Skip to content

Sessions and the Data Model

TabDat operates around an in-memory session model designed for speed, safety, and predictability.

The Active Dataset

At any given time, a TabDat session maintains one active dataset. All standard inspection, transformation, modeling, and plotting commands operate directly on this active dataset.

When you execute a transformation (such as generate, replace, keep, drop, or rename), TabDat updates the active dataset relation:

tabdat> use patients.parquet
Loaded: patients.parquet (1200 rows, 5 columns)

tabdat> generate bmi = weight / (height / 100)^2
Generated: bmi (DOUBLE)

Session-Local Named Tables

In addition to the active dataset, a session can maintain named tables in memory.

Creating Named Tables via SQL

Use the into <table> clause in SQL commands to store query results into a named table:

tabdat> sql select sex, avg(bmi) as mean_bmi from active group by sex into summary_by_sex
Created summary_by_sex: 2 rows, 2 columns

When created, the new named table immediately becomes the active dataset.

Switching Between Tables

You can switch back to any named table using use <table>:

tabdat> use summary_by_sex
Activated: summary_by_sex (2 rows, 2 columns)

Named tables exist purely in memory for the duration of the CLI session. To persist them, use save or export.

Panel Metadata

Panel data structures are defined using panel <id_var> <time_var>:

tabdat> panel id year
Panel declared: id=id, time=year (balanced panel: 500 units x 10 periods)
  • Integrity Requirements: The (id, time) pair must have zero missing values and must uniquely identify every row.
  • Scope: Panel metadata is session-local and retained in memory.
  • Dependent Commands: Panel-aware commands such as xtreg, xtdata, xtlogit, xtabond, and panel did require active panel metadata.
  • Clearing Panel Metadata: Run panel clear to remove panel declarations.