feat: add nodesCosineSimilarity算法-基于节点属性计算余弦相似度 #45

xdddst · 2021-12-15T09:18:34Z

No description provided.

Yanyan-Wang · 2021-12-16T01:55:22Z

packages/graph/src/nodes-cosine-similarity.ts

+      // 计算节点向量和种子节点向量的点积
+      const dot = seedNodeVector.dot(nodeVector);
+      // 计算节点向量和种子节点向量的余弦相似度
+      const cosineSimilarity = dot / (seedNodeNorm2 * nodeNorm2);


可能除零

Yanyan-Wang · 2021-12-16T01:56:31Z

packages/graph/src/utils/node-properties.ts

+            !`${node.properties[propertyKey]}`.match(dateReg) && 
+            isNaN(Number(node.properties[propertyKey])) || propertyKey === 'id') {
+            if (propertyKeyInfo.hasOwnProperty(propertyKey)) {
+            delete propertyKeyInfo[propertyKey];


Yanyan-Wang · 2021-12-16T01:57:19Z

packages/graph/src/utils/node-properties.ts

+})
+
+// 取top50的属性
+const sortKeys = Object.keys(propertyKeyInfo).sort((a,b)=>{


=> 前后空格

sort((a, b) => propertyKeyInfo[b] - propertyKeyInfo[a])

Yanyan-Wang · 2021-12-16T01:57:33Z

packages/graph/src/utils/node-properties.ts

+const sortKeys = Object.keys(propertyKeyInfo).sort((a,b)=>{
+    return propertyKeyInfo[b] - propertyKeyInfo[a];
+});
+return sortKeys.length < n ? sortKeys : sortKeys.slice(0, n);


好多地方都有问题，都检查一下，不一一评论了

Yanyan-Wang · 2021-12-16T01:57:53Z

packages/graph/src/utils/node-properties.ts

+
+// 获取所有属性并排序
+export const getAllSortProperties = (nodes: NodeConfig[] = [], n: number = 100) => {
+    const propertyKeyInfo = {};


这个文件的所有缩进不是两格？

缩进还是有问题

Yanyan-Wang · 2021-12-16T01:59:02Z

packages/graph/src/utils/vector.ts

+      return 0;
+    }
+    let res = 0;
+      for(let key in this.arr) {


for 后面加空格

Yanyan-Wang · 2021-12-16T01:59:57Z

packages/graph/tests/unit/cosineSimilarity-spec.ts

+import propertiesGraphData from './data/cluster-origin-properties-data.json';
+
+describe('nodesCosineSimilarity', () => {
+  it('nodesCosineSimilarity simple demo: ', () => {


最好多1-2个单测，以后改的时候有更强参考性

Yanyan-Wang · 2021-12-16T02:01:15Z

packages/graph/src/utils/node-properties.ts

+        // 目前过滤只保留可以转成数值型的或日期型的, todo: 统一转成one-hot特征向量或者embedding
+        if (!`${node.properties[propertyKey]}`.match(secondReg) && 
+            !`${node.properties[propertyKey]}`.match(dateReg) && 
+            isNaN(Number(node.properties[propertyKey])) || propertyKey === 'id') {


propertyKey === 'id' 和上面所有 && 是同级的？那把这个判断放到最前面，可以有更好性能

Yanyan-Wang · 2021-12-16T03:58:50Z

packages/graph/src/cosine-similarity.ts

+  // 计算元素向量和目标元素向量的点积
+  const dot = targetItemVector.dot(itemVector);
+  // 计算元素向量和目标元素向量的余弦相似度
+  const cosineSimilarity = seedNodeNorm2 * itemNorm2 ? dot / (seedNodeNorm2 * itemNorm2) : 0;


seedNodeNorm2 * itemNorm2 被计算了两次

Yanyan-Wang · 2021-12-16T03:59:36Z

packages/graph/src/utils/node-properties.ts

+
+// 获取所有属性并排序
+export const getAllSortProperties = (nodes: NodeConfig[] = [], n: number = 100) => {
+    const propertyKeyInfo = {};


缩进还是有问题

Yanyan-Wang · 2021-12-16T03:59:50Z

packages/graph/src/utils/vector.ts

+      return 0;
+    }
+    let res = 0;
+      for(let key in this.arr) {


for 后面加空格

Yanyan-Wang · 2021-12-16T04:00:55Z

CHANGELOG.md


+- feat: add consine-similarity algorithm and nodes-consine-similarity algorithm;


下面加一个空行

…gorithm feat: add consine-similarity algorithm and perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code

xdddst · 2021-12-16T04:23:05Z

已修改

Yanyan-Wang reviewed Dec 16, 2021

View reviewed changes

xdddst force-pushed the feat/add-cosine-similarity branch 3 times, most recently from e8bcc1e to cc29aad Compare December 16, 2021 03:58

Yanyan-Wang reviewed Dec 16, 2021

View reviewed changes

xdddst force-pushed the feat/add-cosine-similarity branch 3 times, most recently from b4e35c1 to aaf8b36 Compare December 16, 2021 04:15

feat: add console-similarity algorithm and nodes-cosine-similarity al…

51eb973

…gorithm feat: add consine-similarity algorithm and perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code chore: perf code

xdddst force-pushed the feat/add-cosine-similarity branch from 6206cb3 to 51eb973 Compare December 16, 2021 04:18

Yanyan-Wang approved these changes Dec 16, 2021

View reviewed changes

Yanyan-Wang merged commit 646a2bc into master Dec 16, 2021

Yanyan-Wang deleted the feat/add-cosine-similarity branch December 16, 2021 07:03


		- feat: add consine-similarity algorithm and nodes-consine-similarity algorithm;

feat: add nodesCosineSimilarity算法-基于节点属性计算余弦相似度 #45

feat: add nodesCosineSimilarity算法-基于节点属性计算余弦相似度 #45

Uh oh!

Conversation

xdddst commented Dec 15, 2021

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

xdddst commented Dec 16, 2021

Uh oh!

Uh oh!