cyb70289 commented on a change in pull request #9310: URL: https://github.com/apache/arrow/pull/9310#discussion_r565111310
########## File path: cpp/src/arrow/util/tdigest.h ########## @@ -0,0 +1,419 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +// approximate quantiles from arbitrary length dataset with O(1) space +// based on 'Computing Extremely Accurate Quantiles Using t-Digests' from Dunning & Ertl +// - https://arxiv.org/abs/1902.04023 +// - https://github.com/tdunning/t-digest + +#pragma once + +#include <algorithm> +#include <cmath> +#include <queue> +#include <vector> + +#include "arrow/util/logging.h" + +#ifndef M_PI +#define M_PI 3.14159265358979323846 +#endif + +namespace arrow { +namespace internal { + +namespace detail { + +// a numerically stable lerp is unbelievably complex +// but we are *approximating* the quantile, so let's keep it simple +double Lerp(double a, double b, double t) { return a + t * (b - a); } + +// histogram bin +struct Centroid { + double mean; + double weight; // # data points in this bin + + // merge with another centroid + void Merge(const Centroid& centroid) { + weight += centroid.weight; + mean += (centroid.mean - mean) * centroid.weight / weight; + } +}; + +// scale function K0: linear function, as baseline +struct ScalerK0 { + explicit ScalerK0(uint32_t delta) : delta_norm(delta / 2.0) {} + + double K(double q) const { return delta_norm * q; } + double Q(double k) const { return k / delta_norm; } + + const double delta_norm; +}; + +// scale function K1 +struct ScalerK1 { + explicit ScalerK1(uint32_t delta) : delta_norm(delta / (2.0 * M_PI)) {} + + double K(double q) const { return delta_norm * std::asin(2 * q - 1); } + double Q(double k) const { return (std::sin(k / delta_norm) + 1) / 2; } + + const double delta_norm; +}; + +// implements t-digest merging algorithm +template <class T = ScalerK1> +class TDigestMerger : private T { + public: + explicit TDigestMerger(uint32_t delta) : T(delta) {} + + void Reset(double total_weight, std::vector<Centroid>* tdigest) { + total_weight_ = total_weight; + tdigest_ = tdigest; + tdigest_->resize(0); + weight_so_far_ = 0; + weight_limit_ = -1; // trigger first centroid merge + } + + // merge one centroid from a sorted centroid stream + void Add(const Centroid& centroid) { + auto& td = *tdigest_; + const double weight = weight_so_far_ + centroid.weight; + if (weight <= weight_limit_) { + td[td.size() - 1].Merge(centroid); Review comment: Done ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
