""" Unit tests for nonlinear solvers Author: Ondrej Certik May 2007 """ from numpy.testing import assert_ import pytest from scipy.optimize import _nonlin as nonlin, root from numpy import diag, dot from numpy.linalg import inv import numpy as np from .test_minpack import pressure_network SOLVERS = {'anderson': nonlin.anderson, 'diagbroyden': nonlin.diagbroyden, 'linearmixing': nonlin.linearmixing, 'excitingmixing': nonlin.excitingmixing, 'broyden1': nonlin.broyden1, 'broyden2': nonlin.broyden2, 'krylov': nonlin.newton_krylov} MUST_WORK = {'anderson': nonlin.anderson, 'broyden1': nonlin.broyden1, 'broyden2': nonlin.broyden2, 'krylov': nonlin.newton_krylov} #------------------------------------------------------------------------------- # Test problems #------------------------------------------------------------------------------- def F(x): x = np.asarray(x).T d = diag([3,2,1.5,1,0.5]) c = 0.01 f = -d @ x - c * float(x.T @ x) * x return f F.xin = [1,1,1,1,1] F.KNOWN_BAD = {} F.JAC_KSP_BAD = {} F.ROOT_JAC_KSP_BAD = {} def F2(x): return x F2.xin = [1,2,3,4,5,6] F2.KNOWN_BAD = {'linearmixing': nonlin.linearmixing, 'excitingmixing': nonlin.excitingmixing} F2.JAC_KSP_BAD = {} F2.ROOT_JAC_KSP_BAD = {} def F2_lucky(x): return x F2_lucky.xin = [0,0,0,0,0,0] F2_lucky.KNOWN_BAD = {} F2_lucky.JAC_KSP_BAD = {} F2_lucky.ROOT_JAC_KSP_BAD = {} def F3(x): A = np.array([[-2, 1, 0.], [1, -2, 1], [0, 1, -2]]) b = np.array([1, 2, 3.]) return A @ x - b F3.xin = [1,2,3] F3.KNOWN_BAD = {} F3.JAC_KSP_BAD = {} F3.ROOT_JAC_KSP_BAD = {} def F4_powell(x): A = 1e4 return [A*x[0]*x[1] - 1, np.exp(-x[0]) + np.exp(-x[1]) - (1 + 1/A)] F4_powell.xin = [-1, -2] F4_powell.KNOWN_BAD = {'linearmixing': nonlin.linearmixing, 'excitingmixing': nonlin.excitingmixing, 'diagbroyden': nonlin.diagbroyden} # In the extreme case, it does not converge for nolinear problem solved by # MINRES and root problem solved by GMRES/BiCGStab/CGS/MINRES/TFQMR when using # Krylov method to approximate Jacobian F4_powell.JAC_KSP_BAD = {'minres'} F4_powell.ROOT_JAC_KSP_BAD = {'gmres', 'bicgstab', 'cgs', 'minres', 'tfqmr'} def F5(x): return pressure_network(x, 4, np.array([.5, .5, .5, .5])) F5.xin = [2., 0, 2, 0] F5.KNOWN_BAD = {'excitingmixing': nonlin.excitingmixing, 'linearmixing': nonlin.linearmixing, 'diagbroyden': nonlin.diagbroyden} # In the extreme case, the Jacobian inversion yielded zero vector for nonlinear # problem solved by CGS/MINRES and it does not converge for root problem solved # by MINRES and when using Krylov method to approximate Jacobian F5.JAC_KSP_BAD = {'cgs', 'minres'} F5.ROOT_JAC_KSP_BAD = {'minres'} def F6(x): x1, x2 = x J0 = np.array([[-4.256, 14.7], [0.8394989, 0.59964207]]) v = np.array([(x1 + 3) * (x2**5 - 7) + 3*6, np.sin(x2 * np.exp(x1) - 1)]) return -np.linalg.solve(J0, v) F6.xin = [-0.5, 1.4] F6.KNOWN_BAD = {'excitingmixing': nonlin.excitingmixing, 'linearmixing': nonlin.linearmixing, 'diagbroyden': nonlin.diagbroyden} F6.JAC_KSP_BAD = {} F6.ROOT_JAC_KSP_BAD = {} #------------------------------------------------------------------------------- # Tests #------------------------------------------------------------------------------- class TestNonlin: """ Check the Broyden methods for a few test problems. broyden1, broyden2, and newton_krylov must succeed for all functions. Some of the others don't -- tests in KNOWN_BAD are skipped. """ def _check_nonlin_func(self, f, func, f_tol=1e-2): # Test all methods mentioned in the class `KrylovJacobian` if func == SOLVERS['krylov']: for method in ['gmres', 'bicgstab', 'cgs', 'minres', 'tfqmr']: if method in f.JAC_KSP_BAD: continue x = func(f, f.xin, method=method, line_search=None, f_tol=f_tol, maxiter=200, verbose=0) assert_(np.absolute(f(x)).max() < f_tol) x = func(f, f.xin, f_tol=f_tol, maxiter=200, verbose=0) assert_(np.absolute(f(x)).max() < f_tol) def _check_root(self, f, method, f_tol=1e-2): # Test Krylov methods if method == 'krylov': for jac_method in ['gmres', 'bicgstab', 'cgs', 'minres', 'tfqmr']: if jac_method in f.ROOT_JAC_KSP_BAD: continue res = root(f, f.xin, method=method, options={'ftol': f_tol, 'maxiter': 200, 'disp': 0, 'jac_options': {'method': jac_method}}) assert_(np.absolute(res.fun).max() < f_tol) res = root(f, f.xin, method=method, options={'ftol': f_tol, 'maxiter': 200, 'disp': 0}) assert_(np.absolute(res.fun).max() < f_tol) @pytest.mark.xfail def _check_func_fail(self, *a, **kw): pass def test_problem_nonlin(self): for f in [F, F2, F2_lucky, F3, F4_powell, F5, F6]: for func in SOLVERS.values(): if func in f.KNOWN_BAD.values(): if func in MUST_WORK.values(): self._check_func_fail(f, func) continue self._check_nonlin_func(f, func) @pytest.mark.parametrize("method", ['lgmres', 'gmres', 'bicgstab', 'cgs', 'minres', 'tfqmr']) def test_tol_norm_called(self, method): # Check that supplying tol_norm keyword to nonlin_solve works self._tol_norm_used = False def local_norm_func(x): self._tol_norm_used = True return np.absolute(x).max() nonlin.newton_krylov(F, F.xin, method=method, f_tol=1e-2, maxiter=200, verbose=0, tol_norm=local_norm_func) assert_(self._tol_norm_used) def test_problem_root(self): for f in [F, F2, F2_lucky, F3, F4_powell, F5, F6]: for meth in SOLVERS: if meth in f.KNOWN_BAD: if meth in MUST_WORK: self._check_func_fail(f, meth) continue self._check_root(f, meth) class TestSecant: """Check that some Jacobian approximations satisfy the secant condition""" xs = [np.array([1,2,3,4,5], float), np.array([2,3,4,5,1], float), np.array([3,4,5,1,2], float), np.array([4,5,1,2,3], float), np.array([9,1,9,1,3], float), np.array([0,1,9,1,3], float), np.array([5,5,7,1,1], float), np.array([1,2,7,5,1], float),] fs = [x**2 - 1 for x in xs] def _check_secant(self, jac_cls, npoints=1, **kw): """ Check that the given Jacobian approximation satisfies secant conditions for last `npoints` points. """ jac = jac_cls(**kw) jac.setup(self.xs[0], self.fs[0], None) for j, (x, f) in enumerate(zip(self.xs[1:], self.fs[1:])): jac.update(x, f) for k in range(min(npoints, j+1)): dx = self.xs[j-k+1] - self.xs[j-k] df = self.fs[j-k+1] - self.fs[j-k] assert_(np.allclose(dx, jac.solve(df))) # Check that the `npoints` secant bound is strict if j >= npoints: dx = self.xs[j-npoints+1] - self.xs[j-npoints] df = self.fs[j-npoints+1] - self.fs[j-npoints] assert_(not np.allclose(dx, jac.solve(df))) def test_broyden1(self): self._check_secant(nonlin.BroydenFirst) def test_broyden2(self): self._check_secant(nonlin.BroydenSecond) def test_broyden1_update(self): # Check that BroydenFirst update works as for a dense matrix jac = nonlin.BroydenFirst(alpha=0.1) jac.setup(self.xs[0], self.fs[0], None) B = np.identity(5) * (-1/0.1) for last_j, (x, f) in enumerate(zip(self.xs[1:], self.fs[1:])): df = f - self.fs[last_j] dx = x - self.xs[last_j] B += (df - dot(B, dx))[:,None] * dx[None,:] / dot(dx, dx) jac.update(x, f) assert_(np.allclose(jac.todense(), B, rtol=1e-10, atol=1e-13)) def test_broyden2_update(self): # Check that BroydenSecond update works as for a dense matrix jac = nonlin.BroydenSecond(alpha=0.1) jac.setup(self.xs[0], self.fs[0], None) H = np.identity(5) * (-0.1) for last_j, (x, f) in enumerate(zip(self.xs[1:], self.fs[1:])): df = f - self.fs[last_j] dx = x - self.xs[last_j] H += (dx - dot(H, df))[:,None] * df[None,:] / dot(df, df) jac.update(x, f) assert_(np.allclose(jac.todense(), inv(H), rtol=1e-10, atol=1e-13)) def test_anderson(self): # Anderson mixing (with w0=0) satisfies secant conditions # for the last M iterates, see [Ey]_ # # .. [Ey] V. Eyert, J. Comp. Phys., 124, 271 (1996). self._check_secant(nonlin.Anderson, M=3, w0=0, npoints=3) class TestLinear: """Solve a linear equation; some methods find the exact solution in a finite number of steps""" def _check(self, jac, N, maxiter, complex=False, **kw): np.random.seed(123) A = np.random.randn(N, N) if complex: A = A + 1j*np.random.randn(N, N) b = np.random.randn(N) if complex: b = b + 1j*np.random.randn(N) def func(x): return dot(A, x) - b sol = nonlin.nonlin_solve(func, np.zeros(N), jac, maxiter=maxiter, f_tol=1e-6, line_search=None, verbose=0) assert_(np.allclose(dot(A, sol), b, atol=1e-6)) def test_broyden1(self): # Broyden methods solve linear systems exactly in 2*N steps self._check(nonlin.BroydenFirst(alpha=1.0), 20, 41, False) self._check(nonlin.BroydenFirst(alpha=1.0), 20, 41, True) def test_broyden2(self): # Broyden methods solve linear systems exactly in 2*N steps self._check(nonlin.BroydenSecond(alpha=1.0), 20, 41, False) self._check(nonlin.BroydenSecond(alpha=1.0), 20, 41, True) def test_anderson(self): # Anderson is rather similar to Broyden, if given enough storage space self._check(nonlin.Anderson(M=50, alpha=1.0), 20, 29, False) self._check(nonlin.Anderson(M=50, alpha=1.0), 20, 29, True) def test_krylov(self): # Krylov methods solve linear systems exactly in N inner steps self._check(nonlin.KrylovJacobian, 20, 2, False, inner_m=10) self._check(nonlin.KrylovJacobian, 20, 2, True, inner_m=10) class TestJacobianDotSolve: """Check that solve/dot methods in Jacobian approximations are consistent""" def _func(self, x): return x**2 - 1 + np.dot(self.A, x) def _check_dot(self, jac_cls, complex=False, tol=1e-6, **kw): np.random.seed(123) N = 7 def rand(*a): q = np.random.rand(*a) if complex: q = q + 1j*np.random.rand(*a) return q def assert_close(a, b, msg): d = abs(a - b).max() f = tol + abs(b).max()*tol if d > f: raise AssertionError('%s: err %g' % (msg, d)) self.A = rand(N, N) # initialize x0 = np.random.rand(N) jac = jac_cls(**kw) jac.setup(x0, self._func(x0), self._func) # check consistency for k in range(2*N): v = rand(N) if hasattr(jac, '__array__'): Jd = np.array(jac) if hasattr(jac, 'solve'): Gv = jac.solve(v) Gv2 = np.linalg.solve(Jd, v) assert_close(Gv, Gv2, 'solve vs array') if hasattr(jac, 'rsolve'): Gv = jac.rsolve(v) Gv2 = np.linalg.solve(Jd.T.conj(), v) assert_close(Gv, Gv2, 'rsolve vs array') if hasattr(jac, 'matvec'): Jv = jac.matvec(v) Jv2 = np.dot(Jd, v) assert_close(Jv, Jv2, 'dot vs array') if hasattr(jac, 'rmatvec'): Jv = jac.rmatvec(v) Jv2 = np.dot(Jd.T.conj(), v) assert_close(Jv, Jv2, 'rmatvec vs array') if hasattr(jac, 'matvec') and hasattr(jac, 'solve'): Jv = jac.matvec(v) Jv2 = jac.solve(jac.matvec(Jv)) assert_close(Jv, Jv2, 'dot vs solve') if hasattr(jac, 'rmatvec') and hasattr(jac, 'rsolve'): Jv = jac.rmatvec(v) Jv2 = jac.rmatvec(jac.rsolve(Jv)) assert_close(Jv, Jv2, 'rmatvec vs rsolve') x = rand(N) jac.update(x, self._func(x)) def test_broyden1(self): self._check_dot(nonlin.BroydenFirst, complex=False) self._check_dot(nonlin.BroydenFirst, complex=True) def test_broyden2(self): self._check_dot(nonlin.BroydenSecond, complex=False) self._check_dot(nonlin.BroydenSecond, complex=True) def test_anderson(self): self._check_dot(nonlin.Anderson, complex=False) self._check_dot(nonlin.Anderson, complex=True) def test_diagbroyden(self): self._check_dot(nonlin.DiagBroyden, complex=False) self._check_dot(nonlin.DiagBroyden, complex=True) def test_linearmixing(self): self._check_dot(nonlin.LinearMixing, complex=False) self._check_dot(nonlin.LinearMixing, complex=True) def test_excitingmixing(self): self._check_dot(nonlin.ExcitingMixing, complex=False) self._check_dot(nonlin.ExcitingMixing, complex=True) def test_krylov(self): self._check_dot(nonlin.KrylovJacobian, complex=False, tol=1e-3) self._check_dot(nonlin.KrylovJacobian, complex=True, tol=1e-3) class TestNonlinOldTests: """ Test case for a simple constrained entropy maximization problem (the machine translation example of Berger et al in Computational Linguistics, vol 22, num 1, pp 39--72, 1996.) """ def test_broyden1(self): x = nonlin.broyden1(F,F.xin,iter=12,alpha=1) assert_(nonlin.norm(x) < 1e-9) assert_(nonlin.norm(F(x)) < 1e-9) def test_broyden2(self): x = nonlin.broyden2(F,F.xin,iter=12,alpha=1) assert_(nonlin.norm(x) < 1e-9) assert_(nonlin.norm(F(x)) < 1e-9) def test_anderson(self): x = nonlin.anderson(F,F.xin,iter=12,alpha=0.03,M=5) assert_(nonlin.norm(x) < 0.33) def test_linearmixing(self): x = nonlin.linearmixing(F,F.xin,iter=60,alpha=0.5) assert_(nonlin.norm(x) < 1e-7) assert_(nonlin.norm(F(x)) < 1e-7) def test_exciting(self): x = nonlin.excitingmixing(F,F.xin,iter=20,alpha=0.5) assert_(nonlin.norm(x) < 1e-5) assert_(nonlin.norm(F(x)) < 1e-5) def test_diagbroyden(self): x = nonlin.diagbroyden(F,F.xin,iter=11,alpha=1) assert_(nonlin.norm(x) < 1e-8) assert_(nonlin.norm(F(x)) < 1e-8) def test_root_broyden1(self): res = root(F, F.xin, method='broyden1', options={'nit': 12, 'jac_options': {'alpha': 1}}) assert_(nonlin.norm(res.x) < 1e-9) assert_(nonlin.norm(res.fun) < 1e-9) def test_root_broyden2(self): res = root(F, F.xin, method='broyden2', options={'nit': 12, 'jac_options': {'alpha': 1}}) assert_(nonlin.norm(res.x) < 1e-9) assert_(nonlin.norm(res.fun) < 1e-9) def test_root_anderson(self): res = root(F, F.xin, method='anderson', options={'nit': 12, 'jac_options': {'alpha': 0.03, 'M': 5}}) assert_(nonlin.norm(res.x) < 0.33) def test_root_linearmixing(self): res = root(F, F.xin, method='linearmixing', options={'nit': 60, 'jac_options': {'alpha': 0.5}}) assert_(nonlin.norm(res.x) < 1e-7) assert_(nonlin.norm(res.fun) < 1e-7) def test_root_excitingmixing(self): res = root(F, F.xin, method='excitingmixing', options={'nit': 20, 'jac_options': {'alpha': 0.5}}) assert_(nonlin.norm(res.x) < 1e-5) assert_(nonlin.norm(res.fun) < 1e-5) def test_root_diagbroyden(self): res = root(F, F.xin, method='diagbroyden', options={'nit': 11, 'jac_options': {'alpha': 1}}) assert_(nonlin.norm(res.x) < 1e-8) assert_(nonlin.norm(res.fun) < 1e-8)