Bugs
From OpenOpt
| Bug | Is already fixed in subversion repository |
|---|---|
| some rarely encountered bugs for ipopt, ralg, maybe other nonlinear solvers | Yes |
| can't solve sparse SLE with latest scipy - wrapper to UMFPACK has been moved from scipy.sparse to scikits.umfpack | Not yet (I have some thoubles with scikits.sparse installation on my system). You could use rendering instead and cast to dense format or use another sparse solver, e.g. pcg, gmres from scipy.sparse |
| A bug with incorrect shape that sometimes triggers from CVXOPT-connected solvers (e.g. cvxopt_lp, glpk) | Yes |
| Sometimes OpenOpt incorrectly renders FuncDesigner linear constraints (typo "vstack" instead of "hstack"). Maybe, sometimes it triggers with other automatic differentiation results for nonlinear constraints. | Yes |
| Bug | Is already fixed in subversion repository |
|---|---|
| User-supplied args for OpenOpt problems sometimes are not available | Yes |
| Parameter searchDirectionStrategy of solver de set to "best" yields error | Yes |
| Bug in non-linear problems group | Yes |
| NameError: global name 'isspmatrix' is not defined (in ooMisc.py for some problems/solvers) | Yes |
| ralg badly handles problems with nonlinear equality constraints (compared to previous releases) | Yes (some improvements) |
| Bug | Is already fixed in subversion repository |
|---|---|
| numpy.array operations with FuncDesigner oofuns/oovars | Yes |
| If scipy is not installed then importing from openopt may be failed | Yes |
| FuncDesigner sometimes yields the error "blocks must have rank 2" | Yes |
| Sometimes ipopt wrapper fails on FD model with "blocks must have rank 2" | Yes |
| Bug | Is already fixed in subversion repository |
|---|---|
| ralg incorrectly handles problems with number of linear equality constraints more than 1 | Yes |
| FuncDesigner function sum() sometimes yield incorrect derivative (hence it affects optimization of FD models where the function is used) | Yes |
| FuncDesigner function sin() yields incorrect value | Yes |
| scipy_slsqp incorrectly handles box-bound constraints lb <= x <= ub if some lb or ub coords are +/- inf | Yes |
| Bug | Is already fixed in subversion repository |
|---|---|
| Sometimes some solvers fail with error message "instance has no attribute 'x'" | Yes |
| Bug | Is already fixed in subversion repository |
|---|---|
| Incorrect handling of A, Aeq constraints by scipy_slsqp provided c, h are absent | Yes |
| scipy_cobyla, sometimes incorrect handling of nonlinear equality constraints h(x)=0 | Yes |
| Incorrect handling of several constraints R^nvars -> R by NLP/NSP solver ralg | Yes |
| A bug for named variables syntax | Yes |
| sometimes lpSolve reports "solution is not feasible" while it is feasible | Yes |
Known bugs for release 0.21
| Bug | Is already fixed in subversion repository |
|---|---|
| "python setup.py develop" works but "python setup.py install" doesn't work | Yes |
| NLLSP works incorrectly if no user-supplied gradient is provided | Yes |
| Matplotlib warning /usr/lib/python2.5/site-packages/matplotlib/axes.py:5047: DeprecationWarning: replace "faceted=False" with "edgecolors='none'" | Yes |
