✂ Split Dataset

Split a 3LC table into train, validation, and test sets — creating new table revisions with full lineage tracking.

📊

Stratified

Balanced class distribution

🔗

Full Lineage

EditedTable revisions

🎲

Reproducible

Seeded random splits

1 Select table

2 Set ratios

3 Split

Split Configuration

Source Table

Select the table you want to split into train/val/test sets.

Split Strategy

Stratified sampling ensures each split has the same class ratio as the original.

Split Ratio

70%

20%

10%

Train 70%

Validation 20%

Test 10%

Total: 100% — drag sliders to adjust. Train + Val + Test must equal 100%.

Random Seed

Fixed seed for reproducible splits. Change to get a different random ordering.

How It Works

EditedTable revisions: Each split creates a new table revision (EditedTable) that references the original rows — no data is duplicated.

Stratified sampling: When enabled, the split preserves the class distribution from the original table in each split. Requires a categorical label column.

DAG lineage: All split tables appear as children of the source table in the DAG view, making the relationship clear.