AHC

Classification Ascendante Hiérarchique(CAH) is a type of clustering algorithm that builds a hierarchy of clusters by successively merging smaller clusters into larger ones.

Representations

Keys _id, name, stepId, domainInformation, dataset, project, processingInfo, creationTS, latestUpdateTS are unchanged and follow the classic of representation only dataSpecification keys are changed and are described as above:

dataSpecification :
- keyword : Its value is “hardClustering” for the representation of hardClustering and “prototypes” for the representation of prototypes **. Cf to mandatory keys.
- valueType : It is the type of processing data values and then depends on them. Cf to mandatory keys.
- meaning: Its value is “hierarchical clustering with bisecting Kmeans”. Cf to mandatory keys.
- view: Cf to mandatory keys.
- dataLocationId: Cf to mandatory keys. Currently, only numerical vector can be saved during raw data loading which gives us un single template at this moment.

JSON template for numerical vector processing data representation.

Representation of prototypes:

{
  "dataSpecification": {
    "keyword": "prototypes",
    "valueType": {
      "datatype": "numericalData",
      "structureType": "Hierarchical Clustering"
    },
    "meaning": "hierarchical clustering with bisecting Kmeans",
    "view": {
      "name": "view_02-02-2023_11:07:45",
      "id": "63db99816b332aad216c21e6"
    },
    "dataLocationId": "63db9a82a52c4511df13c8f5",
    "hyperParameters": {
      "maxNbClusters": 8,
      "numberOfClusters": 2,
      "seed": 1
    }
  }
}

Representation of hardClustering:

{
  "dataSpecification": {
    "keyword": "hardClustering",
    "valueType": {
      "datatype": "numericalData",
      "structureType": "Hierarchical Clustering"
    },
    "meaning": "hierarchical clustering with bisecting Kmeans",
    "view": {
      "name": "view_02-02-2023_11:07:45",
      "id": "63db99816b332aad216c21e6"
    },
    "dataLocationId": "63db9a83a52c4511df13c902",
    "hyperParameters": {
      "maxNbClusters": 8,
      "numberOfClusters": 2,
      "seed": 1
    }
  }
}

Observations

We save the observations of prototypes in mongoDB:

{
  "_id": "ObjectId(63db9a83a52c4511df13c8f9)",
  "clustersNumber": 3,
  "prototypes": [
    {
      "clusterId": 0,
      "prototype": [
        {
          "columnName": "cp_a_1_bar",
          "value": 5.632708722788574e-17
        },
        {
          "columnName": "cp_a_2_bar",
          "value": -3.2857467549600007e-16
        },
        {
          "columnName": "cp_r_bar",
          "value": 7.510278297051431e-17
        }
      ]
    }
  ],
  "representationId": "ObjectId(63db9a82a52c4511df13c8f2)"
}

We save the observations of hardClustering in parquet: