Regroup Entities

Drop an entity column and aggregate the remaining rows with a formula — the equivalent of GROUP BY.

The Regroup Entities step removes an entity and regroups the data by the ones that remain, using a formula you choose to aggregate. It is the equivalent of GROUP BY in SQL or groupby in pandas.

Why entities need care

Entity columns are the columns required to uniquely identify a row. Because of that, a combination of entity values can never repeat — which makes removing or changing entities a delicate operation that can corrupt the data if done carelessly.

Say you have a dataset with Date and Country as entities, the most common combination in Alphacast. You will have many rows for the same date, one per country. You cannot simply drop the Country column, because dates would then repeat and entity combinations must be unique.

Regrouping is how you resolve this.

How to regroup

1. Create a pipeline and select the source

   Start from a Fetch Dataset step.

2. Add the Regroup Entities step
3. Choose which entities to drop

   Deselect the entities you want to remove.

4. Choose the aggregation per column

   Decide how rows that now share the same entity values should be combined — the Date, in the example above.

   The aggregation is set per column, so different columns can be combined differently in the same step. The right choice depends entirely on what each column represents.

   Aggregation	What it does
   sum	Total of the group’s values.
   mean / avg	Arithmetic mean, ignoring nulls.
   median	Median, ignoring nulls.
   min / max	Smallest or largest value in the group.
   count	Number of non-zero values.
   product	Product of the group’s values.
   std / var	Standard deviation or variance, ignoring nulls.
   first / last	First or last value in the group.
   meanw	Weighted mean. Requires a weight column.
   sumw	Weighted sum. Requires a weight column.
   drop	Discards the column instead of aggregating it.

   Weighted aggregations are the right choice when combining rates or ratios across entities — a weighted mean of inflation across countries needs a population or GDP weight column, since a plain mean would treat every country equally.

The result keeps every entity except the ones you excluded, and has fewer rows than the original because rows sharing entity values are collapsed together.

This step uses the fixed list of aggregations above, not the formula editor. To compute a value that isn’t in the list, add a Calculate Variable step before or after regrouping.

Next steps

Calculate variable

Compute values the fixed aggregations don’t cover.

Change frequency

Aggregate along the time dimension instead.