NEWS

---

swjm 0.1.3

• Fixed the adaptive step-size rule. The step was previously derived from count_digits(), which counted decimal places rather than magnitude. Two consequences: a dual norm of 1.27 was scored like 0.5 and received a step ten times smaller than intended, so paths starting above lambda = 1 advanced too slowly to traverse the path within max_iter (most visibly for the JFM, where coefficients were left roughly eight times too small); and once the dual norm fell below 1e-4 it printed in scientific notation, count_digits() returned 0, and the step jumped back to 0.1, preventing the path from descending below lambda = 1e-4 at any iteration count.
• The step now starts at eps at the top of the path and is divided by ten each time the dual norm falls a further decade below its initial value. The rule is scale invariant and penalty independent.
• eps is now used. It was previously discarded by the adaptive branch for both model types and all three penalties, so the documented step-size argument had no effect.
• Removed count_digits(), which is no longer used.
• stagewise_fit() and cv_stagewise() gain lambda_min_ratio, which stops the path once the dual norm falls to that fraction of its value at the top of the path. With the step-size fix in place the JFM path reaches its solution in the first fifth of the iterations and then sits static: roughly 75% of the default 5000 iterations were contributing under 1% of the total coefficient movement, while still inflating the lambda grid that cross-validation has to search. The default follows glmnet (0.01 when n > p, 1e-4 otherwise) and cuts JFM fit-plus-CV time by about 57% with coefficient estimates unchanged. The JSCM path is still moving at iteration 5000 and is left essentially untouched. Set lambda_min_ratio = 0 for the previous behaviour.

swjm 0.1.2 (2026-07-30)

• Corrected the stagewise update directions under the scaled penalties. The argmax of the update objective over the scaled-penalty unit ball carries a factor 1/xi on every death-model coordinate (scaled lasso and mixed-sign cooperative branches: e_j sgn(g)/xi; group and same-sign cooperative branches: (g_a, g_b/xi^2)/||(g_a, g_b/xi)||_2). Earlier versions used unit-norm directions in rescaled gradient coordinates, which under-stepped the death-model block by the rescaling factor at every iteration and systematically over-shrank death-model coefficients.
• The gradient-rescaling factor is now frozen at its initial value, so each fit emulates a single fixed scaled penalty and the solution paths are well behaved.