phit

Data & Workflow

Statistics, Modeling, & Visualization

Ridge

Verification

Models

On this page

Numerics

Test suites

Scope

Coefficients, local-face effective df, conditional covariance, fitted values, weighted CV MSE and SE, and lambda selection.

References

Reference
glmnet, MASS, lmridge, and independent matrix calculations
Versions
R 4.6.1 (2026-06-24), glmnet 5.1, MASS 7.3.65, lmridge 1.2.2

Data

Inputs
Twelve fits and six CV cases including weighted day folds over hourly heating rows.

Method

Public glmnet output and independently computed conditional uncertainty, checked in the extension and polyfill.

Check
Comparison with a reference implementation

Tolerances

|x − r| ≤ 1e-7 + 2e-6 × |r|

Applies to: Numerical results

x: package value; r: reference value. Absolute error: |x − r|. Max errors are over the values a row compares.

Results

36 comparisons in 18 cases: 36 pass, 0 fail, 0 skipped.

By case

Results by case
Case Comparisons Largest abs. error Status
standardized 2 6.3e-11 pass
raw 2 4.5e-13 pass
weighted 2 2.7e-10 pass
no_intercept 2 7.6e-10 pass
no_intercept_raw 2 2.4e-10 pass
unpenalized 2 5.6e-12 pass
bounded 2 1.1e-13 pass
bounded_zero_lambda 2 1.1e-13 pass
factors_interactions 2 7.5e-10 pass
weighted_bounded 2 1.1e-10 pass
missing 2 3.2e-10 pass
heating 2 6.9e-8 pass
cv/small_folds_ungrouped 2 4.6e-8 pass
cv/unequal_folds_and_weights 2 1.7e-12 pass
cv/bounded_weighted 2 2.7e-12 pass
cv/no_intercept 2 3.9e-11 pass
cv/missing_original_rows 2 3.6e-12 pass
cv/day_folds_with_age_weights 2 5.1e-9 pass

Every comparison

Every comparison
Case Implementation Quantity Criterion Values Max abs. error Status
standardized native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 332 6.3e-11 pass
raw native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 4.5e-13 pass
weighted native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 2.7e-10 pass
no_intercept native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 319 7.6e-10 pass
no_intercept_raw native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 319 2.4e-10 pass
unpenalized native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 5.6e-12 pass
bounded native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 313 1.1e-13 pass
bounded_zero_lambda native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 321 1.1e-13 pass
factors_interactions native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 345 7.5e-10 pass
weighted_bounded native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 321 1.1e-10 pass
missing native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 325 3.2e-10 pass
heating native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 2,125 6.9e-8 pass
cv/small_folds_ungrouped native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 4.6e-8 pass
cv/unequal_folds_and_weights native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 1.7e-12 pass
cv/bounded_weighted native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 2.7e-12 pass
cv/no_intercept native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 3.9e-11 pass
cv/missing_original_rows native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 3.6e-12 pass
cv/day_folds_with_age_weights native Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 29 5.1e-9 pass
standardized polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 332 6.3e-11 pass
raw polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 4.5e-13 pass
weighted polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 2.7e-10 pass
no_intercept polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 319 7.6e-10 pass
no_intercept_raw polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 319 2.4e-10 pass
unpenalized polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 331 5.6e-12 pass
bounded polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 313 1.2e-14 pass
bounded_zero_lambda polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 321 1.1e-13 pass
factors_interactions polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 345 7.5e-10 pass
weighted_bounded polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 321 1.1e-10 pass
missing polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 325 3.2e-10 pass
heating polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 2,125 6.9e-8 pass
cv/small_folds_ungrouped polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 4.6e-8 pass
cv/unequal_folds_and_weights polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 1.7e-12 pass
cv/bounded_weighted polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 2.7e-12 pass
cv/no_intercept polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 3.9e-11 pass
cv/missing_original_rows polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 34 3.6e-12 pass
cv/day_folds_with_age_weights polyfill Ridge estimates and CV error |x − r| ≤ 1e-7 + 2e-6 × |r| 29 5.1e-9 pass

Limitations

Known differences

  • Explicit Gaussian lambda is divided by the weighted response scale to match glmnet output.
  • Conditional covariance uses RSS/(n-df); lmridge uses a different residual-variance denominator, so covariance per sigma squared is compared separately.
  • Default lambda paths are phit policy; explicit paths are used for glmnet comparisons.

Reproduction

From the repository root 
Rscript docs/verifications/ridge.R
python3 docs/verifications/compare_ridge.py
Raw record
site/verifications/ridge.json
Full results
docs/verifications/ridge-results.json
Input data
tests/fixtures/ridge-r.json