welch / tdigest Goto Github PK

TDigest.prototype._addweight = function(nearest, x, n) {
// add weight at location x to nearest centroid. adding x to
// nearest will not shift its relative position in the tree and
// require reinsertion.
//
if (x !== nearest.mean) {
nearest.mean += n * (x - nearest.mean) / (nearest.n + n);
nearest.mean = nearest.mean.toFixed(number) // add this line?
}
nearest.cumn += n;
nearest.mean_cumn += n / 2;
nearest.n += n;
this.n += n;
};

First of all, thanks for writing this excellent library!

I'm currently using compress and then saving the result as JSON. I would like to be able to use said JSON to recreate the TDigest object at a later point in time so that I can compute percentiles.

I'm not terribly familiar with OOP in JavaScript, so I'm not sure I fully understand your source code. Any advice for how to best do this would be very much appreciated.

Thanks!

The current implementation doesn't allow the internal state to be saved and later restored.

I'm proposing adding the following to TDigest:

• load function ( takes object typically achieved via JSON.parse )
• traverse function ( to traverse the centroid binary tree )
• _insert_centroid ( to allow direct insertion )

"saving" will be accomplished via JSON.stringify ( tdObject )

I'll be forking this repo and doing the above shortly.

var a = [1,2,3,4,5]
var b = [7,8,9,10]

var t1 = new TDigest()
t1.push(a);
var c1 = t1.toArray();

var t2 = new TDigest();
t2.push(b)
var c2 = t2.toArray()

var t = new TDigest();
t.push(c1);
t.push(c2);

t.summary()

i want merge centroids. Am i right?

Hey! I'm the developer of Js-sdsl. Official website: https://js-sdsl.github.io/

Now, we published the version 4.1.4.

I see you are using bintrees.

In benchmark, we have confirmed that Js-sdsl is several times faster than bintrees.

I've submitted a PR to eslint and has been accepted.

See eslint/eslint#16255 and eslint/eslint#16267.

I am willing to submit a pull request for this change.

Looking forward to your reply! :D

Does this apply to big data? 500 million？