介绍

给定一个未知大小的数据集，按随机顺序获得一个随机样本。这通常被用于数据分析，作为一种从数据池或仓库或 CSV 文件中获取小的随机样本的方法。

实现

JavaScript

/*
Given a data set of an unknown size, Get a random sample in a random order
It's used in data analytics, often as a way to get a small random sample from a data lake or warehouse, or from a large CSV file
*/
function shuf (datasetSource, sampleSize) {
  const output = fillBaseSample(datasetSource, sampleSize)
  return randomizeOutputFromDataset(datasetSource, output)
}

/**
 * Fills the output if possible, with the minimum number of values
 * @param {Iterable.<T>} datasetSource The iterable source of data
 * @param {number} sampleSize The size of the sample to extract from the dataset
 * @returns {Array.<T>} The random sample, as an array
 * @template T
 */
function fillBaseSample (datasetSource, sampleSize) {
  let filledIndexes = []
  let output = new Array(sampleSize)

  // Spread data out filling the array
  while (true) {
    const iterator = datasetSource.next()
    if (iterator.done) break

    let insertTo = Math.floor(Math.random() * output.length)
    while (filledIndexes.includes(insertTo)) {
      insertTo++
      if (insertTo === output.length) {
        insertTo = 0
      }
    }
    output[insertTo] = {
      value: iterator.value
    }

    filledIndexes = [...filledIndexes, insertTo]

    if (filledIndexes.length === sampleSize) {
      break
    }
  }

  if (filledIndexes.length < output.length) {
    // Not a large enough dataset to fill the sample - trim empty values
    output = output.filter((_, i) => filledIndexes.includes(i))
  }

  return output.map((o) => o.value)
}

/**
 * Replaces values in the output randomly with new ones from the dataset
 * @param {Iterable.<T>} datasetSource The iterable source of data
 * @param {Array.<T>} output The output so far, filled with data
 * @returns {Array.<T>} The random sample, as an array
 * @template T
 */
function randomizeOutputFromDataset (datasetSource, output) {
  const newOutput = [...output]
  let readSoFar = output.length

  while (true) {
    const iterator = datasetSource.next()
    if (iterator.done) break
    readSoFar++

    const insertTo = Math.floor(Math.random() * readSoFar)
    if (insertTo < newOutput.length) {
      newOutput[insertTo] = iterator.value
    }
  }

  return newOutput
}

// Example

/**
 * Generates a random range of data, with values between 0 and 2^31 - 1
 * @param {number} length The number of data items to generate
 * @returns {Iterable<number>} Random iterable data
*/
function * generateRandomData (length) {
  const maxValue = Math.pow(2, 31) - 1
  for (let i = 0; i < length; i++) {
    yield Math.floor(Math.random() * maxValue)
  }
}

// const source = generateRandomData(1000)
// const result = shuf(source, 10)

参考

• Sampling from a population of unknown size | by Andrea C. | Towards Data Science (opens new window)