Bug: `transformed_range` is empty when `i_best_alpha == 0`

[tobias-zehnder]

Description

In the pcvl branch of alpha selection, transformed_range becomes an empty list when i_best_alpha == 0. This happens because pl_alphas[0] is the maximum value, causing np.argmax() to return 0. As a result, np.arange(i_best_alpha) generates an empty array, leading to a failure in np.argmax(transformed_pl_alphas).

Expected Behavior

The code should handle cases where i_best_alpha == 0 by skipping the transformation step and directly assigning best_alpha, best_l1_ratio, and best_pl_score from the original values.

Proposed Fix

Modify the code to check for i_best_alpha == 0 and handle it explicitly.

Code Reference

sparsesurv/sparsesurv/cv.py

Lines 553 to 565 in 90b5292

	transformed_range = np.arange(i_best_alpha)
	transformed_pl_alphas = pl_alphas[transformed_range] - (
	((pl_best_alpha - pl_alpha_max) / sparsity_best_alpha)
	* n_coefs[transformed_range]
	)
	transformed_i_best_alpha = np.argmax(transformed_pl_alphas)
	this_best_pl_transformed = transformed_pl_alphas[
	transformed_i_best_alpha
	]
	if this_best_pl_transformed > best_pl_score:
	best_alpha = l1_alphas[transformed_i_best_alpha]
	best_l1_ratio = l1_ratio
	best_pl_score = this_best_pl_transformed