Low precision for l1 regularized logistic regression problem.
Closed this issue · 1 comments
tiagoakle commented
This issue was raised by @bodono on the cvxgrp/cvxpy repository but belongs here.
When solving an l1 logistic regression problem ECOS produces a very inexact solution, see
https://github.com/bodono/apgpy/blob/master/notebooks/L1LogisticRegression.ipynb
However, when we disable the balancing the issue disappears.