用聚合填充 MongoDB 中二维时间序列数据的天空白

如何解决用聚合填充 MongoDB 中二维时间序列数据的天空白

我有一个二维时间序列数据的集合如下：

[
  { 
    "value" : 9,"timestamp" : "2020-12-30T02:06:33.000+0000","recipeId" : 15
  },{ 
    "value" : 2,"timestamp" : "2020-12-30T12:04:23.000+0000","recipeId" : 102
  },{ 
    "value" : 5,"timestamp" : "2020-12-30T15:09:23.000+0000",...
]

记录有一个 recipeId，这是我正在寻找的第一级分组。应该总结一天食谱的所有 values。我想要每个 recipeId 的时间序列数组。我需要用 0 填补缺失的日子。我希望为提供的开始和结束日期范围创建此构造。

对于 2020-12-29 到 2020-12-31 的日期范围，有些人喜欢这样：

[
 [
  { 
    "sum" : 0,"timestamp" : "2020-12-29",{ 
    "sum" : 9,"timestamp" : "2020-12-30",{ 
    "sum" : 0,"timestamp" : "2020-12-31",...
 ],[
  { 
    "sum" : 0,"recipeId" : 0
  },{ 
    "sum" : 7,...
 ]
]

这是我目前所拥有的，它只能部分解决我的要求。我无法正确完成最后几个阶段：

[
  {
    "$match": {
      "timestamp": {
        "$gte": "2020-12-29T00:00:00.000Z","$lte": "2020-12-31T00:00:00.000Z"
      }
    }
  },{
    "$addFields": {
      "timestamp": {
        "$dateFromParts": {
          "year": { "$year": "$timestamp" },"month": { "$month": "$timestamp" },"day": { "$dayOfMonth": "$timestamp" }
        }
      },"dateRange": {
        "$map": {
          "input": {
            "$range": [
              0,{
                "$trunc": {
                  "$divide": [
                    {
                      "$subtract": [
                        "2020-12-31T00:00:00.000Z","2020-12-29T00:00:00.000Z"
                      ]
                    },1000
                  ]
                }
              },86400
            ]
          },"in": {
            "$add": [
              "2020-12-29T00:00:00.000Z",{ "$multiply": ["$$this",1000] }
            ]
          }
        }
      }
    }
  },{ "$unwind": "$dateRange" },{
    "$group": {
      "_id": { "date": "$dateRange","recipeId": "$recipeId" },"count": {
        "$sum": { "$cond": [{ "$eq": ["$dateRange","$timestamp"] },1,0] }
      }
    }
  },{
    "$group": {
      "_id": "$_id.date","total": { "$sum": "$count" },"byRecipeId": {
        "$push": {
          "k": { "$toString": "$_id.recipeId" },"v": { "$sum": "$count" }
        }
      }
    }
  },{ "$sort": { "_id": 1 } },{
    "$project": {
      "_id": 0,"timestamp": "$_id","total": "$total","byRecipeId": {
        "$arrayToObject": {
          "$filter": { "input": "$byRecipeId","cond": "$$this.v" }
        }
      }
    }
  }
]

导致：

[
    {
        "timestamp": "2020-12-29T00:00:00.000Z","total": 21,"byRecipeId": {}
    },{
        "timestamp": "2020-12-30T00:00:00.000Z","total": 0,"byRecipeId": {
            "15": 9,"102": 7
        }
    },{
        "timestamp": "2020-12-31T00:00:00.000Z","byRecipeId": {}
    }
]

我当然愿意接受替代解决方案。例如，我看到了这篇文章：https://medium.com/@alexandro.ramr777/fill-missing-values-using-mongodb-aggregation-framework-f011114e83e0 但它不涉及多维。

解决方法

您可以使用 $redcue 函数。此代码填补了当天的分钟数。应该很容易调整它以提供缺失的日子。

{
   $addFields: {
      data: {
         $reduce: {
            input: { $range: [0,24 * 60] },initialValue: [],in: {
               $let: {
                  vars: { 
                     ts: { 
                        $add: [
                           moment().startOf('day').toDate(),{ $multiply: ["$$this",1000 * 60] }
                        ] 
                     } 
                  },in: {
                     $concatArrays: [
                        "$$value",[{
                           $cond: {
                              if: { $in: ["$$ts","$data.timestamp"] },then: { 
                                 $first: { 
                                    $filter: { 
                                       input: "$data",cond: { $eq: ["$$this.timestamp","$$ts"] } 
                                    } 
                                 } 
                              },else: { timestamp: "$$ts",total: 0 }
                           }
                        }]
                     ]
                  }
               }
            }
         }
      }
   }
}

在我看来，$reduce 比 $map 更优雅，但根据我的经验，$reduce 的性能要差得多。