i = 0; i < sampledLength - 2; i++) {
|
||||
// 当前桶的范围
|
||||
const avgRangeStart = Math.floor((i + 1) * bucketSize) + 1
|
||||
const avgRangeEnd = Math.floor((i + 2) * bucketSize) + 1
|
||||
const avgRangeLength = Math.min(avgRangeEnd, dataLength) - avgRangeStart
|
||||
|
||||
// 计算下一个桶的平均点(用于三角形计算)
|
||||
let avgX = 0
|
||||
let avgY = 0
|
||||
|
||||
for (let j = avgRangeStart; j < Math.min(avgRangeEnd, dataLength); j++) {
|
||||
const point = data[j]!
|
||||
avgX += point.x
|
||||
avgY += point.y
|
||||
}
|
||||
|
||||
if (avgRangeLength > 0) {
|
||||
avgX /= avgRangeLength
|
||||
avgY /= avgRangeLength
|
||||
}
|
||||
|
||||
// 当前桶的范围
|
||||
const rangeStart = Math.floor(i * bucketSize) + 1
|
||||
const rangeEnd = Math.floor((i + 1) * bucketSize) + 1
|
||||
|
||||
// 上一个选中的点
|
||||
const prevPoint = sampled[sampledIndex - 1]!
|
||||
|
||||
// 在当前桶中找到形成最大三角形面积的点
|
||||
let maxArea = -1
|
||||
let maxAreaIndex = rangeStart
|
||||
|
||||
for (let j = rangeStart; j < Math.min(rangeEnd, dataLength); j++) {
|
||||
const point = data[j]!
|
||||
|
||||
// 计算三角形面积(使用叉积公式的绝对值)
|
||||
// Area = |((x1 - x3)(y2 - y1) - (x1 - x2)(y3 - y1))| / 2
|
||||
// 为了性能,我们省略除以2,因为只需要比较相对大小
|
||||
const area = Math.abs(
|
||||
(prevPoint.x - avgX) * (point.y - prevPoint.y) -
|
||||
(prevPoint.x - point.x) * (avgY - prevPoint.y),
|
||||
)
|
||||
|
||||
if (area > maxArea) {
|
||||
maxArea = area
|
||||
maxAreaIndex = j
|
||||
}
|
||||
}
|
||||
|
||||
// 选择形成最大面积的点
|
||||
sampled[sampledIndex] = data[maxAreaIndex]!
|
||||
sampledIndex++
|
||||
}
|
||||
|
||||
return sampled
|
||||
}
|
||||
|
||||
/**
|
||||
* 最小-最大抽样算法
|
||||
*
|
||||
* 这是一种简单但有效的抽样方法,将数据分成桶,每个桶选择最小值和最大值。
|
||||
* 适合展示数据的整体范围和波动,但可能会丢失一些细节特征。
|
||||
*
|
||||
* @param data 原始数据点数组
|
||||
* @param threshold 目标点数(必须 >= 2,最终点数可能略多于阈值)
|
||||
* @returns 抽样后的数据点数组
|
||||
*
|
||||
* @example
|
||||
* const original = Array.from({ length: 10000 }, (_, i) => ({ x: i, y: Math.sin(i / 100) }))
|
||||
* const sampled = downsampleMinMax(original, 500)
|
||||
*/
|
||||
export function downsampleMinMax(data: WaveformPoint[], threshold: number): WaveformPoint[] {
|
||||
if (!Array.isArray(data) || data.length === 0) {
|
||||
return []
|
||||
}
|
||||
|
||||
const dataLength = data.length
|
||||
|
||||
// 如果数据点数少于阈值,直接返回
|
||||
if (threshold >= dataLength || threshold <= 1) {
|
||||
return data
|
||||
}
|
||||
|
||||
const sampled: WaveformPoint[] = []
|
||||
|
||||
// 计算每个桶的大小
|
||||
const bucketSize = Math.max(1, Math.floor(dataLength / Math.floor(threshold / 2)))
|
||||
|
||||
for (let i = 0; i < dataLength; i += bucketSize) {
|
||||
const bucketEnd = Math.min(i + bucketSize, dataLength)
|
||||
let minPoint = data[i]!
|
||||
let maxPoint = data[i]!
|
||||
|
||||
// 在当前桶中找到最小和最大的Y值
|
||||
for (let j = i + 1; j < bucketEnd; j++) {
|
||||
const point = data[j]!
|
||||
if (point.y < minPoint.y) {
|
||||
minPoint = point
|
||||
}
|
||||
if (point.y > maxPoint.y) {
|
||||
maxPoint = point
|
||||
}
|
||||
}
|
||||
|
||||
// 按X坐标顺序添加最小值和最大值
|
||||
if (minPoint.x < maxPoint.x) {
|
||||
sampled.push(minPoint)
|
||||
if (minPoint !== maxPoint) {
|
||||
sampled.push(maxPoint)
|
||||
}
|
||||
} else {
|
||||
sampled.push(maxPoint)
|
||||
if (minPoint !== maxPoint) {
|
||||
sampled.push(minPoint)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return sampled
|
||||
}
|
||||
|
||||
/**
|
||||
* 自适应抽样策略
|
||||
*
|
||||
* 根据数据量自动选择最合适的抽样算法和阈值
|
||||
*
|
||||
* @param data 原始数据点数组
|
||||
* @param maxPoints 最大显示点数(可选,默认为5000)
|
||||
* @returns 抽样后的数据点数组和使用的算法信息
|
||||
*/
|
||||
export function adaptiveSampling(
|
||||
data: WaveformPoint[],
|
||||
maxPoints: number = 5000,
|
||||
): { points: WaveformPoint[]; algorithm: 'none' | 'lttb' | 'minmax'; originalCount: number } {
|
||||
const dataLength = data.length
|
||||
|
||||
// 不需要抽样
|
||||
if (dataLength <= maxPoints) {
|
||||
return { points: data, algorithm: 'none', originalCount: dataLength }
|
||||
}
|
||||
|
||||
// 根据数据量选择算法
|
||||
// LTTB 适合保持波形形状,但对极大数据集可能较慢
|
||||
// MinMax 适合快速预览大数据集的范围
|
||||
if (dataLength > maxPoints * 10) {
|
||||
// 超大数据集,使用更快的 MinMax
|
||||
return {
|
||||
points: downsampleMinMax(data, maxPoints),
|
||||
algorithm: 'minmax',
|
||||
originalCount: dataLength,
|
||||
}
|
||||
} else {
|
||||
// 使用 LTTB 以获得更好的视觉质量
|
||||
return {
|
||||
points: downsampleLTTB(data, maxPoints),
|
||||
algorithm: 'lttb',
|
||||
originalCount: dataLength,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* 计算推荐的抽样阈值
|
||||
*
|
||||
* 基于视口宽度和像素密度计算合理的抽样点数
|
||||
*
|
||||
* @param viewportWidth 视口宽度(像素)
|
||||
* @param pixelRatio 设备像素比(默认为 window.devicePixelRatio 或 1)
|
||||
* @param pointsPerPixel 每像素点数(默认为 2,意味着每像素最多2个数据点)
|
||||
* @returns 推荐的抽样点数
|
||||
*/
|
||||
export function calculateSamplingThreshold(
|
||||
viewportWidth: number,
|
||||
pixelRatio: number = typeof window !== 'undefined' ? window.devicePixelRatio : 1,
|
||||
pointsPerPixel: number = 2,
|
||||
): number {
|
||||
return Math.max(100, Math.floor(viewportWidth * pixelRatio * pointsPerPixel))
|
||||
}
|
||||
Reference in New Issue
Block a user