CoxNetCVResult

Result of cross-validated penalizer selection for CoxNet.

Usage

CoxNetCVResult(
    penalizers,
    mean_scores,
    std_scores,
    n_nonzero,
    metric,
    l1_ratio,
    k,
)

Returned by cv_coxnet(). Stores the cross-validation scores at every tested penalizer value, identifies the best penalizer (highest concordance or lowest Brier score), and applies the one-standard-error rule for a more parsimonious alternative.

Penalizers are ordered from largest (most regularized, sparsest model) to smallest (least regularized, densest model).

Attributes

penalizers_: Penalizer values tested, sorted descending.
mean_scores_: Mean cross-validation score at each penalizer.
std_scores_: Standard deviation of per-fold scores at each penalizer.
n_nonzero_: Average number of non-zero coefficients at each penalizer (averaged over cross-validation folds).
metric_: Metric used: "concordance" or "brier".
l1_ratio_: Elastic-net mixing parameter fixed during the path search.
k_: Number of folds used.
best_penalizer_: Penalizer with the best mean cross-validation score.
best_score_: Mean score at best_penalizer_.
penalizer_1se_: Largest penalizer (most regularized) whose mean score is within one standard error of best_score_. Prefer this when parsimony matters.
score_1se_: Mean score at penalizer_1se_.

Examples

Run cross-validated penalizer selection and inspect the result:

import greenwood as gw

# Load data and write the model formula
lung = gw.load_dataset("lung", backend="polars")
formula = "Surv(time, status) ~ age + sex + ph.ecog + ph.karno + wt.loss"

# Run cross-validated penalizer selection
cv_result = gw.cv_coxnet(formula, data=lung, seed=23)
cv_result

CoxNetCV (concordance, ↑ higher is better, l1_ratio=1.0, 5-fold)

  best penalizer : 0.10767  (mean concordance: 0.6405)
  1-SE  penalizer : 0.18923  (mean concordance: 0.6193)

  50 penalizers tested, range [0.000218, 0.218]

The best penalizer and the more conservative one-standard-error penalizer are available as attributes:

# Retrieve the best penalizer value
cv_result.best_penalizer_

0.10767101882396797

# Retrieve the more conservative one-standard-error penalizer
cv_result.penalizer_1se_

0.189232959572064

Methods

Name	Description
to_frame()	Return the CV path as a tidy DataFrame, one row per penalizer.

to_frame()

Return the CV path as a tidy DataFrame, one row per penalizer.

Usage

Source

to_frame(
    *,
    format=None,
)

Parameters

format: str | None = None: Output format: None (default), "pandas", "polars", or "pyarrow". When None, a backend is auto-detected.

Returns

DataFrame: Columns: penalizer, mean_score, std_score, n_nonzero. Rows are ordered from largest to smallest penalizer.

Examples

Export the full CV path as a Polars DataFrame for plotting or further analysis:

import greenwood as gw

# Load data and run cross-validated penalizer selection
lung = gw.load_dataset("lung", backend="polars")
formula = "Surv(time, status) ~ age + sex + ph.ecog + ph.karno + wt.loss"
result = gw.cv_coxnet(formula, data=lung, seed=23)

# Export the CV path as a Polars DataFrame
result.to_frame(format="polars")

shape: (50, 4)

penalizer	mean_score	std_score	n_nonzero
f64	f64	f64	f64
0.217882	0.510069	0.022515	0.2
0.189233	0.619313	0.051879	1.6
0.164351	0.618485	0.052703	1.8
0.142741	0.63301	0.06026	2.2
0.123972	0.637076	0.063636	2.4
…	…	…	…
0.000383	0.625341	0.061453	5.0
0.000333	0.62533	0.061629	5.0
0.000289	0.62533	0.061629	5.0
0.000251	0.62533	0.061629	5.0
0.000218	0.62533	0.061629	5.0