OpenSearch/docs/reference/aggregations/pipeline/inference-bucket-aggregation.asciidoc

[role="xpack"]
[testenv="basic"]
[[search-aggregations-pipeline-inference-bucket-aggregation]]
=== {infer-cap} Bucket Aggregation

A parent pipeline aggregation which loads a pre-trained model and performs 
{infer} on the collated result fields from the parent bucket aggregation.

To use the {infer} bucket aggregation, you need to have the same security 
privileges that are required for using the <<get-inference>>.

[[inference-bucket-agg-syntax]]
==== Syntax

A `inference` aggregation looks like this in isolation:

[source,js]
--------------------------------------------------
{
  "inference": {
    "model_id": "a_model_for_inference", <1>
    "inference_config": { <2>
      "regression_config": {
        "num_top_feature_importance_values": 2
      }
    },
    "buckets_path": {
      "avg_cost": "avg_agg", <3>
          "max_cost": "max_agg"
    }
  }
}
--------------------------------------------------
// NOTCONSOLE
<1> The ID of model to use.
<2> The optional inference config which overrides the model's default settings
<3> Map the value of `avg_agg` to the model's input field `avg_cost`


[[inference-bucket-params]]
.`inference` Parameters
[options="header"]
|===
|Parameter Name |Description |Required |Default Value
| `model_id`         | The ID of the model to load and infer against       | Required  | -
| `inference_config` | Contains the inference type and its options. There are two types: <<inference-agg-regression-opt,`regression`>> and <<inference-agg-classification-opt,`classification`>>  | Optional | -
| `buckets_path`     | Defines the paths to the input aggregations and maps the aggregation names to the field names expected by the model.
See <<buckets-path-syntax>> for more details | Required       | -
|===


==== Configuration options for {infer} models

The `inference_config` setting is optional and usually isn't required as the 
pre-trained models come equipped with sensible defaults. In the context of 
aggregations some options can overridden for each of the 2 types of model.

[discrete]
[[inference-agg-regression-opt]]
===== Configuration options for {regression} models

`num_top_feature_importance_values`::
(Optional, integer)
include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-regression-num-top-feature-importance-values]

[discrete]
[[inference-agg-classification-opt]]
===== Configuration options for {classification} models

`num_top_classes`::
(Optional, integer)
include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-classification-num-top-classes]

`num_top_feature_importance_values`::
(Optional, integer)
include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-classification-num-top-feature-importance-values]

`prediction_field_type`::
(Optional, string)
include::{es-repo-dir}/ml/ml-shared.asciidoc[tag=inference-config-classification-prediction-field-type]


[[inference-bucket-agg-example]]
==== Example 

The following snippet aggregates a web log by `client_ip` and extracts a number 
of features via metric and bucket sub-aggregations as input to the {infer} 
aggregation configured with a model trained to identify suspicious client IPs:

[source,console]
-------------------------------------------------
GET kibana_sample_data_logs/_search
{
  "size": 0,
  "aggs": {
    "client_ip": { <1>
      "composite": {
        "sources": [
          {
            "client_ip": {
              "terms": {
                "field": "clientip"
              }
            }
          }
        ]
      },
      "aggs": { <2>
        "url_dc": {
          "cardinality": {
            "field": "url.keyword"
          }
        },
        "bytes_sum": {
          "sum": {
            "field": "bytes"
          }
        },
        "geo_src_dc": {
          "cardinality": {
            "field": "geo.src"
          }
        },
        "geo_dest_dc": {
          "cardinality": {
            "field": "geo.dest"
          }
        },
        "responses_total": {
          "value_count": {
            "field": "timestamp"
          }
        },
        "success": {
          "filter": {
            "term": {
              "response": "200"
            }
          }
        },
        "error404": {
          "filter": {
            "term": {
              "response": "404"
            }
          }
        },
        "error503": {
          "filter": {
            "term": {
              "response": "503"
            }
          }
        },
        "malicious_client_ip": { <3>
          "inference": {
            "model_id": "malicious_clients_model",
            "buckets_path": {
              "response_count": "responses_total",
              "url_dc": "url_dc",
              "bytes_sum": "bytes_sum",
              "geo_src_dc": "geo_src_dc",
              "geo_dest_dc": "geo_dest_dc",
              "success": "success._count",
              "error404": "error404._count",
              "error503": "error503._count"
            }
          }
        }
      }
    }
  }
}
-------------------------------------------------
// TEST[skip:setup kibana sample data]

<1> A composite bucket aggregation that aggregates the data by `client_ip`.
<2> A series of metrics and bucket sub-aggregations.
<3> {infer-cap} bucket aggregation that contains the model ID and maps the 
aggregation names to the model's input fields.