Use case: predict the time of the next purchase about elasticsearch-linear-regression HOT 1 CLOSED

Hi Milan,
to calculate the "leg" field when new events arrive isn't possible using pure Elasticsearch functionality since features like triggers in SQL are not yet provided. Thus you should consider to have 3rd party tools in place which perform data preprocessing before indexing to Elasticsearch.
To avoid calculating the forecast for each User/Product you could use sub aggregates which would group the pairs for you automatically e.g.

/buying_habit/_search?size=0
{
    "aggs": {
        "users": {
            "terms": {
                "field": "user.keyword",
                "size": 1000
            },
            "aggs": {
                "products":
                {
                    "terms": {
                        "field": "product.keyword",
                        "size": 100
                    },
                    "aggs": {
                        "demand_p": {
                            "linreg_predict": {
                                "fields": ["quantity", "lag"],
                                "inputs": [2]
                            }
                        }
                    }
                }
            }
        }
    }
}

This query should calculate the forecast for the top 1000 buyers and for the top 100 products these bought in the past. If this solution works for you and you need the forecast for all users, you would have to execute the query multiple times each for a user partition since increasing the size of the terms aggregation would end in a long runtime. Partitioning users could be achieved e.g. by filtering by user first name character.
If you need more support don't hesitate to contact me via mail: [email protected]. I home it's OK if I close this issue.

from elasticsearch-linear-regression.