Time Features Constructor

The Time Features Constructor node builds time–based features (shifts, rolling statistics, etc.) on top of any Orange data table. It usually sits between the data-source node and the nodes that will consume the enriched data set (for example Variable Dependency Graph, Save to DB or any learner).

Overview

  • Receives a data table (Data) and, optionally, a Variable Definitions table.

  • Lets you define new variables manually or by importing an existing Variable/Expression table.

  • Outputs the updated data table and the updated Variable Definitions table.

Inputs and outputs

  • Input – Data Mandatory. Original data on which features will be computed.

  • Input – Variable Definitions Optional. A two-column table (Variable, Expression) that lists expressions to evaluate.

  • Output – Data The original data plus every generated feature.

  • Output – Variable Definitions A full list of variables (original + derived) with their expressions.

Variable Definitions format

  • Variable – Name of the variable, original or derived.

  • Expression – Formula used to compute the variable. Original variables carry the value nan.

Typical workflow

  1. Connect data – Attach any data-providing node to Data. The data set must have a file/stream name so that new columns are tagged properly.

  2. Create features

    Automatic mode – Send a pre-built Variable/Expression table into the Variable Definitions input. The node evaluates every expression automatically.

    Manual mode

    • Choose New → Numeric to insert an empty row into Variables to generate.

    • Type the expression or use the drop-downs:

      – Variable chooser (original or already generated variables). – Operators +  -  *  /  % and parentheses. – Basic functions abs  int  float  pow. – Time functions listed below.

    • Edit, duplicate or delete pending variables as needed.

    • Press Send to evaluate every pending expression.

  3. Outputs – Pressing Send (or supplying an input table) triggers feature computation, table update and the emission of both outputs.

Time functions

  • shift(var, lag) – Shift var lag steps (negative or positive).

  • sum(var, inf, sup) – Sum of var from t – inf to t – sup.

  • min(var, inf, sup) – Minimum in the window.

  • max(var, inf, sup) – Maximum in the window.

  • sd(var, inf, sup) – Standard deviation in the window.

  • mean(var, inf, sup) – Arithmetic mean in the window.

  • count(var, inf, sup) – Number of non-null values in the window.

  • shiftBatch(var, n) – Convenience macro that generates n consecutive backward lags for var: t-1 … t-n. For example shiftBatch(x, 3) automatically creates

    • x_t-1 (same as shift(x, 1))

    • x_t-2 (shift(x, 2))

    • x_t-3 (shift(x, 3))

    Positive n therefore means “look n steps into the past”.

For all window functions inf and sup may be positive or negative; inf must be the smaller (earlier) offset.

Important notes

  • No nested window functions – You cannot write shift(mean(var,0,1),2). First create mean(var,0,1), then shift the new variable.

  • Duplicate names – The node auto-renames duplicates by appending _1, _2…

  • Invalid expressions – Highlighted in red and blocked until fixed.

  • Reset – Clears every pending and generated variable.

Keyboard shortcuts

  • Enter – Confirm expression editing.

  • Delete / Backspace – Remove the selected pending variable.

Summary

  • Requires Data; Variable Definitions is optional.

  • Generates shifts and rolling statistics (including shiftBatch).

  • Outputs an enriched data table and an updated definition table.

  • Manual editor with helper drop-downs and validation.

  • Press Send to compute all pending features.