Tests appeared broken in the main branch because of updates in scikit-learn and scipy.
This fixes several issues by (i) letting scipy run with default precision when testing bfgs/lbfgs, (ii) testing optimum in function values for the lbfgs tests on binary logistic regression, (iii) fixing the argument for the isotonic pav test.
Tests appeared broken in the main branch because of updates in scikit-learn and scipy. This fixes several issues by (i) letting scipy run with default precision when testing bfgs/lbfgs, (ii) testing optimum in function values for the lbfgs tests on binary logistic regression, (iii) fixing the argument for the isotonic pav test.