Tabular datasets

Tabular lives under category Other. Use it for supervised Advanced Training on CSV-style tables: several input columns and one target, plus a Task Type.

Enterprise only

Tabular (supervised) datasets require an Enterprise plan on the organization. See Tabular and non-tabular access.

Tabular datasets require an Enterprise plan on the organization. See Tabular and non-tabular access on the datasets overview.

Create

  1. New dataset → Other (available on an Enterprise plan) → Tabular

  2. Task Type — Regression, binary classification, or multiclass classification. Object detection is not offered here because it needs image files; use Non-tabular.

  3. Files — CSV or Hugging Face

Data tab

Task Type has its own dropdown with helper text: Select the type of machine learning task for this dataset.

Below that, Select Input and Target Columns:

  • Inputs — one or many columns

  • Target — a single column

You can change task type on this tab after create. Save inputs and target before preprocessing or training.

Preprocessing

On an Enterprise plan, open Preprocessing. Run Preprocess dataset, supply Encoder Code and an Encoder class, then Run Preprocessing. For tabular data the button stays disabled until inputs and target are set on Data.

A warning appears if you run preprocessing again: previous metrics and outputs will be overwritten.

Experiments

Attach the dataset on the Advanced Training Environment step. The wizard reviews columns, then offers Supervised on Agent. LatentPPO is not available for tabular datasets.

Tabular vs non-tabular

Tabular

Non-tabular

Data

Columns in a table file

Files and folders

Typical tasks

Regression, classification on numbers

Vision, encoders, detection

Object detection at create

No

Yes

Preprocessing tab

Yes (Enterprise only)

Yes (Enterprise only)