Variable Dependency Graph

The Variable Dependency Graph widget converts a Variable / Expression table (produced by Time Features Constructor) into an interactive directed network that highlights how every generated feature depends on the rest of the data set.

Overview

  • Input – a single Orange Table whose first attribute column is Variable and whose first meta column is Expression (each row comes straight from Time Features Constructor).

  • Output – a Network object (orange-network package) whose nodes represent variables and whose edges represent dependencies discovered in the expressions.

  • Typical use-case – plug the table emitted by Time Features Constructor into Variable Dependency Graph and then feed the resulting network to Network Explorer for visual analysis.

What the widget does

  1. Parse every expression Each row is scanned for references to other variables (exact matches only – whitespace and dashes are normalised).

  2. Detect temporal shifts If an expression contains shift(var, ±k) the edge between var and the derived feature is annotated with the lag k. Positive values are shown as 1, 2, … (no sign).

  3. Build a sparse adjacency matrix Edges carry the numerical lag (default = 1 when no shift is present).

  4. Create the `Network` object

    • edge_labels – string version of the lag ("1", "3", …), ready for display in Network Explorer (turn on Show edge weights).

    • node meta fields - var_name → the column name displayed as the node label. - var_type → Derived or Original (can be mapped to colour).

User interface

  • Generate button – forces a refresh (table changes are detected automatically, but you can click to be sure).

  • No additional settings: the widget is intentionally minimal – all styling, filtering and layout happen in Network Explorer.

Practical tips

  • When you open the network in Network Explorer: - Set Label → var_name to show readable names. - Set Colour → var_type to distinguish originals (orange) from derived features (blue). - Enable Show edge weights to see the temporal lags on the arrows.

  • The widget ignores expressions that reference variables not present in the table – useful when you experiment with placeholder names.