"Greet each day with a smile, so it doesn't know what you're planning on doing to it." - Wayne

Formula to Model Matrix in One Function Call

wyane is a Python package that converts statistical formulas into model matrices using Polars DataFrames. It does one thing: takes a formula and a DataFrame, returns a model matrix.

Requirements: Python 3.13+ (due to fiasto-py dependency)

uv pip install wayne-trade
# or
pip install wayne-trade

import wayne
import polars as pl

# Load your data
df = pl.read_csv("data/mtcars.csv")

# Define your formula
formula = 'mpg ~ cyl + wt*hp + poly(disp, 4) - 1'

# Get the model matrix
model_matrix = wayne.trade_formula_for_matrix(df, formula)
print(model_matrix)

Wayne provides a clean interface to fiasto-py's parsing functionality without requiring users to directly import fiasto-py and deal with its maturin/pyo3 complexity:

import wayne

# Parse a formula to get detailed structure
result = wayne.speak_fiasto("mpg ~ cyl + wt*hp + poly(disp, 4) - 1")
print(result['columns'].keys())  # All variables and their metadata
print(result['metadata']['has_intercept'])  # Formula metadata
print(result['all_generated_columns'])  # All generated column names

• Single Function: trade_formula_for_matrix(df, formula)
• Polars Integration: Works with Polars DataFrames
• R-Style Formulas: Supports R/Wilkinson notation
• Fast Parsing: Rust + fiasto for formula parsing
• Orthogonal Polynomials: NumPy-based polynomial generation
• Interactions: Handles interaction terms (e.g., x*z)
• Intercept Control: Add/remove intercept with +1/-1

Wayne supports R-style statistical formulas:

• Basic terms: y ~ x + z
• Interactions: y ~ x*z (expands to x + z + x:z)
• Polynomial terms: y ~ poly(x, 3) (generates 3 orthogonal polynomial columns)
• Intercept control: y ~ x - 1 (removes intercept)
• Complex formulas: y ~ x + z + poly(w, 3) + x:z - 1

See the examples/ directory for complete examples:

# Main example
uv run python examples/final_example.py

# Single function demo
uv run python examples/single_function_example.py

The model matrix contains:

• Main effects: Original variables from your formula
• Interaction terms: Generated interaction columns (e.g., wt_hp for wt*hp)
• Polynomial terms: Orthogonal polynomial columns (e.g., disp_poly_1, disp_poly_2, etc.)

For developers and maintainers:

• Release Process: See RELEASE.md for how to release new versions to PyPI
• Automatic Releases: Pushing changes to CHANGELOG.md triggers automatic PyPI releases
• Testing: Run uv run -m pytest to run the test suite
• Polynomial Validation: Run uv run compare_polynomials.py to verify R compatibility
• Intercept: Optional intercept column

Wayne creates model matrices that are ready for:

• Linear regression
• Generalized linear models
• Mixed effects models
• Any statistical modeling that needs a design matrix

MIT