cc() truncates periodic coordinates for integer input with fractional bounds

Open
#272 0 comments 0 reactions 0 assignees View on GitHub

@ktz03 is already working on this.

Since Oct 12, 2026.

  • #273 by @ktz03 — open

Assessment

Difficulty
2/5
Estimated time
1-3 hours
Newbie friendliness
35/100
Issue type
Bug
Clarity
Clearly specified
Activity status
Stale
Tech stack
numpy, python
Domain
data

Research direction

Start with _map_cyclic() in the cyclic spline implementation and trace how it handles integer coordinates with fractional bounds; compare that path with the natural spline cr() control described in the issue. Reproduce the provided direct cc() and DesignInfo prediction examples with integer and float inputs, then verify both produce matching bases while preserving input immutability.

Written by the indexing model from the issue text.

Description

cc() silently truncates periodic coordinates when its input has an integer dtype and its exterior bounds are non-integral. The same numeric values produce different spline bases when passed as integers and floats, including when predicting with an existing DesignInfo.

Reproduced with the latest PyPI release, Patsy 1.0.3, and current master 8ce7441586fc228b9ff487a97e3a6ebe9fde6aad on Windows with Python 3.13.15, NumPy 2.5.4 and SciPy 1.18.1:

import numpy as np
from patsy import cc, dmatrix, build_design_matrices

x = np.arange(-1, 6)
kwargs = dict(knots=[1., 2., 3.], lower_bound=0.1, upper_bound=3.4)
integer_basis = cc(x, **kwargs)
float_basis = cc(x.astype(float), **kwargs)
print(np.max(np.abs(integer_basis - float_basis)))
# 1.1751149047931713 (expected: 0)

formula = "0 + cc(x, knots=[1., 2., 3.], lower_bound=0.1, upper_bound=3.4)"
training = dmatrix(formula, {"x": np.array([1., 2., 3.])})
integer_prediction = build_design_matrices([training.design_info], {"x": x})[0]
float_prediction = build_design_matrices([training.design_info], {"x": x.astype(float)})[0]
print(np.max(np.abs(integer_prediction - float_prediction)))
# 1.1751149047931713 (expected: 0)

For example, the period is 3.3, so x=4 maps to approximately 0.7. _map_cyclic() writes that coordinate into a copy of the integer input and truncates it to zero; its subsequent lower-bound mapping then moves it to another incorrect coordinate. The returned design matrix contains no warning about this loss of precision.

Integer and float input agree for points inside the bounds, for whole-number exterior bounds, and for the corresponding natural spline cr() control. A fix can convert integer coordinates to floating point before periodic mapping while preserving the existing floating-point input dtype and input immutability.

Dominant language
Python
Stars
990
Forks
113
Avg merge
2d 8h
Merged PRs (30d)
3

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from pydata/patsy

All issues in pydata/patsy

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.