Skip to content

Alias field type is not handled well in DF analytics. #50787

Closed
@przemekwitek

Description

@przemekwitek

With the recent change (#50219) that copies the mapping type from dependent variable to prediction field, QA regression has been found.

Wei posted the bug description:
The failure is related to handling DFA field alias. the failed test is to run an analytics against a field with alias, it passed on master before Dec11, and failed recently with this error:

"failure_reason" : """[dfa_wine_quality_red_alias_1578498260_000_0] Failed to join results: failures while writing results [failure in bulk execution:
[0]: index [dest_wine_quality_red_alias_1578516260980], id [-BZHR2wB9mzBfTtIWLPc], message [org.elasticsearch.index.mapper.MapperParsingException: failed to parse]

This is the job configuration:

{
  "id": "dfa_breast-cancer-alias_1578499743_000_0",
  "source": {
    "index": [
      "breast-cancer-alias"
    ],
    "query": {
      "match_all": {}
    }
  },
  "dest": {
    "index": "dest_breast_cancer_alias_1578517743585",
    "results_field": "ml"
  },
  "analysis": {
    "classification": {
      "dependent_variable": "class_alias",
      "num_top_classes": 2,
      "prediction_field_name": "class_alias_prediction",
      "training_percent": 100,
      "randomize_seed": 4381108523829301000
    }
  },
  "analyzed_fields": {
    "includes": [],
    "excludes": [
      "class",
      "breast-quad"
    ]
  },
  "model_memory_limit": "1gb",
  "create_time": 1578517744361,
  "version": "8.0.0",
  "allow_lazy_start": false
}

This is the mapping of breast-cancer-alias index:

"mappings" : {
      "properties" : {
        "age" : {
          "type" : "keyword"
        },
        "breast" : {
          "type" : "keyword"
        },
        "breast-quad" : {
          "type" : "keyword"
        },
        "breast-quad_alias" : {
          "type" : "alias",
          "path" : "breast-quad"
        },
        "class" : {
          "type" : "keyword"
        },
        "class_alias" : {
          "type" : "alias",
          "path" : "class"
        },
        "deg-malig" : {
          "type" : "long"
        },
        "inv-nodes" : {
          "type" : "keyword"
        },
        "irradiat" : {
          "type" : "keyword"
        },
        "menopause" : {
          "type" : "keyword"
        },
        "node-caps" : {
          "type" : "keyword"
        },
        "tumor-size" : {
          "type" : "keyword"
        }
      }
    },

Metadata

Metadata

Assignees

No one assigned

    Labels

    :mlMachine learning>bug

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions