DOCUMENTATION

Hand a SQL result to the ML Lab

Filter, join or aggregate in SQL, then train on the result — with no intermediate file.

The gesture

  1. Open the Data Studio and load a file.
  2. In the SQL console, write your query. The active dataset is queryable by name, and you can attach other files — CSV, Parquet or JSON.
  3. Once the result is right, click Send to the ML Lab.

The result becomes the lab's active dataset. Nothing is written to disk on the way.

Open the Data Studio

## When this is the right tool
  • Filter before training: keep one region, one period, one segment.
  • Join several files on a key, when the Data Studio's own join is not enough.
  • Aggregate: go from one row per event to one row per customer, which is what most real ML problems need.

The cost, announced

The SQL engine is an 18–22 MB download, optional. Until you open it, it is not downloaded. It is single-threaded: LabML does not serve the headers that would unlock multi-threading.

That is also why cross-column consistency checks do not go through SQL: they have to work for everyone, including people who never open the console.

Where to go next