浅谈sparse vec检索工程化实现 - JadePeng

前面我们通过两篇文章: BGE M3-Embedding 模型介绍 和 Sparse稀疏检索介绍与实践 介绍了sparse 稀疏检索，今天我们来看看如何建立一个工程化的系统来实现sparse vec的检索。

之前提过milvus最新的V2.4支持sparse检索，我们先看看milvus的实现。

milvus的sparse检索实现

milvus 检索底层引擎是knowhere，主要代码在src/index/sparse 里。

首先，通过数据结构SparseRow，用于表示稀疏向量，支持浮点数（float）类型的数据

class SparseRow {
    static_assert(std::is_same_v<T, fp32>, "SparseRow supports float only");

 public:
    // construct an SparseRow with memory allocated to hold `count` elements.
    SparseRow(size_t count = 0)
        : data_(count ? new uint8_t[count * element_size()] : nullptr), count_(count), own_data_(true) {
    }

    SparseRow(size_t count, uint8_t* data, bool own_data) : data_(data), count_(count), own_data_(own_data) {
    }

    // copy constructor and copy assignment operator perform deep copy
    SparseRow(const SparseRow<T>& other) : SparseRow(other.count_) {
        std::memcpy(data_, other.data_, data_byte_size());
    }

    SparseRow(SparseRow<T>&& other) noexcept : SparseRow() {
        swap(*this, other);
    }

    SparseRow&
    operator=(const SparseRow<T>& other) {
        if (this != &other) {
            SparseRow<T> tmp(other);
            swap(*this, tmp);
        }
        return *this;
    }

    SparseRow&
    operator=(SparseRow<T>&& other) noexcept {
        swap(*this, other);
        return *this;
    }

    ~SparseRow() {
        if (own_data_ && data_ != nullptr) {
            delete[] data_;
            data_ = nullptr;
        }
    }

    size_t
    size() const {
        return count_;
    }

    size_t
    memory_usage() const {
        return data_byte_size() + sizeof(*this);
    }

    // return the number of bytes used by the underlying data array.
    size_t
    data_byte_size() const {
        return count_ * element_size();
    }

    void*
    data() {
        return data_;
    }

    const void*
    data() const {
        return data_;
    }

    // dim of a sparse vector is the max index + 1, or 0 for an empty vector.
    int64_t
    dim() const {
        if (count_ == 0) {
            return 0;
        }
        auto* elem = reinterpret_cast<const ElementProxy*>(data_) + count_ - 1;
        return elem->index + 1;
    }

    SparseIdVal<T>
    operator[](size_t i) const {
        auto* elem = reinterpret_cast<const ElementProxy*>(data_) + i;
        return {elem->index, elem->value};
    }

    void
    set_at(size_t i, table_t index, T value) {
        auto* elem = reinterpret_cast<ElementProxy*>(data_) + i;
        elem->index = index;
        elem->value = value;
    }

    float
    dot(const SparseRow<T>& other) const {
        float product_sum = 0.0f;
        size_t i = 0;
        size_t j = 0;
        // TODO: improve with _mm_cmpistrm or the AVX512 alternative.
        while (i < count_ && j < other.count_) {
            auto* left = reinterpret_cast<const ElementProxy*>(data_) + i;
            auto* right = reinterpret_cast<const ElementProxy*>(other.data_) + j;

            if (left->index < right->index) {
                ++i;
            } else if (left->index > right->index) {
                ++j;
            } else {
                product_sum += left->value * right->value;
                ++i;
                ++j;
            }
        }
        return product_sum;
    }

    friend void
    swap(SparseRow<T>& left, SparseRow<T>& right) {
        using std::swap;
        swap(left.count_, right.count_);
        swap(left.data_, right.data_);
        swap(left.own_data_, right.own_data_);
    }

    static inline size_t
    element_size() {
        return sizeof(table_t) + sizeof(T);
    }

 private:
    // ElementProxy is used to access elements in the data_ array and should
    // never be actually constructed.
    struct __attribute__((packed)) ElementProxy {
        table_t index;
        T value;
        ElementProxy() = delete;
        ElementProxy(const ElementProxy&) = delete;
    };
    // data_ must be sorted by column id. use raw pointer for easy mmap and zero
    // copy.
    uint8_t* data_;
    size_t count_;
    bool own_data_;
};

然后索引具体是在InvertedIndex 类里, 对应sparse_inverted_index.h 文件，首先看定义的一些private 字段。

    std::vector<SparseRow<T>> raw_data_;
    mutable std::shared_mutex mu_;

    std::unordered_map<table_t, std::vector<SparseIdVal<T>>> inverted_lut_;
    bool use_wand_ = false;
    // If we want to drop small values during build, we must first train the
    // index with all the data to compute value_threshold_.
    bool drop_during_build_ = false;
    // when drop_during_build_ is true, any value smaller than value_threshold_
    // will not be added to inverted_lut_. value_threshold_ is set to the
    // drop_ratio_build-th percentile of all absolute values in the index.
    T value_threshold_ = 0.0f;
    std::unordered_map<table_t, T> max_in_dim_;
    size_t max_dim_ = 0;

• raw_data_ 是原始的数据
• inverted_lut_ 可以理解为一个倒排表
• use_wand_ 用于控制查询时，是否使用WAND算法，WAND算法是经典的查询优化算法，可以通过类似跳表的方式跳过一些数据，减少计算量，提升查询效率
• max_in_dim_ 是为wand服务的

索引构建流程

构建，主要是对外提供一个Add数据的方法：

    Status
    Add(const SparseRow<T>* data, size_t rows, int64_t dim) {
        std::unique_lock<std::shared_mutex> lock(mu_);
        auto current_rows = n_rows_internal();
        if (current_rows > 0 && drop_during_build_) {
            LOG_KNOWHERE_ERROR_ << "Not allowed to add data to a built index with drop_ratio_build > 0.";
            return Status::invalid_args;
        }
        if ((size_t)dim > max_dim_) {
            max_dim_ = dim;
        }

        raw_data_.insert(raw_data_.end(), data, data + rows);
        for (size_t i = 0; i < rows; ++i) {
            add_row_to_index(data[i], current_rows + i);
        }
        return Status::success;
    }

这里会更新数据的max_dim，数据追加到raw_data_，然后add_row_to_index，将新的doc放入inverted_lut_, 并更新max_in_dim_，用于记录最大值，方便wand查询时跳过计算。

    inline void
    add_row_to_index(const SparseRow<T>& row, table_t id) {
        for (size_t j = 0; j < row.size(); ++j) {
            auto [idx, val] = row[j];
            // Skip values close enough to zero(which contributes little to
            // the total IP score).
            if (drop_during_build_ && fabs(val) < value_threshold_) {
                continue;
            }
            if (inverted_lut_.find(idx) == inverted_lut_.end()) {
                inverted_lut_[idx];
                if (use_wand_) {
                    max_in_dim_[idx] = 0;
                }
            }
            inverted_lut_[idx].emplace_back(id, val);
            if (use_wand_) {
                max_in_dim_[idx] = std::max(max_in_dim_[idx], val);
            }
        }
    }

索引保存与load

保存时，是自定义的二进制文件：

    Status
    Save(MemoryIOWriter& writer) {
        /**
         * zero copy is not yet implemented, now serializing in a zero copy
         * compatible way while still copying during deserialization.
         *
         * Layout:
         *
         * 1. int32_t rows, sign indicates whether to use wand
         * 2. int32_t cols
         * 3. for each row:
         *     1. int32_t len
         *     2. for each non-zero value:
         *        1. table_t idx
         *        2. T val
         *     With zero copy deserization, each SparseRow object should
         *     reference(not owning) the memory address of the first element.
         *
         * inverted_lut_ and max_in_dim_ not serialized, they will be
         * constructed dynamically during deserialization.
         *
         * Data are densly packed in serialized bytes and no padding is added.
         */
        std::shared_lock<std::shared_mutex> lock(mu_);
        writeBinaryPOD(writer, n_rows_internal() * (use_wand_ ? 1 : -1));
        writeBinaryPOD(writer, n_cols_internal());
        writeBinaryPOD(writer, value_threshold_);
        for (size_t i = 0; i < n_rows_internal(); ++i) {
            auto& row = raw_data_[i];
            writeBinaryPOD(writer, row.size());
            if (row.size() == 0) {
                continue;
            }
            writer.write(row.data(), row.size() * SparseRow<T>::element_size());
        }
        return Status::success;
    }

索引文件格式：

• 1. int32_t rows 总记录数，通过±符号来区分是否 use wand
• 1. int32_t cols 列数
• 1. for each row:

• 1. int32_t len 长度
  

• 2. for each non-zero value:
  

•    1. table_t idx term的id编号
  

•    2. T val   term的权重
  

注意，这里inverted_lut_倒排表是没有存储的，是在加载的时候重建，所以load的过程，就是一个逆过程：

Status
    Load(MemoryIOReader& reader) {
        std::unique_lock<std::shared_mutex> lock(mu_);
        int64_t rows;
        readBinaryPOD(reader, rows);
        use_wand_ = rows > 0;
        rows = std::abs(rows);
        readBinaryPOD(reader, max_dim_);
        readBinaryPOD(reader, value_threshold_);

        raw_data_.reserve(rows);

        for (int64_t i = 0; i < rows; ++i) {
            size_t count;
            readBinaryPOD(reader, count);
            raw_data_.emplace_back(count);
            if (count == 0) {
                continue;
            }
            reader.read(raw_data_[i].data(), count * SparseRow<T>::element_size());
            add_row_to_index(raw_data_[i], i);
        }

        return Status::success;
    }

我们来回顾，compute_lexical_matching_score其实就是计算共同term的weight score相乘，然后加起来，所以可以想象下，暴力检索大概就是把所有term对应的doc取并集，然后计算lexical_matching_score，最后取topk。

我们来看milvus的实现，先看暴力检索：

    // find the top-k candidates using brute force search, k as specified by the capacity of the heap.
    // any value in q_vec that is smaller than q_threshold and any value with dimension >= n_cols() will be ignored.
    // TODO: may switch to row-wise brute force if filter rate is high. Benchmark needed.
    void
    search_brute_force(const SparseRow<T>& q_vec, T q_threshold, MaxMinHeap<T>& heap, const BitsetView& bitset) const {
        auto scores = compute_all_distances(q_vec, q_threshold);
        for (size_t i = 0; i < n_rows_internal(); ++i) {
            if ((bitset.empty() || !bitset.test(i)) && scores[i] != 0) {
                heap.push(i, scores[i]);
            }
        }
    }

    std::vector<float>
    compute_all_distances(const SparseRow<T>& q_vec, T q_threshold) const {
        std::vector<float> scores(n_rows_internal(), 0.0f);
        for (size_t idx = 0; idx < q_vec.size(); ++idx) {
            auto [i, v] = q_vec[idx];
            if (v < q_threshold || i >= n_cols_internal()) {
                continue;
            }
            auto lut_it = inverted_lut_.find(i);
            if (lut_it == inverted_lut_.end()) {
                continue;
            }
            // TODO: improve with SIMD
            auto& lut = lut_it->second;
            for (size_t j = 0; j < lut.size(); j++) {
                auto [idx, val] = lut[j];
                scores[idx] += v * float(val);
            }
        }
        return scores;
    }

• 核心在compute_all_distances里，先通过q_vec得到每一个term id，然后从inverted_lut_里找到term对应的doc list，然后计算score，相同doc id的score累加
• 最后用MaxMinHeap堆，来取topk

暴力检索能保准精准性，但是效率比较低。我们来看使用wand优化的检索：

// any value in q_vec that is smaller than q_threshold will be ignored.
    void
    search_wand(const SparseRow<T>& q_vec, T q_threshold, MaxMinHeap<T>& heap, const BitsetView& bitset) const {
        auto q_dim = q_vec.size();
        std::vector<std::shared_ptr<Cursor<std::vector<SparseIdVal<T>>>>> cursors(q_dim);
        auto valid_q_dim = 0;
        // 倒排链
        for (size_t i = 0; i < q_dim; ++i) {
	        // idx(term_id)
            auto [idx, val] = q_vec[i];
            if (std::abs(val) < q_threshold || idx >= n_cols_internal()) {
                continue;
            }
            auto lut_it = inverted_lut_.find(idx);
            if (lut_it == inverted_lut_.end()) {
                continue;
            }
            auto& lut = lut_it->second;
            // max_in_dim_ 记录了term index 的最大score
            cursors[valid_q_dim++] = std::make_shared<Cursor<std::vector<SparseIdVal<T>>>>(
                lut, n_rows_internal(), max_in_dim_.find(idx)->second * val, val, bitset);
        }
        if (valid_q_dim == 0) {
            return;
        }
        cursors.resize(valid_q_dim);
        auto sort_cursors = [&cursors] {
            std::sort(cursors.begin(), cursors.end(),
                      [](auto& x, auto& y) { return x->cur_vec_id() < y->cur_vec_id(); });
        };
        sort_cursors();
        // 堆未满，或者新的score > 堆顶的score
        auto score_above_threshold = [&heap](float x) { return !heap.full() || x > heap.top().val; };
        while (true) {
	        // 上边界
            float upper_bound = 0;
            // pivot 满足条件的倒排链的序号
            size_t pivot;
            bool found_pivot = false;
            for (pivot = 0; pivot < cursors.size(); ++pivot) {
	            // 有倒排结束
                if (cursors[pivot]->is_end()) {
                    break;
                }
                upper_bound += cursors[pivot]->max_score();
                if (score_above_threshold(upper_bound)) {
                    found_pivot = true;
                    break;
                }
            }
            if (!found_pivot) {
                break;
            }
            // 找到满足upper_bound 满足条件的pivot_id
            table_t pivot_id = cursors[pivot]->cur_vec_id();
            // 如果第一个倒排链的当前vec_id (doc_id) 等于pivot_id，可以直接从第0个倒排链开始，计算score
            if (pivot_id == cursors[0]->cur_vec_id()) {
                float score = 0;
                // 遍历所有cursors，累加score
                for (auto& cursor : cursors) {
                    if (cursor->cur_vec_id() != pivot_id) {
                        break;
                    }
                    score += cursor->cur_distance() * cursor->q_value();
                    // 倒排链移到下一位
                    cursor->next();
                }
                // 放入堆
                heap.push(pivot_id, score);
                // 重排cursors，保证最小的vec_id在最前面
                sort_cursors();
            } else {
                // 第一个倒排链的当前vec_id不等于pivot_id, pivot>=1
                // 那么从pivot(满足threshold的倒排链序号)往前找是否有cur_vec_id==pivot_id的
                size_t next_list = pivot;
                for (; cursors[next_list]->cur_vec_id() == pivot_id; --next_list) {
                }
                // 这里的next_list的cur_vec_id 不一定等与pivot_id，将list seek到pivot_id
                // seek后，cursors[next_list].cur_vec_id() >= pivot_id，通过seek，可以跳过一些vec id
                cursors[next_list]->seek(pivot_id);
                // 从next_list + 1开始
                for (size_t i = next_list + 1; i < cursors.size(); ++i) {
                    // 如果当前cur_vec_id >= 上一个则停止
                    if (cursors[i]->cur_vec_id() >= cursors[i - 1]->cur_vec_id()) {
                        break;
                    }
                    // 否则，交换倒排链，可以确保==pivot_id的倒排链交换到前面
                    std::swap(cursors[i], cursors[i - 1]);
                }
            }
        }
    }

• 首先是倒排链取出来放入cursors，然后对cursors按照vec_id排序，将vec_id较小的排到倒排链的首位
• 通过score_above_threshold，遍历cursors找符合条件的cursor 索引号pivot，这里通过堆未满，或者新的score > 堆顶的score来判断，可以跳过一些score小的
• 然后找到pivot cursor对应的pivot_id，也就是doc id，然后判断第一个倒排链的cur_vec_id 是否等于pivot_id：
  • 如果等于，就可以遍历倒排链，计算pivot_id的score，然后放入小顶堆中排序，然后重排倒排链
  • 如果不等于，那么就需要想办法将cur_vec_id == pivot_id的往前放，同时跳过倒排链中vec_id < cur_vec_id的数据（减枝）

用golang实现轻量级sparse vec检索

用类似milvus的方法，我们简单实现一个golang版本的

package main

import (
	"container/heap"
	"encoding/binary"
	"fmt"
	"io"
	"math/rand"
	"os"
	"sort"
	"time"
)

type Cursor struct {
	docIDs     []int32
	weights    []float64
	maxScore   float64
	termWeight float64
	currentIdx int
}

func NewCursor(docIDs []int32, weights []float64, maxScore float64, weight float64) *Cursor {
	return &Cursor{
		docIDs:     docIDs,
		weights:    weights,
		maxScore:   maxScore,
		termWeight: weight,
		currentIdx: 0,
	}
}

func (c *Cursor) Next() {
	c.currentIdx++
}

func (c *Cursor) Seek(docId int32) {
	for {
		if c.IsEnd() {
			break
		}
		if c.CurrentDocID() < docId {
			c.Next()
		} else {
			break
		}
	}
}

func (c *Cursor) IsEnd() bool {
	return c.currentIdx >= len(c.docIDs)
}

func (c *Cursor) CurrentDocID() int32 {
	return c.docIDs[c.currentIdx]
}

func (c *Cursor) CurrentDocWeight() float64 {
	return c.weights[c.currentIdx]
}

// DocVectors type will map docID to its vector
type DocVectors map[int32]map[int32]float64

// InvertedIndex type will map termID to sorted list of docIDs
type InvertedIndex map[int32][]int32

// TermMaxScore will keep track of maximum scores for terms
type TermMaxScores map[int32]float64

// SparseIndex class struct
type SparseIndex struct {
	docVectors    DocVectors
	invertedIndex InvertedIndex
	termMaxScores TermMaxScores
	dim           int32
}

// NewSparseIndex initializes a new SparseIndex with empty structures
func NewSparseIndex() *SparseIndex {
	return &SparseIndex{
		docVectors:    make(DocVectors),
		invertedIndex: make(InvertedIndex),
		termMaxScores: make(TermMaxScores),
		dim:           0,
	}
}

// Add method for adding documents to the sparse index
func (index *SparseIndex) Add(docID int32, vec map[int32]float64) {
	index.docVectors[docID] = vec

	for termID, score := range vec {
		index.invertedIndex[termID] = append(index.invertedIndex[termID], docID)

		// Track max score for each term
		if maxScore, ok := index.termMaxScores[termID]; !ok || score > maxScore {
			index.termMaxScores[termID] = score
		}
		if termID > index.dim {
			index.dim = termID
		}
	}
}

// Save index to file
func (index *SparseIndex) Save(filename string) error {
	file, err := os.Create(filename)
	if err != nil {
		return err
	}
	defer file.Close()

	// Write the dimension
	binary.Write(file, binary.LittleEndian, index.dim)

	// Write each document vector
	for docID, vec := range index.docVectors {
		binary.Write(file, binary.LittleEndian, docID)
		vecSize := int32(len(vec))
		binary.Write(file, binary.LittleEndian, vecSize)

		for termID, score := range vec {
			binary.Write(file, binary.LittleEndian, termID)
			binary.Write(file, binary.LittleEndian, score)
		}
	}

	return nil
}

// Load index from file
func (index *SparseIndex) Load(filename string) error {
	file, err := os.Open(filename)
	if err != nil {
		return err
	}
	defer file.Close()

	var dim int32
	binary.Read(file, binary.LittleEndian, &dim)
	index.dim = dim

	for {
		var docID int32
		err := binary.Read(file, binary.LittleEndian, &docID)
		if err == io.EOF {
			break // End of file
		} else if err != nil {
			return err // Some other error
		}

		var vecSize int32
		binary.Read(file, binary.LittleEndian, &vecSize)
		vec := make(map[int32]float64)

		for i := int32(0); i < vecSize; i++ {
			var termID int32
			var score float64
			binary.Read(file, binary.LittleEndian, &termID)
			binary.Read(file, binary.LittleEndian, &score)
			vec[termID] = score
		}

		index.Add(docID, vec) // Rebuild the index
	}
	return nil
}

func (index *SparseIndex) bruteSearch(queryVec map[int32]float64, K int) []int32 {

	scores := computeAllDistances(queryVec, index)

	// 取top k
	docHeap := &DocScoreHeap{}
	for docID, score := range scores {
		if docHeap.Len() < K {
			heap.Push(docHeap, &DocScore{docID, score})
		} else if (*docHeap)[0].score < score {
			heap.Pop(docHeap)
			heap.Push(docHeap, &DocScore{docID, score})
		}
	}

	topDocs := make([]int32, 0, K)
	for docHeap.Len() > 0 {
		el := heap.Pop(docHeap).(*DocScore)
		topDocs = append(topDocs, el.docID)
	}

	sort.Slice(topDocs, func(i, j int) bool {
		return topDocs[i] < topDocs[j]
	})

	return topDocs
}

func computeAllDistances(queryVec map[int32]float64, index *SparseIndex) map[int32]float64 {
	scores := make(map[int32]float64)
	for term, qWeight := range queryVec {
		if postingList, exists := index.invertedIndex[term]; exists {
			for _, docID := range postingList {
				docVec := index.docVectors[docID]
				docWeight, exists := docVec[term]
				if !exists {
					continue
				}
				score := qWeight * docWeight

				if _, ok := scores[docID]; !ok {
					scores[docID] = score
				} else {
					scores[docID] += score
				}
			}
		}
	}
	return scores
}

// TopK retrieves the top K documents nearest to the query vector
func (index *SparseIndex) WandSearch(queryVec map[int32]float64, K int) []int32 {
	docHeap := &DocScoreHeap{}

	// 倒排链
	postingLists := make([]*Cursor, len(queryVec))
	idx := 0
	for term, termWeight := range queryVec {
		if postingList, exists := index.invertedIndex[term]; exists {
			// 包含term的doc，term对应的weight
			weights := make([]float64, len(postingList))
			for i, docID := range postingList {
				weights[i] = index.docVectors[docID][term]
			}
			postingLists[idx] = NewCursor(postingList, weights, index.termMaxScores[term]*termWeight, termWeight)
			idx += 1
		}
	}

	sortPostings := func() {
		for i := range postingLists {
			if postingLists[i].IsEnd() {
				return
			}
		}
		// 将postingLists按照首个docid排序
		sort.Slice(postingLists, func(i, j int) bool {
			return postingLists[i].CurrentDocID() < postingLists[j].CurrentDocID()
		})
	}

	sortPostings()

	scoreAboveThreshold := func(value float64) bool {
		return docHeap.Len() < K || (*docHeap)[0].score < value
	}

	for {
		upperBound := 0.0
		foundPivot := false
		pivot := 0
		for idx := range postingLists {
			if postingLists[idx].IsEnd() {
				break
			}

			upperBound += postingLists[idx].maxScore
			if scoreAboveThreshold(upperBound) {
				foundPivot = true
				pivot = idx
				break
			}
		}

		if !foundPivot {
			break
		}

		// 找到满足upper_bound 满足条件的pivot_id
		pivotId := postingLists[pivot].CurrentDocID()
		if pivotId == postingLists[0].CurrentDocID() {
			//	如果第一个倒排链的当前vec_id (doc_id) 等于pivot_id，可以直接从第0个倒排链开始，计算score
			score := 0.0
			// 遍历所有cursors，累加score
			for idx := range postingLists {
				cursor := postingLists[idx]
				if cursor.CurrentDocID() != pivotId {
					break
				}
				score += cursor.CurrentDocWeight() * cursor.termWeight
				// 移到下一个docid
				postingLists[idx].Next()
			}

			// 放入堆s
			if docHeap.Len() < K {
				heap.Push(docHeap, &DocScore{pivotId, score})
			} else if (*docHeap)[0].score < score {
				heap.Pop(docHeap)
				heap.Push(docHeap, &DocScore{pivotId, score})
			}

			// 重排cursors，保证最小的vec_id在最前面
			sortPostings()
		} else {
			// 第一个倒排链的当前vec_id不等于pivot_id, pivot>=1
			// 那么从pivot(满足threshold的倒排链序号)往前找是否有cur_vec_id==pivot_id的
			nextList := pivot
			for ; postingLists[nextList].CurrentDocID() == pivotId; nextList-- {
			}
			// 这里的next_list的cur_vec_id 不一定等与pivot_id，将list seek到pivot_id
			// seek后，cursors[next_list].cur_vec_id() >= pivot_id，通过seek，可以跳过一些vec id
			postingLists[nextList].Seek(pivotId)
			// 从next_list + 1开始

			for i := nextList + 1; i < len(postingLists); i++ {
				// 如果当前cur_vec_id >= 上一个则停止
				if postingLists[i].CurrentDocID() >= postingLists[i-1].CurrentDocID() {
					break
				}
				// 否则，交换倒排链，可以确保==pivot_id的倒排链交换到前面
				temp := postingLists[i]
				postingLists[i] = postingLists[i-1]
				postingLists[i-1] = temp
			}
		}
	}

	topDocs := make([]int32, 0, K)
	for docHeap.Len() > 0 {
		el := heap.Pop(docHeap).(*DocScore)
		topDocs = append(topDocs, el.docID)
	}
	sort.Slice(topDocs, func(i, j int) bool {
		return topDocs[i] < topDocs[j]
	})

	return topDocs
}

// Helper structure to manage the priority queue for the top-K documents
type DocScore struct {
	docID int32
	score float64
}

type DocScoreHeap []*DocScore

func (h DocScoreHeap) Len() int           { return len(h) }
func (h DocScoreHeap) Less(i, j int) bool { return h[i].score < h[j].score }
func (h DocScoreHeap) Swap(i, j int)      { h[i], h[j] = h[j], h[i] }

func (h *DocScoreHeap) Push(x interface{}) {
	*h = append(*h, x.(*DocScore))
}

func (h *DocScoreHeap) Pop() interface{} {
	old := *h
	n := len(old)
	x := old[n-1]
	*h = old[0 : n-1]
	return x
}

func main() {
	index := NewSparseIndex()

	rand.Seed(time.Now().UnixNano())
	// Add document vectors as needed
	for i := 1; i <= 1000; i++ {
		// 打印当前i的值
		index.Add(int32(i), map[int32]float64{101: rand.Float64(),
			150: rand.Float64(),
			190: rand.Float64(),
			500: rand.Float64()})
	}
	//index.Save("index.bin")
	//index.Load("index.bin")
	topDocs := index.WandSearch(map[int32]float64{101: rand.Float64(), 150: rand.Float64(), 190: rand.Float64(),
		500: rand.Float64()}, 10)
	fmt.Println("Top Docs:", topDocs)
}

• 代码实现了索引的构建、保存和加载，检索方面实现了暴力检索和WAND检索
• 注意，添加doc时，需要保障doc有序，实际应用中，docid可以引擎维护一个真实id到递增docid的映射
• 代码中已经有注释，这里不再赘述，注意代码未充分调试，可能有bug
• 代码实现倒排表全放到内存，效率高，但对内存要求高

sparse 检索整体类似传统的文本检索，因此传统的工程优化方法可以运用到sparse检索中，本文分析了milvus的实现，并实现了一个golang版本的sparse检索。