Table Tools
Overview
core.table_tools performs two common table operations without requiring
a separate notebook: calculate a new column from an expression, or create a new
dataset from rows matching a boolean condition.
What You Do
To add a derived column:
Open Table Transform and review the available columns.
Enter an expression and preview the result on a bounded number of rows.
Choose a new column name.
Apply the expression and register the transformed dataset.
To create a subset:
Enter a boolean expression.
Preview how many rows match and inspect a sample.
Choose a dataset ID and display name.
Register the subset and optionally make it active.
Panel
Table Transform
The panel brings together column inspection, expression preview and the two registered transformations. Previewing first helps identify missing columns, invalid syntax and unexpectedly broad filters before creating a full result.
Actions
Add Derived Column
Add Derived Column evaluates an expression and adds its result under a new column name. Existing columns are protected from accidental overwrite.
DuckDB-backed sources can retain a lazy relational expression. Sources that require pandas materialisation are written through the Parquet compatibility path so that the resulting dataset can be accessed lazily afterwards.
Create Subset Dataset
Create Subset Dataset evaluates a boolean condition and registers only the matching rows. Existing semantic mappings are copied to the derived dataset where their columns remain available.
Expression Examples
Create a colour-like derived value:
mag_g - mag_r
Create a high-score, low-redshift subset:
(score > 0.9) & (redshift < 1.0)
Use explicit parentheses when combining conditions so that the intended order is clear.
Outputs and Provenance
Both operations produce registered datasets rather than changing another plugin’s private dataframe. This allows the result to appear in the global dataset selector and participate in mappings, selections, persistence and later actions.
Caution
Expressions are powerful and a syntactically valid result can still be scientifically wrong. Always review the preview, null behaviour, data type and output row count before using a transformed dataset for training or publication.