Linear-time shuffling of large datasets for Node.js
This package uses the Rao algorithm to shuffle data sets that are too large to fit in memory. The algorithm is described pretty well by Chris Hardin. The input stream is randomly scattered into "piles" which are stored on disk. Then each pile is shuffled in-memory with Fisher-Yates.
If your data set is extremely large, then even your piles may not fit in memory. In that case, the algorithm could recurse until the piles are small enough, but that feature is not yet implemented here.
Because the input elements are written to disk as part of the shuffle,
big-shuffle
can only take string data. If you need to shuffle other
types of times, serialize them to string
first.
Support for shuffling Buffer
and Uint8Array
objects may be added
later if there is demand.
npm install big-shuffle
For TypeScript users:
import { shuffle } from 'big-shuffle';
import * as path from 'path';
const inArray = [];
function *asyncRange(max: number) {
for (let i = 0; i < max; i++) {
yield i.toString(10);
}
}
const shuffled = shuffle(asyncRange(1000000));
for await (const i of shuffled) {
console.log();
}
This will generate, shuffle, and print a million random numbers.
Should work the same for JavaScript users after a few changes.
function shuffle(
inStream: AsyncIterable<string>,
numPiles: number = 1000, // More piles reduces memory usage but requires more open file descriptors
pileDir: string = path.join(__dirname, 'shuffle_piles'), // Filesystem path where the files are located
): Promise<AsyncIterable<string>>;
Note that the shuffled iterable will not yield any records until the input iterable is fully consumed.
class ShuffleTransform extends stream.Transform{
constructor(
numPiles: number = 1000, // More piles reduces memory usage but requires more open file descriptors
pileDir: string = path.join(__dirname, 'shuffle_piles'), // Filesystem path where the files are located
)
}