You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
1008 lines
46 KiB
1008 lines
46 KiB
4 months ago
|
import { __spreadArray } from "tslib";
|
||
|
|
import floydWarshall from './floydWarshall';
|
||
|
|
import gSpan from './gSpan/gSpan';
|
||
|
|
import dijkstra from './dijkstra';
|
||
|
|
import { uniqueId } from './util';
|
||
|
|
/**
|
||
|
|
* 为 graphData 中每个节点生成邻居单元数组
|
||
|
|
* @param graphData
|
||
|
|
* @param spm
|
||
|
|
* @param nodeLabelProp
|
||
|
|
* @param k k-近邻
|
||
|
|
*/
|
||
|
|
var findKNeighborUnits = function findKNeighborUnits(graphData, spm, nodeLabelProp, k) {
|
||
|
|
if (nodeLabelProp === void 0) {
|
||
|
|
nodeLabelProp = 'cluster';
|
||
|
|
}
|
||
|
|
if (k === void 0) {
|
||
|
|
k = 2;
|
||
|
|
}
|
||
|
|
var units = [];
|
||
|
|
var nodes = graphData.nodes;
|
||
|
|
spm.forEach(function (row, i) {
|
||
|
|
units.push(findKNeighborUnit(nodes, row, i, nodeLabelProp, k));
|
||
|
|
});
|
||
|
|
return units;
|
||
|
|
};
|
||
|
|
var findKNeighborUnit = function findKNeighborUnit(nodes, row, i, nodeLabelProp, k) {
|
||
|
|
var unitNodeIdxs = [i];
|
||
|
|
var neighbors = [];
|
||
|
|
var labelCountMap = {};
|
||
|
|
row.forEach(function (v, j) {
|
||
|
|
if (v <= k && i !== j) {
|
||
|
|
unitNodeIdxs.push(j);
|
||
|
|
neighbors.push(nodes[j]);
|
||
|
|
var label = nodes[j][nodeLabelProp];
|
||
|
|
if (!labelCountMap[label]) labelCountMap[label] = {
|
||
|
|
count: 1,
|
||
|
|
dists: [v]
|
||
|
|
};else {
|
||
|
|
labelCountMap[label].count++;
|
||
|
|
labelCountMap[label].dists.push(v);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
});
|
||
|
|
// 将 labelCountMap 中的 dists 按照从小到大排序，方便后面使用
|
||
|
|
Object.keys(labelCountMap).forEach(function (label) {
|
||
|
|
labelCountMap[label].dists = labelCountMap[label].dists.sort(function (a, b) {
|
||
|
|
return a - b;
|
||
|
|
});
|
||
|
|
});
|
||
|
|
return {
|
||
|
|
nodeIdx: i,
|
||
|
|
nodeId: nodes[i].id,
|
||
|
|
nodeIdxs: unitNodeIdxs,
|
||
|
|
neighbors: neighbors,
|
||
|
|
neighborNum: unitNodeIdxs.length - 1,
|
||
|
|
nodeLabelCountMap: labelCountMap
|
||
|
|
};
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 随机寻找点对，满足距离小于 k
|
||
|
|
* @param k 参数 k，表示 k-近邻
|
||
|
|
* @param nodeNum 参数 length
|
||
|
|
* @param maxNodePairNum 寻找点对的数量不超过 maxNodePairNum
|
||
|
|
* @param spm 最短路径矩阵
|
||
|
|
*/
|
||
|
|
var findNodePairsRandomly = function findNodePairsRandomly(k, nodeNum, maxNodePairNum, kNeighborUnits, spm) {
|
||
|
|
// 每个节点需要随机找出的点对数
|
||
|
|
var nodePairNumEachNode = Math.ceil(maxNodePairNum / nodeNum);
|
||
|
|
var nodePairMap = {};
|
||
|
|
var foundNodePairCount = 0;
|
||
|
|
// 遍历节点，为每个节点随机找出 nodePairNumEachNode 个点对，满足距离小于 k。找到的点对数量超过 maxNodePairNum 或所有节点遍历结束时终止
|
||
|
|
kNeighborUnits.forEach(function (unit, i) {
|
||
|
|
// 若未达到 nodePairNumEachNode，或循环次数小于最大循环次数(2 * nodeNum)，继续循环
|
||
|
|
var nodePairForICount = 0;
|
||
|
|
var outerLoopCount = 0;
|
||
|
|
var neighbors = unit.nodeIdxs; // the first one is the center node
|
||
|
|
var neighborNum = unit.neighborNum - 1;
|
||
|
|
while (nodePairForICount < nodePairNumEachNode) {
|
||
|
|
// 另一端节点在节点数组中的的 index
|
||
|
|
var oidx = neighbors[1 + Math.floor(Math.random() * neighborNum)];
|
||
|
|
var innerLoopCount = 0;
|
||
|
|
// 若随机得到的另一端 idx 不符合条件，则继续 random。条件是不是同一个节点、这个点对没有被记录过、距离小于 k
|
||
|
|
while (nodePairMap["".concat(i, "-").concat(oidx)] || nodePairMap["".concat(oidx, "-").concat(i)]) {
|
||
|
|
oidx = Math.floor(Math.random() * nodeNum);
|
||
|
|
innerLoopCount++;
|
||
|
|
if (innerLoopCount > 2 * nodeNum) break; // 循环次数大于最大循环次数(2 * nodeNum)跳出循环，避免死循环
|
||
|
|
}
|
||
|
|
|
||
|
|
if (innerLoopCount < 2 * nodeNum) {
|
||
|
|
// 未达到最大循环次数，说明找到了合适的另一端
|
||
|
|
nodePairMap["".concat(i, "-").concat(oidx)] = {
|
||
|
|
start: i,
|
||
|
|
end: oidx,
|
||
|
|
distance: spm[i][oidx]
|
||
|
|
};
|
||
|
|
nodePairForICount++;
|
||
|
|
foundNodePairCount++;
|
||
|
|
// 如果当前找到的点对数量达到了上限，返回结果
|
||
|
|
if (foundNodePairCount >= maxNodePairNum) return nodePairMap;
|
||
|
|
}
|
||
|
|
outerLoopCount++;
|
||
|
|
if (outerLoopCount > 2 * nodeNum) break; // 循环次数大于最大循环次数(2 * nodeNum)跳出循环，避免死循环
|
||
|
|
}
|
||
|
|
// 这个节点没有找到足够 nodePairNumEachNode 的点对。更新 nodePairNumEachNode，让后续节点找更多的点对
|
||
|
|
if (nodePairForICount < nodePairNumEachNode) {
|
||
|
|
var gap = nodePairNumEachNode - nodePairForICount;
|
||
|
|
nodePairNumEachNode = (nodePairNumEachNode + gap) / (nodeNum - i - 1);
|
||
|
|
}
|
||
|
|
});
|
||
|
|
return nodePairMap;
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 计算所有 nodePairMap 中节点对的相交邻居诱导子图
|
||
|
|
* @param nodePairMap 节点对 map，key 为 node1.id-node2.id，value 为 { startNodeIdx, endNodeIdx, distance }
|
||
|
|
* @param neighborUnits 每个节点的邻居元数组
|
||
|
|
* @param graphData 原图数据
|
||
|
|
* @param edgeMap 边的 map，方便检索
|
||
|
|
* @param cachedInducedGraphMap 缓存的结果，下次进入该函数将继续更新该缓存，若 key 在缓存中存在则不需要重复计算
|
||
|
|
*/
|
||
|
|
var getIntersectNeighborInducedGraph = function getIntersectNeighborInducedGraph(nodePairMap, neighborUnits, graphData, cachedInducedGraphMap) {
|
||
|
|
var nodes = graphData.nodes;
|
||
|
|
if (!cachedInducedGraphMap) cachedInducedGraphMap = {};
|
||
|
|
Object.keys(nodePairMap).forEach(function (key) {
|
||
|
|
var _a, _b;
|
||
|
|
if (cachedInducedGraphMap && cachedInducedGraphMap[key]) return;
|
||
|
|
cachedInducedGraphMap[key] = {
|
||
|
|
nodes: [],
|
||
|
|
edges: []
|
||
|
|
};
|
||
|
|
var pair = nodePairMap[key];
|
||
|
|
var startUnitNodeIds = (_a = neighborUnits[pair.start]) === null || _a === void 0 ? void 0 : _a.nodeIdxs;
|
||
|
|
var endUnitNodeIds = (_b = neighborUnits[pair.end]) === null || _b === void 0 ? void 0 : _b.nodeIdxs;
|
||
|
|
if (!startUnitNodeIds || !endUnitNodeIds) return; // 不存在邻元，返回空图
|
||
|
|
var endSet = new Set(endUnitNodeIds);
|
||
|
|
var intersect = startUnitNodeIds.filter(function (x) {
|
||
|
|
return endSet.has(x);
|
||
|
|
}); // 可能会爆栈（在 1580 + 6 nodes full-connected 时出现）
|
||
|
|
if (!intersect || !intersect.length) return; // 没有交集，返回空图
|
||
|
|
var intersectIdMap = {};
|
||
|
|
var intersectLength = intersect.length;
|
||
|
|
for (var i = 0; i < intersectLength; i++) {
|
||
|
|
var node = nodes[intersect[i]];
|
||
|
|
cachedInducedGraphMap[key].nodes.push(node); // 将交集中的点加入诱导子图
|
||
|
|
intersectIdMap[node.id] = true;
|
||
|
|
}
|
||
|
|
// 遍历所有边数据，如果边的两端都在交集中，将该边加入诱导子图
|
||
|
|
graphData.edges.forEach(function (edge) {
|
||
|
|
if (intersectIdMap[edge.source] && intersectIdMap[edge.target]) cachedInducedGraphMap[key].edges.push(edge);
|
||
|
|
});
|
||
|
|
});
|
||
|
|
return cachedInducedGraphMap;
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 计算 strcutre 在 graph 上的匹配数量
|
||
|
|
* @param graph 图数据
|
||
|
|
* @param structure 目前支持只有两个节点一条边的最简单结构
|
||
|
|
* @param nodeLabelProp 节点类型字段名
|
||
|
|
* @param edgeLabelProp 边类型字段名
|
||
|
|
*/
|
||
|
|
var getMatchedCount = function getMatchedCount(graph, structure, nodeLabelProp, edgeLabelProp) {
|
||
|
|
var _a, _b;
|
||
|
|
var nodeMap = {};
|
||
|
|
graph.nodes.forEach(function (node) {
|
||
|
|
nodeMap[node.id] = node;
|
||
|
|
});
|
||
|
|
var count = 0;
|
||
|
|
if (!((_a = structure === null || structure === void 0 ? void 0 : structure.edges) === null || _a === void 0 ? void 0 : _a.length) || ((_b = structure === null || structure === void 0 ? void 0 : structure.nodes) === null || _b === void 0 ? void 0 : _b.length) < 2) return 0;
|
||
|
|
graph.edges.forEach(function (e) {
|
||
|
|
var sourceLabel = nodeMap[e.source][nodeLabelProp];
|
||
|
|
var targetLabel = nodeMap[e.target][nodeLabelProp];
|
||
|
|
var strNodeLabel1 = structure === null || structure === void 0 ? void 0 : structure.nodes[0][nodeLabelProp];
|
||
|
|
var strNodeLabel2 = structure === null || structure === void 0 ? void 0 : structure.nodes[1][nodeLabelProp];
|
||
|
|
var strEdgeLabel = structure === null || structure === void 0 ? void 0 : structure.edges[0][edgeLabelProp];
|
||
|
|
if (e[edgeLabelProp] !== strEdgeLabel) return;
|
||
|
|
if (sourceLabel === strNodeLabel1 && targetLabel === strNodeLabel2 || sourceLabel === strNodeLabel2 && targetLabel === strNodeLabel1) {
|
||
|
|
count++;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
return count;
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* structures 中寻找最具有代表性的一个。这个结构是使得 matchedCountMap 的分组方式类内间距最小，类间间距最大
|
||
|
|
* @param matchedCountMap 每个 structure 分类后的各图匹配数量，格式 { [strcture.idx]: { [interInducedGraphKey]: count } }
|
||
|
|
* @param structureNum strcuture 个数，与 matchedCountMap.length 对应
|
||
|
|
* @param structures
|
||
|
|
*/
|
||
|
|
var findRepresentStructure = function findRepresentStructure(matchedCountMap, structureNum, structures) {
|
||
|
|
var maxOffset = Infinity,
|
||
|
|
representClusterType = 0;
|
||
|
|
var _loop_1 = function _loop_1(i) {
|
||
|
|
// 一种分组的 map，key 是 intGraph 的 key，value 是 structures[i] 的匹配个数
|
||
|
|
var countMapI = matchedCountMap[i];
|
||
|
|
// 按照 value 为该组排序，生成 keys 的数组：
|
||
|
|
var sortedGraphKeys = Object.keys(countMapI).sort(function (a, b) {
|
||
|
|
return countMapI[a] - countMapI[b];
|
||
|
|
});
|
||
|
|
// 共 100 个 graphKeys，将 graphKeys 按顺序分为 groupNum 组
|
||
|
|
var groupNum = 10;
|
||
|
|
var clusters = []; // 总共有 groupNum 个项
|
||
|
|
sortedGraphKeys.forEach(function (key, j) {
|
||
|
|
if (!clusters[j % groupNum]) clusters[j % groupNum] = {
|
||
|
|
graphs: [],
|
||
|
|
totalCount: 0,
|
||
|
|
aveCount: 0
|
||
|
|
};
|
||
|
|
clusters[j % groupNum].graphs.push(key);
|
||
|
|
clusters[j % groupNum].totalCount += countMapI[key];
|
||
|
|
});
|
||
|
|
// 计算 cluster 与 cluster 之间的距离 innerDist，每个 cluster 内部的距离 intraDist
|
||
|
|
var aveIntraDist = 0; // 该类的类内平均值
|
||
|
|
var aveCounts = []; // 类内平均匹配数量，将用于计算类间距离
|
||
|
|
clusters.forEach(function (graphsInCluster) {
|
||
|
|
// 类内均值
|
||
|
|
var aveCount = graphsInCluster.totalCount / graphsInCluster.graphs.length;
|
||
|
|
graphsInCluster.aveCount = aveCount;
|
||
|
|
aveCounts.push(aveCount);
|
||
|
|
// 对于每类，计算类内间距平均值
|
||
|
|
var aveIntraPerCluster = 0;
|
||
|
|
var graphsNum = graphsInCluster.length;
|
||
|
|
graphsInCluster.graphs.forEach(function (graphKey1, j) {
|
||
|
|
var graph1Count = countMapI[graphKey1];
|
||
|
|
graphsInCluster.graphs.forEach(function (graphKey2, k) {
|
||
|
|
if (j === k) return;
|
||
|
|
aveIntraPerCluster += Math.abs(graph1Count - countMapI[graphKey2]);
|
||
|
|
});
|
||
|
|
});
|
||
|
|
aveIntraPerCluster /= graphsNum * (graphsNum - 1) / 2;
|
||
|
|
aveIntraDist += aveIntraPerCluster;
|
||
|
|
});
|
||
|
|
aveIntraDist /= clusters.length;
|
||
|
|
// 用类内均值计算类间距
|
||
|
|
var aveInterDist = 0; // 类间间距平均值
|
||
|
|
aveCounts.forEach(function (aveCount1, j) {
|
||
|
|
aveCounts.forEach(function (aveCount2, k) {
|
||
|
|
if (j === k) return;
|
||
|
|
aveInterDist += Math.abs(aveCount1 - aveCount2);
|
||
|
|
});
|
||
|
|
aveInterDist /= aveCounts.length * (aveCounts.length - 1) / 2;
|
||
|
|
});
|
||
|
|
// 寻找 (类间间距均值-类内间距均值) 最大的一种分组方式（对应的 structure 就是最终要找的唯一 DS(G)）
|
||
|
|
var offset = aveInterDist - aveIntraDist;
|
||
|
|
if (maxOffset < offset) {
|
||
|
|
maxOffset = offset;
|
||
|
|
representClusterType = i;
|
||
|
|
}
|
||
|
|
};
|
||
|
|
for (var i = 0; i < structureNum; i++) {
|
||
|
|
_loop_1(i);
|
||
|
|
}
|
||
|
|
return {
|
||
|
|
structure: structures[representClusterType],
|
||
|
|
structureCountMap: matchedCountMap[representClusterType]
|
||
|
|
};
|
||
|
|
};
|
||
|
|
var getNodeMaps = function getNodeMaps(nodes, nodeLabelProp) {
|
||
|
|
var nodeMap = {},
|
||
|
|
nodeLabelMap = {};
|
||
|
|
nodes.forEach(function (node, i) {
|
||
|
|
nodeMap[node.id] = {
|
||
|
|
idx: i,
|
||
|
|
node: node,
|
||
|
|
degree: 0,
|
||
|
|
inDegree: 0,
|
||
|
|
outDegree: 0
|
||
|
|
};
|
||
|
|
var label = node[nodeLabelProp];
|
||
|
|
if (!nodeLabelMap[label]) nodeLabelMap[label] = [];
|
||
|
|
nodeLabelMap[label].push(node);
|
||
|
|
});
|
||
|
|
return {
|
||
|
|
nodeMap: nodeMap,
|
||
|
|
nodeLabelMap: nodeLabelMap
|
||
|
|
};
|
||
|
|
};
|
||
|
|
var getEdgeMaps = function getEdgeMaps(edges, edgeLabelProp, nodeMap) {
|
||
|
|
var edgeMap = {},
|
||
|
|
edgeLabelMap = {};
|
||
|
|
edges.forEach(function (edge, i) {
|
||
|
|
edgeMap["".concat(uniqueId)] = {
|
||
|
|
idx: i,
|
||
|
|
edge: edge
|
||
|
|
};
|
||
|
|
var label = edge[edgeLabelProp];
|
||
|
|
if (!edgeLabelMap[label]) edgeLabelMap[label] = [];
|
||
|
|
edgeLabelMap[label].push(edge);
|
||
|
|
var sourceNode = nodeMap[edge.source];
|
||
|
|
if (sourceNode) {
|
||
|
|
sourceNode.degree++;
|
||
|
|
sourceNode.outDegree++;
|
||
|
|
}
|
||
|
|
var targetNode = nodeMap[edge.target];
|
||
|
|
if (targetNode) {
|
||
|
|
targetNode.degree++;
|
||
|
|
targetNode.inDegree++;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
return {
|
||
|
|
edgeMap: edgeMap,
|
||
|
|
edgeLabelMap: edgeLabelMap
|
||
|
|
};
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 输出最短路径的 map，key 为 sourceNode.id-targetNode.id，value 为这两个节点的最短路径长度
|
||
|
|
* @param nodes
|
||
|
|
* @param spm
|
||
|
|
* @param directed
|
||
|
|
*/
|
||
|
|
var getSpmMap = function getSpmMap(nodes, spm, directed) {
|
||
|
|
var length = spm.length;
|
||
|
|
var map = {};
|
||
|
|
spm.forEach(function (row, i) {
|
||
|
|
var start = directed ? 0 : i + 1;
|
||
|
|
var iId = nodes[i].id;
|
||
|
|
for (var j = start; j < length; j++) {
|
||
|
|
if (i === j) continue;
|
||
|
|
var jId = nodes[j].id;
|
||
|
|
var dist = row[j];
|
||
|
|
map["".concat(iId, "-").concat(jId)] = dist;
|
||
|
|
if (!directed) map["".concat(jId, "-").concat(iId)] = dist;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
return map;
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 计算一对节点（node1，node2）的 NDS 距离
|
||
|
|
* @param graph 原图数据
|
||
|
|
* @param node1
|
||
|
|
* @param node2
|
||
|
|
*/
|
||
|
|
var getNDSDist = function getNDSDist(graph, node1, node2, nodeMap, spDist, kNeighborUnits, structure, nodeLabelProp, edgeLabelProp, cachedNDSMap, cachedInterInducedGraph) {
|
||
|
|
var _a;
|
||
|
|
var key = "".concat(node1.id, "-").concat(node2.id);
|
||
|
|
if (cachedNDSMap && cachedNDSMap[key]) return cachedNDSMap[key];
|
||
|
|
var interInducedGraph = cachedInterInducedGraph ? cachedInterInducedGraph[key] : undefined;
|
||
|
|
// 若没有缓存相交邻居诱导子图，计算
|
||
|
|
if (!interInducedGraph) {
|
||
|
|
var pairMap = (_a = {}, _a[key] = {
|
||
|
|
start: nodeMap[node1.id].idx,
|
||
|
|
end: nodeMap[node2.id].idx,
|
||
|
|
distance: spDist
|
||
|
|
}, _a);
|
||
|
|
cachedInterInducedGraph = getIntersectNeighborInducedGraph(pairMap, kNeighborUnits, graph, cachedInterInducedGraph);
|
||
|
|
interInducedGraph = cachedInterInducedGraph[key];
|
||
|
|
}
|
||
|
|
return getMatchedCount(interInducedGraph, structure, nodeLabelProp, edgeLabelProp);
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* 计算 pattern 上绩点的度数并存储到 minPatternNodeLabelDegreeMap
|
||
|
|
*/
|
||
|
|
var stashPatternNodeLabelDegreeMap = function stashPatternNodeLabelDegreeMap(minPatternNodeLabelDegreeMap, neighborLabel, patternNodeMap, patternNodeLabelMap) {
|
||
|
|
var _a, _b, _c;
|
||
|
|
var minPatternNodeLabelDegree = (_a = minPatternNodeLabelDegreeMap[neighborLabel]) === null || _a === void 0 ? void 0 : _a.degree;
|
||
|
|
var minPatternNodeLabelInDegree = (_b = minPatternNodeLabelDegreeMap[neighborLabel]) === null || _b === void 0 ? void 0 : _b.inDegree;
|
||
|
|
var minPatternNodeLabelOutDegree = (_c = minPatternNodeLabelDegreeMap[neighborLabel]) === null || _c === void 0 ? void 0 : _c.outDegree;
|
||
|
|
if (minPatternNodeLabelDegreeMap[neighborLabel] === undefined) {
|
||
|
|
minPatternNodeLabelDegree = Infinity;
|
||
|
|
minPatternNodeLabelInDegree = Infinity;
|
||
|
|
minPatternNodeLabelOutDegree = Infinity;
|
||
|
|
patternNodeLabelMap[neighborLabel].forEach(function (patternNodeWithLabel) {
|
||
|
|
var patternNodeDegree = patternNodeMap[patternNodeWithLabel.id].degree;
|
||
|
|
if (minPatternNodeLabelDegree > patternNodeDegree) minPatternNodeLabelDegree = patternNodeDegree;
|
||
|
|
var patternNodeInDegree = patternNodeMap[patternNodeWithLabel.id].inDegree;
|
||
|
|
if (minPatternNodeLabelInDegree > patternNodeInDegree) minPatternNodeLabelInDegree = patternNodeInDegree;
|
||
|
|
var patternNodeOutDegree = patternNodeMap[patternNodeWithLabel.id].outDegree;
|
||
|
|
if (minPatternNodeLabelOutDegree > patternNodeOutDegree) minPatternNodeLabelOutDegree = patternNodeOutDegree;
|
||
|
|
});
|
||
|
|
minPatternNodeLabelDegreeMap[neighborLabel] = {
|
||
|
|
degree: minPatternNodeLabelDegree,
|
||
|
|
inDegree: minPatternNodeLabelInDegree,
|
||
|
|
outDegree: minPatternNodeLabelOutDegree
|
||
|
|
};
|
||
|
|
}
|
||
|
|
return {
|
||
|
|
minPatternNodeLabelDegree: minPatternNodeLabelDegree,
|
||
|
|
minPatternNodeLabelInDegree: minPatternNodeLabelInDegree,
|
||
|
|
minPatternNodeLabelOutDegree: minPatternNodeLabelOutDegree
|
||
|
|
};
|
||
|
|
};
|
||
|
|
/**
|
||
|
|
* GADDI 模式匹配
|
||
|
|
* @param graphData 原图数据
|
||
|
|
* @param pattern 搜索图（需要在原图上搜索的模式）数据
|
||
|
|
* @param directed 是否计算有向图，默认 false
|
||
|
|
* @param k 参数 k，表示 k-近邻
|
||
|
|
* @param length 参数 length
|
||
|
|
* @param nodeLabelProp 节点数据中代表节点标签（分类信息）的属性名。默认为 cluster
|
||
|
|
* @param edgeLabelProp 边数据中代表边标签（分类信息）的属性名。默认为 cluster
|
||
|
|
*/
|
||
|
|
var GADDI = function GADDI(graphData, pattern, directed, k, length, nodeLabelProp, edgeLabelProp) {
|
||
|
|
var _a;
|
||
|
|
if (directed === void 0) {
|
||
|
|
directed = false;
|
||
|
|
}
|
||
|
|
if (nodeLabelProp === void 0) {
|
||
|
|
nodeLabelProp = 'cluster';
|
||
|
|
}
|
||
|
|
if (edgeLabelProp === void 0) {
|
||
|
|
edgeLabelProp = 'cluster';
|
||
|
|
}
|
||
|
|
if (!graphData || !graphData.nodes) return;
|
||
|
|
// 分为三步：
|
||
|
|
// 0. 预计算：节点/边数，邻接矩阵、最短路径矩阵
|
||
|
|
// 1. 处理原图 graphData。再分为 1~5 小步
|
||
|
|
// 2. 匹配
|
||
|
|
// console.log("----- stage-pre: preprocessing -------");
|
||
|
|
// -------- 第零步，预计算：节点/边数，邻接矩阵、最短路径矩阵-------
|
||
|
|
var nodeNum = graphData.nodes.length;
|
||
|
|
if (!nodeNum) return;
|
||
|
|
// console.log("----- stage-pre.1: calc shortest path matrix for graph -------");
|
||
|
|
var spm = floydWarshall(graphData, directed);
|
||
|
|
// console.log(
|
||
|
|
// "----- stage-pre.2: calc shortest path matrix for pattern -------"
|
||
|
|
// );
|
||
|
|
var patternSpm = floydWarshall(pattern, directed);
|
||
|
|
// console.log(
|
||
|
|
// "----- stage-pre.3: calc shortest path matrix map for graph -------"
|
||
|
|
// );
|
||
|
|
var spmMap = getSpmMap(graphData.nodes, spm, directed);
|
||
|
|
// console.log(
|
||
|
|
// "----- stage-pre.4: calc shortest path matrix map for pattern -------"
|
||
|
|
// );
|
||
|
|
var patternSpmMap = getSpmMap(pattern.nodes, patternSpm, directed);
|
||
|
|
// console.log("----- stage-pre.5: establish maps -------");
|
||
|
|
// 节点的 map，以 id 为 id 映射，方便后续快速检索
|
||
|
|
var _b = getNodeMaps(graphData.nodes, nodeLabelProp),
|
||
|
|
nodeMap = _b.nodeMap,
|
||
|
|
nodeLabelMap = _b.nodeLabelMap;
|
||
|
|
var _c = getNodeMaps(pattern.nodes, nodeLabelProp),
|
||
|
|
patternNodeMap = _c.nodeMap,
|
||
|
|
patternNodeLabelMap = _c.nodeLabelMap;
|
||
|
|
// 计算节点度数
|
||
|
|
getEdgeMaps(graphData.edges, edgeLabelProp, nodeMap);
|
||
|
|
var patternEdgeLabelMap = getEdgeMaps(pattern.edges, edgeLabelProp, patternNodeMap).edgeLabelMap;
|
||
|
|
// 若未指定 length，自动计算 pattern 半径（最短路径最大值）
|
||
|
|
var patternSpmSpread = [];
|
||
|
|
patternSpm === null || patternSpm === void 0 ? void 0 : patternSpm.forEach(function (row) {
|
||
|
|
patternSpmSpread = patternSpmSpread.concat(row);
|
||
|
|
});
|
||
|
|
if (!length) length = Math.max.apply(Math, __spreadArray(__spreadArray([], patternSpmSpread, false), [2], false));
|
||
|
|
if (!k) k = length;
|
||
|
|
// console.log("params", directed, length, k);
|
||
|
|
// console.log("----- stage-pre.6: calc k neighbor units -------");
|
||
|
|
// 计算每个节点的 k 邻元集合
|
||
|
|
var kNeighborUnits = findKNeighborUnits(graphData, spm, nodeLabelProp, k);
|
||
|
|
var patternKNeighborUnits = findKNeighborUnits(pattern, patternSpm, nodeLabelProp, k);
|
||
|
|
// console.log(
|
||
|
|
// "----- stage0: going to processing graph and find intersect neighbor induced graphs -------"
|
||
|
|
// );
|
||
|
|
// console.log("----- stage0.1: going to select random node pairs -------");
|
||
|
|
// -------- 第一步，处理原图 graphData-------
|
||
|
|
// 1.1. 随机选择最多 100 个点对，满足距离小于 Length 和 k
|
||
|
|
// 当 graphData 少于 20 个节点，则不能找出 100 个点对，只找出不多于 n(n-1)/2 个点对
|
||
|
|
var maxNodePairNum = Math.min(100, nodeNum * (nodeNum - 1) / 2);
|
||
|
|
var nodePairsMap = findNodePairsRandomly(k, nodeNum, maxNodePairNum, kNeighborUnits, spm);
|
||
|
|
// console.log(
|
||
|
|
// "----- stage0.2: going to calculate intersect neighbor induced graphs -------"
|
||
|
|
// );
|
||
|
|
// 1.2. 生成上面节点对的相应相交邻居诱导子图。格式为 {'beginNodeIdx-endNodeIdx': {nodes: [], edges: []}}
|
||
|
|
var intGMap = getIntersectNeighborInducedGraph(nodePairsMap, kNeighborUnits, graphData);
|
||
|
|
// 1.3. 使用 gSpan 算法（frequent graph mining）计算 ISIntG 的前 10 个频率最高的子结构（3-4条边）
|
||
|
|
var top = 10,
|
||
|
|
minSupport = 1,
|
||
|
|
minNodeNum = 1,
|
||
|
|
maxNodeNum = 4;
|
||
|
|
var params = {
|
||
|
|
graphs: intGMap,
|
||
|
|
nodeLabelProp: nodeLabelProp,
|
||
|
|
edgeLabelProp: edgeLabelProp,
|
||
|
|
minSupport: minSupport,
|
||
|
|
minNodeNum: minNodeNum,
|
||
|
|
maxNodeNum: maxNodeNum,
|
||
|
|
directed: directed
|
||
|
|
};
|
||
|
|
// console.log(
|
||
|
|
// "----- stage1: (gSpan) going to find frequent structure dsG -------"
|
||
|
|
// );
|
||
|
|
// console.log("----- stage1.1: going to run gSpan -------");
|
||
|
|
// 暂时假设生成的 sub structure 都只有一条边
|
||
|
|
var freStructures = gSpan(params).slice(0, top);
|
||
|
|
// structureNum 可能小于 top
|
||
|
|
var structureNum = freStructures.length;
|
||
|
|
// 1.4. 计算上述 10 个子结构在 intGMap 中每个诱导子图的匹配个数
|
||
|
|
var matchedCountMap = [];
|
||
|
|
freStructures.forEach(function (structure, i) {
|
||
|
|
matchedCountMap[i] = {};
|
||
|
|
Object.keys(intGMap).forEach(function (key) {
|
||
|
|
var graph = intGMap[key];
|
||
|
|
var subStructureCount = getMatchedCount(graph, structure, nodeLabelProp, edgeLabelProp);
|
||
|
|
matchedCountMap[i][key] = subStructureCount;
|
||
|
|
});
|
||
|
|
});
|
||
|
|
// console.log(
|
||
|
|
// "----- stage1.1: going to find the most represent strucutre -------"
|
||
|
|
// );
|
||
|
|
// 1.5. 对于每个子结构，根据匹配个数为 intGMap 中的诱导子图分组，生成 structureNum 种分组
|
||
|
|
// 计算每种分组的类间距和类内间距，找到类间距最大、类内间距最小的一种分组，这种分组对应的子结构被选为唯一代表性子结构 DS(G)
|
||
|
|
var _d = findRepresentStructure(matchedCountMap, structureNum, freStructures),
|
||
|
|
dsG = _d.structure,
|
||
|
|
ndsDist = _d.structureCountMap;
|
||
|
|
// -------- 第二步，匹配-------
|
||
|
|
// 2.1 找到从 Q 中的一个节点作为起始节点，寻找 G 中的匹配。这个其实节点的标签可以在 G 中找到最多的节点
|
||
|
|
var beginPNode = pattern.nodes[0],
|
||
|
|
candidates = [],
|
||
|
|
label = (_a = pattern.nodes[0]) === null || _a === void 0 ? void 0 : _a[nodeLabelProp],
|
||
|
|
maxNodeNumWithSameLabel = -Infinity;
|
||
|
|
pattern.nodes.forEach(function (node) {
|
||
|
|
var pLabel = node[nodeLabelProp];
|
||
|
|
var nodesWithSameLabel = nodeLabelMap[pLabel];
|
||
|
|
if ((nodesWithSameLabel === null || nodesWithSameLabel === void 0 ? void 0 : nodesWithSameLabel.length) > maxNodeNumWithSameLabel) {
|
||
|
|
maxNodeNumWithSameLabel = nodesWithSameLabel.length;
|
||
|
|
candidates = nodesWithSameLabel;
|
||
|
|
label = pLabel;
|
||
|
|
beginPNode = node;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
// console.log("----- stage2: going to find candidates -------");
|
||
|
|
// 全局缓存，避免重复计算
|
||
|
|
var minPatternNodeLabelDegreeMap = {}; // key 是 label，value 是该 label 节点的最小度数
|
||
|
|
var patternIntGraphMap = {},
|
||
|
|
patternNDSDist = {},
|
||
|
|
// key 为 node.id-node.id
|
||
|
|
patternNDSDistMap = {}; // key 为 node.id-label2，value nds距离值数组（按从大到小排序，无需关心具体对应哪个 node2）
|
||
|
|
// 2.2.2 对于 Q 中的另一个标签的 k 个节点，计算它们到 node 的最短路径以及 NDS 距离
|
||
|
|
var patternSpDist = {};
|
||
|
|
var patternSpDistBack = {};
|
||
|
|
Object.keys(patternNodeLabelMap).forEach(function (label2, j) {
|
||
|
|
patternSpDist[label2] = [];
|
||
|
|
if (directed) {
|
||
|
|
patternSpDistBack[label2] = [];
|
||
|
|
}
|
||
|
|
var maxDist = -Infinity;
|
||
|
|
var patternNodesWithLabel2 = patternNodeLabelMap[label2];
|
||
|
|
var patternNodePairMap = {};
|
||
|
|
patternNodesWithLabel2.forEach(function (nodeWithLabel2) {
|
||
|
|
var dist = patternSpmMap["".concat(beginPNode.id, "-").concat(nodeWithLabel2.id)];
|
||
|
|
dist && patternSpDist[label2].push(dist);
|
||
|
|
if (maxDist < dist) maxDist = dist;
|
||
|
|
patternNodePairMap["".concat(beginPNode.id, "-").concat(nodeWithLabel2.id)] = {
|
||
|
|
start: 0,
|
||
|
|
end: patternNodeMap[nodeWithLabel2.id].idx,
|
||
|
|
distance: dist
|
||
|
|
};
|
||
|
|
if (directed) {
|
||
|
|
var distBack = patternSpmMap["".concat(nodeWithLabel2.id, "-").concat(beginPNode.id)];
|
||
|
|
distBack && patternSpDistBack[label2].push(distBack);
|
||
|
|
}
|
||
|
|
});
|
||
|
|
// spDist[label2] 按照从小到大排序
|
||
|
|
patternSpDist[label2] = patternSpDist[label2].sort(function (a, b) {
|
||
|
|
return a - b;
|
||
|
|
});
|
||
|
|
if (directed) patternSpDistBack[label2] = patternSpDistBack[label2].sort(function (a, b) {
|
||
|
|
return a - b;
|
||
|
|
});
|
||
|
|
// 计算 Q 中所有 label2 节点到 beginPNode 的 NDS 距离
|
||
|
|
// 所有 label2 节点到 beginPNode 的邻居相交诱导子图：
|
||
|
|
// key: node1.id-node2.id
|
||
|
|
patternIntGraphMap = getIntersectNeighborInducedGraph(patternNodePairMap, patternKNeighborUnits, pattern, patternIntGraphMap);
|
||
|
|
// pattern 中 beginNode 到当前 label2 节点 的 NDS 距离（数组，无需关心具体对应到哪个节点）
|
||
|
|
var currentPatternNDSDistArray = [];
|
||
|
|
Object.keys(patternNodePairMap).forEach(function (key) {
|
||
|
|
if (patternNDSDist[key]) {
|
||
|
|
currentPatternNDSDistArray.push(patternNDSDist[key]);
|
||
|
|
return; // 缓存过则不需要再次计算
|
||
|
|
}
|
||
|
|
|
||
|
|
var patternIntGraph = patternIntGraphMap[key];
|
||
|
|
patternNDSDist[key] = getMatchedCount(patternIntGraph, dsG, nodeLabelProp, edgeLabelProp);
|
||
|
|
currentPatternNDSDistArray.push(patternNDSDist[key]);
|
||
|
|
});
|
||
|
|
// 根据值为 currentPatternNDSDist 从大到小排序
|
||
|
|
currentPatternNDSDistArray = currentPatternNDSDistArray.sort(function (a, b) {
|
||
|
|
return b - a;
|
||
|
|
});
|
||
|
|
patternNDSDistMap["".concat(beginPNode.id, "-").concat(label2)] = currentPatternNDSDistArray;
|
||
|
|
if (label2 === label) return;
|
||
|
|
var candidatesNum = (candidates === null || candidates === void 0 ? void 0 : candidates.length) || 0;
|
||
|
|
var _loop_4 = function _loop_4(m) {
|
||
|
|
var cNode = candidates[m];
|
||
|
|
// prune1：若 candidates 中节点 cNode 的 kNeighborUnits 中标签为 label2 的节点个数少于 pattern 中 label2 个数，删去它
|
||
|
|
var graphNeighborUnit = kNeighborUnits[nodeMap[cNode.id].idx];
|
||
|
|
var graphNeighborUnitCountMap = graphNeighborUnit.nodeLabelCountMap[label2];
|
||
|
|
var patternLabel2Num = patternNodeLabelMap[label2].length;
|
||
|
|
if (!graphNeighborUnitCountMap || graphNeighborUnitCountMap.count < patternLabel2Num) {
|
||
|
|
candidates.splice(m, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
// prune2：若 candidates 中节点 cNode 到 kNeighborUnits 中标签为 label2 的节点最短路径大于 patternSpDist[label2]，删去它
|
||
|
|
// (prune2 规则即：candidate 相关的最短路径的最大 spDist[label2].length 个，按照大小顺序依次和 patternSpDist[label2] 中的值比较，只要遇到一个是 G > Q 的，就删去这个 candidate)
|
||
|
|
var prune2Invalid = false;
|
||
|
|
for (var n = 0; n < patternLabel2Num; n++) {
|
||
|
|
if (graphNeighborUnitCountMap.dists[n] > patternSpDist[label2][n]) {
|
||
|
|
prune2Invalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
if (prune2Invalid) {
|
||
|
|
candidates.splice(m, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
// prune3：若 candidates 中节点 cNode 到 kNeighborUnits 中标签为 label2 的节点 NDS 距离小于 patternNDSDist[beginNode.id-label2]，删去它
|
||
|
|
// TODO：prune3，currentPatternNDSDistArray 与 currentNDSDist 的比较
|
||
|
|
// 计算 G 中所有 label2 节点到 cNode 的 NDS 距离
|
||
|
|
// 所有 label2 节点到 cNode 的邻居相交诱导子图：
|
||
|
|
var cNodePairMap = {};
|
||
|
|
graphNeighborUnit.neighbors.forEach(function (neighborNode) {
|
||
|
|
var dist = spmMap["".concat(cNode.id, "-").concat(neighborNode.id)];
|
||
|
|
cNodePairMap["".concat(cNode.id, "-").concat(neighborNode.id)] = {
|
||
|
|
start: nodeMap[cNode.id].idx,
|
||
|
|
end: nodeMap[neighborNode.id].idx,
|
||
|
|
distance: dist
|
||
|
|
};
|
||
|
|
});
|
||
|
|
// 更新 intGMap
|
||
|
|
intGMap = getIntersectNeighborInducedGraph(cNodePairMap, kNeighborUnits, graphData, intGMap);
|
||
|
|
// candidate 到它周围 label2 节点的 NDS 距离, key 是 node.id-node.id
|
||
|
|
var currentNDSDistArray = [];
|
||
|
|
Object.keys(cNodePairMap).forEach(function (key) {
|
||
|
|
if (ndsDist[key]) {
|
||
|
|
currentNDSDistArray.push(ndsDist[key]);
|
||
|
|
return; // 缓存过则不需要再次计算
|
||
|
|
}
|
||
|
|
|
||
|
|
var intGraph = intGMap[key];
|
||
|
|
ndsDist[key] = getMatchedCount(intGraph, dsG, nodeLabelProp, edgeLabelProp);
|
||
|
|
currentNDSDistArray.push(ndsDist[key]);
|
||
|
|
});
|
||
|
|
// 根据值为 currentNDSDistArray 从大到小排序
|
||
|
|
currentNDSDistArray = currentNDSDistArray.sort(function (a, b) {
|
||
|
|
return b - a;
|
||
|
|
});
|
||
|
|
var prune3Invalid = false;
|
||
|
|
for (var n = 0; n < patternLabel2Num; n++) {
|
||
|
|
if (currentNDSDistArray[n] < currentPatternNDSDistArray[n]) {
|
||
|
|
prune3Invalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
if (prune3Invalid) {
|
||
|
|
candidates.splice(m, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
};
|
||
|
|
for (var m = candidatesNum - 1; m >= 0; m--) {
|
||
|
|
_loop_4(m);
|
||
|
|
}
|
||
|
|
});
|
||
|
|
var candidateGraphs = [];
|
||
|
|
// console.log(
|
||
|
|
// "----- stage3: going to splice neighbors for each candidate graph -------"
|
||
|
|
// );
|
||
|
|
// candidates 经过筛选后，以每个 candidate 为中心，生成 Length-neighbor 的邻居诱导子图
|
||
|
|
// 并在诱导子图中去除不可能在 Q 上找到匹配的点：在 Q 上不存在的 label，其他 label 到 candidate 的最大最短距离符合 Q、NDS 距离符合 Q
|
||
|
|
candidates === null || candidates === void 0 ? void 0 : candidates.forEach(function (candidate) {
|
||
|
|
var nodeIdx = nodeMap[candidate.id].idx;
|
||
|
|
var lengthNeighborUnit = findKNeighborUnit(graphData.nodes, spm[nodeIdx], nodeIdx, nodeLabelProp, length);
|
||
|
|
var neighborNodes = lengthNeighborUnit.neighbors;
|
||
|
|
// 删除不可能找到匹配的邻居点
|
||
|
|
var neighborNum = neighborNodes.length;
|
||
|
|
var unmatched = false;
|
||
|
|
for (var i = neighborNum - 1; i >= 0; i--) {
|
||
|
|
// 如果通过裁剪，符合条件的节点数量已过少，说明不能匹配这个 candidate 相关的图
|
||
|
|
if (neighborNodes.length + 1 < pattern.nodes.length) {
|
||
|
|
unmatched = true;
|
||
|
|
return;
|
||
|
|
}
|
||
|
|
var neighborNode = neighborNodes[i];
|
||
|
|
var neighborLabel = neighborNode[nodeLabelProp];
|
||
|
|
// prune1: 若该邻居点的 label 不存在于 pattern 中，移除这个点
|
||
|
|
if (!patternNodeLabelMap[neighborLabel] || !patternNodeLabelMap[neighborLabel].length) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
// prune2: 若该邻居点到 candidate 的最短路径比和它有相同 label 的节点到 beginPNode 的最大最短路径长度长，移除这个点
|
||
|
|
// prune2.1: 如果没有这个标签到 beginPNode 的距离记录，说明 pattern 上（可能 beginPNode 是这个 label）没有其他这个 label 的节点
|
||
|
|
if (!patternSpDist[neighborLabel] || !patternSpDist[neighborLabel].length) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
var key = "".concat(candidate.id, "-").concat(neighborNode.id);
|
||
|
|
// prune2.2
|
||
|
|
var distToCandidate = spmMap[key];
|
||
|
|
var idx = patternSpDist[neighborLabel].length - 1;
|
||
|
|
var maxDistWithLabelInPattern = patternSpDist[neighborLabel][idx]; // patternSpDist[neighborLabel] 已经按照从小到大排序
|
||
|
|
if (distToCandidate > maxDistWithLabelInPattern) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
if (directed) {
|
||
|
|
var keyBack = "".concat(neighborNode.id, "-").concat(candidate.id);
|
||
|
|
var distFromCandidate = spmMap[keyBack];
|
||
|
|
idx = patternSpDistBack[neighborLabel].length - 1;
|
||
|
|
var maxBackDistWithLabelInPattern = patternSpDistBack[neighborLabel][idx];
|
||
|
|
if (distFromCandidate > maxBackDistWithLabelInPattern) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
// prune3: 若该邻居点到 candidate 的 NDS 距离比和它有相同 label 的节点到 beginPNode 的最小 NDS 距离小，移除这个点
|
||
|
|
var ndsToCandidate = ndsDist[key] ? ndsDist[key] : getNDSDist(graphData, candidate, neighborNode, nodeMap, distToCandidate, kNeighborUnits, dsG, nodeLabelProp, edgeLabelProp, ndsDist, intGMap);
|
||
|
|
var patternKey = "".concat(beginPNode.id, "-").concat(neighborLabel);
|
||
|
|
var minNdsWithLabelInPattern = patternNDSDistMap[patternKey][patternNDSDistMap[patternKey].length - 1]; // patternNDSDist[key] 一定存在
|
||
|
|
if (ndsToCandidate < minNdsWithLabelInPattern) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
// prune4: 若该邻居点的度数小于 pattern 同 label 节点最小度数，删去该点
|
||
|
|
var _a = stashPatternNodeLabelDegreeMap(minPatternNodeLabelDegreeMap, neighborLabel, patternNodeMap, patternNodeLabelMap),
|
||
|
|
minPatternNodeLabelDegree = _a.minPatternNodeLabelDegree,
|
||
|
|
minPatternNodeLabelInDegree = _a.minPatternNodeLabelInDegree,
|
||
|
|
minPatternNodeLabelOutDegree = _a.minPatternNodeLabelOutDegree;
|
||
|
|
if (nodeMap[neighborNode.id].degree < minPatternNodeLabelDegree) {
|
||
|
|
neighborNodes.splice(i, 1);
|
||
|
|
continue;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
// 节点在个数上符合匹配（不少于 pattern 的节点个数），现在筛选相关边
|
||
|
|
if (!unmatched) {
|
||
|
|
candidateGraphs.push({
|
||
|
|
nodes: [candidate].concat(neighborNodes)
|
||
|
|
});
|
||
|
|
}
|
||
|
|
});
|
||
|
|
// console.log(
|
||
|
|
// "----- stage4: going to splice edges and neighbors for each candidate graph -------"
|
||
|
|
// );
|
||
|
|
var undirectedLengthsToBeginPNode = dijkstra(pattern, beginPNode.id, false).length;
|
||
|
|
var undirectedLengthsToBeginPNodeLabelMap = {};
|
||
|
|
if (directed) {
|
||
|
|
Object.keys(undirectedLengthsToBeginPNode).forEach(function (nodeId) {
|
||
|
|
var nodeLabel = patternNodeMap[nodeId].node[nodeLabelProp];
|
||
|
|
if (!undirectedLengthsToBeginPNodeLabelMap[nodeLabel]) undirectedLengthsToBeginPNodeLabelMap[nodeLabel] = [undirectedLengthsToBeginPNode[nodeId]];else undirectedLengthsToBeginPNodeLabelMap[nodeLabel].push(undirectedLengthsToBeginPNode[nodeId]);
|
||
|
|
});
|
||
|
|
Object.keys(undirectedLengthsToBeginPNodeLabelMap).forEach(function (pLabel) {
|
||
|
|
undirectedLengthsToBeginPNodeLabelMap[pLabel].sort(function (a, b) {
|
||
|
|
return a - b;
|
||
|
|
});
|
||
|
|
});
|
||
|
|
} else {
|
||
|
|
undirectedLengthsToBeginPNodeLabelMap = patternSpDist;
|
||
|
|
}
|
||
|
|
// 现在 candidateGraphs 里面只有节点，进行边的筛选
|
||
|
|
var candidateGraphNum = candidateGraphs.length;
|
||
|
|
var _loop_2 = function _loop_2(i) {
|
||
|
|
var candidateGraph = candidateGraphs[i];
|
||
|
|
var candidate = candidateGraph.nodes[0];
|
||
|
|
var candidateNodeLabelCountMap = {};
|
||
|
|
var candidateNodeMap = {};
|
||
|
|
candidateGraph.nodes.forEach(function (node, q) {
|
||
|
|
candidateNodeMap[node.id] = {
|
||
|
|
idx: q,
|
||
|
|
node: node,
|
||
|
|
degree: 0,
|
||
|
|
inDegree: 0,
|
||
|
|
outDegree: 0
|
||
|
|
};
|
||
|
|
var cNodeLabel = node[nodeLabelProp];
|
||
|
|
if (!candidateNodeLabelCountMap[cNodeLabel]) candidateNodeLabelCountMap[cNodeLabel] = 1;else candidateNodeLabelCountMap[cNodeLabel]++;
|
||
|
|
});
|
||
|
|
// 根据 candidate 和 neighborNodes 中的节点生成 G 的诱导子图
|
||
|
|
// 即，将 graphData 上两端都在 candidateGraph.nodes 中的边放入 candidateEdges
|
||
|
|
var candidateEdges = [];
|
||
|
|
var edgeLabelCountMap = {};
|
||
|
|
graphData.edges.forEach(function (edge) {
|
||
|
|
if (candidateNodeMap[edge.source] && candidateNodeMap[edge.target]) {
|
||
|
|
candidateEdges.push(edge);
|
||
|
|
if (!edgeLabelCountMap[edge[edgeLabelProp]]) edgeLabelCountMap[edge[edgeLabelProp]] = 1;else edgeLabelCountMap[edge[edgeLabelProp]]++;
|
||
|
|
candidateNodeMap[edge.source].degree++;
|
||
|
|
candidateNodeMap[edge.target].degree++;
|
||
|
|
candidateNodeMap[edge.source].outDegree++;
|
||
|
|
candidateNodeMap[edge.target].inDegree++;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
// prune：若有一个 edgeLabel 在 candidateGraph 上的个数少于 pattern，去除该图
|
||
|
|
var pattenrEdgeLabelNum = Object.keys(patternEdgeLabelMap).length;
|
||
|
|
var prunedByEdgeLabel = false;
|
||
|
|
for (var e = 0; e < pattenrEdgeLabelNum; e++) {
|
||
|
|
var label_1 = Object.keys(patternEdgeLabelMap)[e];
|
||
|
|
if (!edgeLabelCountMap[label_1] || edgeLabelCountMap[label_1] < patternEdgeLabelMap[label_1].length) {
|
||
|
|
prunedByEdgeLabel = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
if (prunedByEdgeLabel) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
// 遍历 candidateEdges，进行边的筛选
|
||
|
|
var candidateEdgeNum = candidateEdges.length;
|
||
|
|
// prune：若边数过少，去除该图
|
||
|
|
if (candidateEdgeNum < pattern.edges.length) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "break";
|
||
|
|
}
|
||
|
|
var candidateGraphInvalid = false;
|
||
|
|
var _loop_5 = function _loop_5(e) {
|
||
|
|
var edge = candidateEdges[e];
|
||
|
|
var edgeLabel = edge[edgeLabelProp];
|
||
|
|
var patternEdgesWithLabel = patternEdgeLabelMap[edgeLabel];
|
||
|
|
// prune 1: 若边的 label 不存在于 pattern 边 label 中，去除该边
|
||
|
|
if (!patternEdgesWithLabel || !patternEdgesWithLabel.length) {
|
||
|
|
edgeLabelCountMap[edgeLabel]--;
|
||
|
|
// 若这个 label 的 count 减少之后，该 label 的边数不足，去除该图
|
||
|
|
if (patternEdgesWithLabel && edgeLabelCountMap[edgeLabel] < patternEdgesWithLabel.length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
return "break";
|
||
|
|
}
|
||
|
|
candidateEdges.splice(e, 1);
|
||
|
|
candidateNodeMap[edge.source].degree--;
|
||
|
|
candidateNodeMap[edge.target].degree--;
|
||
|
|
candidateNodeMap[edge.source].outDegree--;
|
||
|
|
candidateNodeMap[edge.target].inDegree--;
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
// prune 2: 若边的 label +两端 label 的三元组关系不能在 pattern 中找到，去除该边
|
||
|
|
var sourceLabel = candidateNodeMap[edge.source].node[nodeLabelProp];
|
||
|
|
var targetLabel = candidateNodeMap[edge.target].node[nodeLabelProp];
|
||
|
|
var edgeMatched = false;
|
||
|
|
patternEdgesWithLabel.forEach(function (patternEdge) {
|
||
|
|
var patternSource = patternNodeMap[patternEdge.source].node;
|
||
|
|
var patternTarget = patternNodeMap[patternEdge.target].node;
|
||
|
|
if (patternSource[nodeLabelProp] === sourceLabel && patternTarget[nodeLabelProp] === targetLabel) edgeMatched = true;
|
||
|
|
if (!directed && patternSource[nodeLabelProp] === targetLabel && patternTarget[nodeLabelProp] === sourceLabel) edgeMatched = true;
|
||
|
|
});
|
||
|
|
if (!edgeMatched) {
|
||
|
|
edgeLabelCountMap[edgeLabel]--;
|
||
|
|
// 若这个 label 的 count 减少之后，该 label 的边数不足，去除该图
|
||
|
|
if (patternEdgesWithLabel && edgeLabelCountMap[edgeLabel] < patternEdgesWithLabel.length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
return "break";
|
||
|
|
}
|
||
|
|
candidateEdges.splice(e, 1);
|
||
|
|
candidateNodeMap[edge.source].degree--;
|
||
|
|
candidateNodeMap[edge.target].degree--;
|
||
|
|
candidateNodeMap[edge.source].outDegree--;
|
||
|
|
candidateNodeMap[edge.target].inDegree--;
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
};
|
||
|
|
for (var e = candidateEdgeNum - 1; e >= 0; e--) {
|
||
|
|
var state_2 = _loop_5(e);
|
||
|
|
if (state_2 === "break") break;
|
||
|
|
}
|
||
|
|
// prune2: 删除边的过程中，发现边数过少/边 label 数过少时，去除该图
|
||
|
|
if (candidateGraphInvalid) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
candidateGraph.edges = candidateEdges;
|
||
|
|
var lengthsToCandidate = dijkstra(candidateGraph, candidateGraph.nodes[0].id, false).length;
|
||
|
|
Object.keys(lengthsToCandidate).reverse().forEach(function (targetId) {
|
||
|
|
if (targetId === candidateGraph.nodes[0].id || candidateGraphInvalid) return;
|
||
|
|
// prune4: 通过上述裁剪，可能导致该邻居子图变为不连通。裁剪掉目前在这个邻居子图中和 candidate（第一个节点）不连通的节点
|
||
|
|
if (lengthsToCandidate[targetId] === Infinity) {
|
||
|
|
var targetNodeLabel = candidateNodeMap[targetId].node[nodeLabelProp];
|
||
|
|
candidateNodeLabelCountMap[targetNodeLabel]--;
|
||
|
|
if (candidateNodeLabelCountMap[targetNodeLabel] < patternNodeLabelMap[targetNodeLabel].length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
return;
|
||
|
|
}
|
||
|
|
var idx = candidateGraph.nodes.indexOf(candidateNodeMap[targetId].node);
|
||
|
|
candidateGraph.nodes.splice(idx, 1);
|
||
|
|
candidateNodeMap[targetId] = undefined;
|
||
|
|
return;
|
||
|
|
}
|
||
|
|
// prune5: 经过边裁剪后，可能又出现了最短路径过长的节点 （比 pattern 中同 label 的节点到 beginNode 最大最短距离远），删去这些节点
|
||
|
|
var nLabel = nodeMap[targetId].node[nodeLabelProp];
|
||
|
|
if (!undirectedLengthsToBeginPNodeLabelMap[nLabel] || !undirectedLengthsToBeginPNodeLabelMap[nLabel].length || lengthsToCandidate[targetId] > undirectedLengthsToBeginPNodeLabelMap[nLabel][undirectedLengthsToBeginPNodeLabelMap[nLabel].length - 1]) {
|
||
|
|
var targetNodeLabel = candidateNodeMap[targetId].node[nodeLabelProp];
|
||
|
|
candidateNodeLabelCountMap[targetNodeLabel]--;
|
||
|
|
if (candidateNodeLabelCountMap[targetNodeLabel] < patternNodeLabelMap[targetNodeLabel].length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
return;
|
||
|
|
}
|
||
|
|
var idx = candidateGraph.nodes.indexOf(candidateNodeMap[targetId].node);
|
||
|
|
candidateGraph.nodes.splice(idx, 1);
|
||
|
|
candidateNodeMap[targetId] = undefined;
|
||
|
|
}
|
||
|
|
});
|
||
|
|
if (candidateGraphInvalid) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
var degreeChanged = true;
|
||
|
|
var loopCount = 0;
|
||
|
|
while (degreeChanged && !candidateGraphInvalid) {
|
||
|
|
degreeChanged = false;
|
||
|
|
// candidate 度数不足，删去该图
|
||
|
|
var condition = directed ? candidateNodeMap[candidate.id].degree < patternNodeMap[beginPNode.id].degree || candidateNodeMap[candidate.id].inDegree < patternNodeMap[beginPNode.id].inDegree || candidateNodeMap[candidate.id].outDegree < patternNodeMap[beginPNode.id].outDegree : candidateNodeMap[candidate.id].degree < patternNodeMap[beginPNode.id].degree;
|
||
|
|
if (condition) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
// candidate label 个数不足，删去该图
|
||
|
|
if (candidateNodeLabelCountMap[candidate[nodeLabelProp]] < patternNodeLabelMap[candidate[nodeLabelProp]].length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
// prune6：去除度数过小的节点
|
||
|
|
var currentCandidateNodeNum = candidateGraph.nodes.length;
|
||
|
|
for (var o = currentCandidateNodeNum - 1; o >= 0; o--) {
|
||
|
|
var cgNode = candidateGraph.nodes[o];
|
||
|
|
var nodeDegree = candidateNodeMap[cgNode.id].degree;
|
||
|
|
var nodeInDegree = candidateNodeMap[cgNode.id].inDegree;
|
||
|
|
var nodeOutDegree = candidateNodeMap[cgNode.id].outDegree;
|
||
|
|
var cNodeLabel = cgNode[nodeLabelProp];
|
||
|
|
var _e = stashPatternNodeLabelDegreeMap(minPatternNodeLabelDegreeMap, cNodeLabel, patternNodeMap, patternNodeLabelMap),
|
||
|
|
minPatternNodeLabelDegree = _e.minPatternNodeLabelDegree,
|
||
|
|
minPatternNodeLabelInDegree = _e.minPatternNodeLabelInDegree,
|
||
|
|
minPatternNodeLabelOutDegree = _e.minPatternNodeLabelOutDegree;
|
||
|
|
var deleteCondition = directed ? nodeDegree < minPatternNodeLabelDegree || nodeInDegree < minPatternNodeLabelInDegree || nodeOutDegree < minPatternNodeLabelOutDegree : nodeDegree < minPatternNodeLabelDegree;
|
||
|
|
if (deleteCondition) {
|
||
|
|
candidateNodeLabelCountMap[cgNode[nodeLabelProp]]--;
|
||
|
|
// 节点 label 个数不足
|
||
|
|
if (candidateNodeLabelCountMap[cgNode[nodeLabelProp]] < patternNodeLabelMap[cgNode[nodeLabelProp]].length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
candidateGraph.nodes.splice(o, 1);
|
||
|
|
candidateNodeMap[cgNode.id] = undefined;
|
||
|
|
degreeChanged = true;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
if (candidateGraphInvalid || !degreeChanged && loopCount !== 0) break;
|
||
|
|
// 经过 prune5 节点裁剪，删去端点已经不在 candidateGraph 中的边
|
||
|
|
candidateEdgeNum = candidateEdges.length;
|
||
|
|
for (var y = candidateEdgeNum - 1; y >= 0; y--) {
|
||
|
|
var cedge = candidateEdges[y];
|
||
|
|
if (!candidateNodeMap[cedge.source] || !candidateNodeMap[cedge.target]) {
|
||
|
|
candidateEdges.splice(y, 1);
|
||
|
|
var edgeLabel = cedge[edgeLabelProp];
|
||
|
|
edgeLabelCountMap[edgeLabel]--;
|
||
|
|
if (candidateNodeMap[cedge.source]) {
|
||
|
|
candidateNodeMap[cedge.source].degree--;
|
||
|
|
candidateNodeMap[cedge.source].outDegree--;
|
||
|
|
}
|
||
|
|
if (candidateNodeMap[cedge.target]) {
|
||
|
|
candidateNodeMap[cedge.target].degree--;
|
||
|
|
candidateNodeMap[cedge.target].inDegree--;
|
||
|
|
}
|
||
|
|
// 边 label 数量不足
|
||
|
|
if (patternEdgeLabelMap[edgeLabel] && edgeLabelCountMap[edgeLabel] < patternEdgeLabelMap[edgeLabel].length) {
|
||
|
|
candidateGraphInvalid = true;
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
degreeChanged = true;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
loopCount++;
|
||
|
|
}
|
||
|
|
if (candidateGraphInvalid) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
// prune: 若节点/边数过少，节点/边 label 过少，去掉这个图
|
||
|
|
if (candidateGraphInvalid || candidateGraph.nodes.length < pattern.nodes.length || candidateEdges.length < pattern.edges.length) {
|
||
|
|
candidateGraphs.splice(i, 1);
|
||
|
|
return "continue";
|
||
|
|
}
|
||
|
|
};
|
||
|
|
for (var i = candidateGraphNum - 1; i >= 0; i--) {
|
||
|
|
var state_1 = _loop_2(i);
|
||
|
|
if (state_1 === "break") break;
|
||
|
|
}
|
||
|
|
// 此时已经生成的多个 candidateGraphs，可能有重复
|
||
|
|
// console.log(
|
||
|
|
// "----- stage5: going to splice dulplicated candidate graphs -------"
|
||
|
|
// );
|
||
|
|
// 删去 candidateGraphs 中一模一样的子图，通过边的 node-node-edgeLabel 作为 key，这类边个数作为 value，进行匹配
|
||
|
|
var currentLength = candidateGraphs.length;
|
||
|
|
var _loop_3 = function _loop_3(i) {
|
||
|
|
var cg1 = candidateGraphs[i];
|
||
|
|
var cg1EdgeMap = {}; // [node1.id-node2.id-edge.label]: count
|
||
|
|
cg1.edges.forEach(function (edge) {
|
||
|
|
var key = "".concat(edge.source, "-").concat(edge.target, "-").concat(edge.label);
|
||
|
|
if (!cg1EdgeMap[key]) cg1EdgeMap[key] = 1;else cg1EdgeMap[key]++;
|
||
|
|
});
|
||
|
|
var _loop_6 = function _loop_6(j) {
|
||
|
|
var cg2 = candidateGraphs[j];
|
||
|
|
var cg2EdgeMap = {}; // [node1.id-node2.id-edge.label]: count
|
||
|
|
cg2.edges.forEach(function (edge) {
|
||
|
|
var key = "".concat(edge.source, "-").concat(edge.target, "-").concat(edge.label);
|
||
|
|
if (!cg2EdgeMap[key]) cg2EdgeMap[key] = 1;else cg2EdgeMap[key]++;
|
||
|
|
});
|
||
|
|
var same = true;
|
||
|
|
if (Object.keys(cg2EdgeMap).length !== Object.keys(cg1EdgeMap).length) {
|
||
|
|
same = false;
|
||
|
|
} else {
|
||
|
|
Object.keys(cg1EdgeMap).forEach(function (key) {
|
||
|
|
if (cg2EdgeMap[key] !== cg1EdgeMap[key]) same = false;
|
||
|
|
});
|
||
|
|
}
|
||
|
|
if (same) {
|
||
|
|
candidateGraphs.splice(j, 1);
|
||
|
|
}
|
||
|
|
};
|
||
|
|
for (var j = currentLength - 1; j > i; j--) {
|
||
|
|
_loop_6(j);
|
||
|
|
}
|
||
|
|
currentLength = candidateGraphs.length;
|
||
|
|
};
|
||
|
|
for (var i = 0; i <= currentLength - 1; i++) {
|
||
|
|
_loop_3(i);
|
||
|
|
}
|
||
|
|
return candidateGraphs;
|
||
|
|
};
|
||
|
|
export default GADDI;
|