feature-calculator

name	feature-calculator
description	Calculate derived columns and financial metrics from transaction data — revenue, profit margin, time-based features, moving averages, and growth rates.
version	0.7.0

Feature Calculator

Skill Summary

Computes derived features and financial metrics from raw transaction data. Handles common calculations like revenue (quantity × price), profit margin (using cost lookup), time-based feature extraction (month, weekday, week number), rolling window aggregates (moving averages), and period-over-period growth rates. Designed for retail/e-commerce transaction pipelines.

Available Calculations

• revenue: quantity * unit_price
• cost_total: quantity * unit_cost (requires cost lookup)
• margin: revenue - cost_total
• margin_pct: margin / revenue * 100
• time_features: Extract year, month, weekday, week_number from a date column
• moving_avg: Rolling N-day moving average of a numeric column
• growth_rate: Period-over-period percentage change
• cumulative_sum: Running cumulative total
• rank: Rank within a group (e.g., rank products by revenue)

Inputs

• --input / -i: Path to transaction data (CSV or Parquet) (required)
• --features: Comma-separated list of features to calculate (default: all)
• --cost-lookup: Path to product catalog/cost lookup JSON
• --date-column: Column name containing dates (default: date)
• --window: Rolling window size in days for moving average (default: 7)
• --group-by: Column for group-level calculations
• --output / -o: Output path

Processing Steps

1. Load transaction data
2. Parse date column to datetime
3. Calculate requested features:
   • Arithmetic: revenue, cost, margin
   • Temporal: extract date components
   • Rolling: compute moving averages per group
   • Growth: calculate period-over-period changes
4. Join with cost lookup if margin calculations requested
5. Save enriched dataset

Output

• Enriched dataset with new columns appended (CSV or Parquet)
• feature_summary.json: Statistics for each computed feature (min, max, mean, null count)
• Console: List of features calculated, row count

Implementation

• Command: python3 ./skills/feature_calculator/calculate.py -i <file> -o <output> --features <list>
• Dependencies: pandas, numpy