Python Polynomial.coordinates方法代码示例

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_orientation_reprojection(self):
        true_s = np.array([.1, .2, -.3])
        true_r = utils.cayley(true_s)
        true_k = 1. / (1. + np.dot(true_s, true_s))
        true_vars = np.r_[true_s, true_k]

        xs = np.random.rand(8, 3)
        true_ys = np.dot(xs, true_r.T)

        sym_vars = Polynomial.coordinates(4, ctype=float)
        x, y, z, w = sym_vars
        sym_s = sym_vars[:3]
        sym_k = sym_vars[3]
        sym_r = utils.cayley_mat(sym_s)
        sym_rd = utils.cayley_denom(sym_s)

        residuals = (np.dot(xs, sym_r.T) - true_ys * sym_rd).flatten()
        cost = sum(r**2 for r in residuals)
        gradients = cost.partial_derivatives()
        constraint = sym_k * (1 + np.dot(sym_s, sym_s)) - 1

        print 'Cost:', cost
        print 'Constraint:', constraint
        print 'At ground truth:'
        print '  Cost = ', cost(*true_vars)
        print '  Constraint = ', constraint(*true_vars)
        print '  Gradients = ', [p(*true_vars) for p in gradients]
        expansions = [solvers.all_monomials(sym_vars, 2) for _ in range(cost.num_vars)]
        for a in expansions:
            a.extend([z*z*w, x*x*w, y*y*w, z*z*w*w, z*w*w])
        minima = optimize.minimize_globally(cost,
                                            [constraint],
                                            expansions=expansions,
                                            diagnostic_solutions=[true_vars])
        print 'Minima: ', minima

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
 def test_two_vars(self):
     x, y = Polynomial.coordinates(2)
     equations = [x**2 + y**2 - 1,
                  x-y]
     expansion_monomials = [
         [x, y],
         [x, y, x*y, x*x, y*y]
     ]
     result = solvers.solve_via_basis_selection(equations, expansion_monomials, x+y+1)
     expected_solutions = [np.sqrt([.5, .5]), -np.sqrt([.5, .5])]
     self.assert_solutions_found(result, expected_solutions)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_estimate_pose_and_depths(self):
        np.random.seed(0)
        num_landmarks = 5

        # Generate ground truth
        true_cayleys = np.array([.1, .2, -.3])
        true_orientation = utils.cayley(true_cayleys)
        true_position = np.array([2., 3., 10.])

        # Generate observed quantities
        true_landmarks = np.random.randn(num_landmarks, 3)
        true_pfeatures = np.dot(true_landmarks - true_position, true_orientation.T)
        true_depths = np.sqrt(np.sum(np.square(true_pfeatures), axis=1))
        true_features = true_pfeatures / true_depths[:, None]
        true_vars = np.hstack((true_orientation.flatten(), true_position, true_depths))

        # Create symbolic quantities
        sym_vars = Polynomial.coordinates(12 + num_landmarks, ctype=float)
        sym_orientation = np.reshape(sym_vars[:9], (3, 3))
        sym_position = np.array(sym_vars[9:12])
        sym_depths = np.array(sym_vars[12:])

        residuals = []
        for i, (landmark, feature, sym_depth) in enumerate(zip(true_landmarks, true_features, sym_depths)):
            residual = np.dot(sym_orientation, landmark - sym_position) - sym_depth * feature
            residuals.extend(residual)

        constraints = (np.dot(sym_orientation.T, sym_orientation) - np.eye(3)).flatten()

        cost = sum(r**2 for r in residuals)
        gradients = cost.partial_derivatives()

        print 'Cost:', cost
        print 'Constraints:'
        for constraint in constraints:
            print '  ', constraint

        print 'At ground truth:'
        print '  Cost = ', cost(*true_vars)
        print '  Constraints = ', utils.evaluate_array(constraints, *true_vars)
        print '  Gradients = ', [p(*true_vars) for p in gradients]
        expansions = solvers.all_monomials(sym_vars, 2)
        minima = optimize.minimize_globally(cost,
                                            constraints,
                                            expansions=expansions,
                                            #diagnostic_solutions=[true_vars],
                                            )

        estimated_r = np.reshape(minima, (3, 3))
        error = np.linalg.norm(estimated_r - true_orientation)

        print 'Minima:\n', estimated_r
        print 'Ground truth:\n', true_orientation
        print 'Error:', error

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
 def test_synthetic_ideal(self):
     zeros = [[-2., -3., 5., 6.], [4.5, 5., -1., 8.]]
     equations = ideal_from_variety(zeros, ctype=float)
     coords = Polynomial.coordinates(len(zeros[0]))
     expansion_monomials = solvers.all_monomials(coords, degree=1)
     lambda_poly = sum(xi * (i + 1) for i, xi in enumerate(coords)) + 1
     result = solvers.solve_via_basis_selection(
         equations,
         expansion_monomials,
         lambda_poly,
         diagnostic_solutions=[zeros[0]])
     self.assert_solutions_found(result, zeros)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
 def test_two_circles(self):
     x, y = Polynomial.coordinates(2)
     equations = [
         (x+2)**2 + (y+2)**2 - 25,
         (x-6)**2 + (y+2)**2 - 25,
     ]
     expansion_monomials = [x, y, x*y, x*x, y*y, x*x*y]
     expected_solutions = [(2, 1), (2, -5)]
     result = solvers.solve_via_basis_selection(equations,
                                                expansion_monomials,
                                                x+2*y-3,
                                                diagnostic_solutions=expected_solutions)
     self.assert_solutions_found(result, expected_solutions)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
def find_critical_points(polynomial, constraints=None, expansions=2, lambda_poly=None, seed=0, **kwargs):
    """Find all local minima, local maxima, and saddle points of the given polynomial by solving the first order
    conditions."""
    system = polynomial.astype(float).partial_derivatives()
    system = filter(lambda p: p.total_degree > 0, system)
    if constraints is not None:
        system.extend(constraints)
    coords = Polynomial.coordinates(polynomial.num_vars, ctype=float)
    if lambda_poly is None:
        rng = random.Random(seed)  # use this rng for repeatability
        lambda_poly = sum(p * rng.random() for p in coords) + rng.random()
    if isinstance(expansions, numbers.Integral):
        expansions = [solvers.all_monomials(coords, expansions)] * len(system)
    return solvers.solve_via_basis_selection(system, expansions, lambda_poly, **kwargs)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
 def test_three_circles(self):
     x, y, z = Polynomial.coordinates(3)
     equations = [
         (x-6)**2 + (y-1)**2 + (z-1)**2 - 25,
         (x-1)**2 + (y-6)**2 + (z-1)**2 - 25,
         (x-1)**2 + (y-1)**2 + (z-6)**2 - 25,
     ]
     expansion_monomials = solvers.all_monomials((x, y, z), degree=2)
     expected_solutions = [(1, 1, 1)]
     lambda_poly = x + 2*y + 3*z + 4
     result = solvers.solve_via_basis_selection(equations,
                                                expansion_monomials,
                                                lambda_poly,
                                                diagnostic_solutions=expected_solutions)
     self.assert_solutions_found(result, expected_solutions)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_three_vars(self):
        x, y, z = Polynomial.coordinates(3)
        equations = [
            x**2 + y**2 + z**2 - 1,
            x - y,
            x - z
        ]
        expansion_monomials = [
            [],
            [x, y, z],
            [x, y, z]
        ]

        result = solvers.solve_via_basis_selection(equations, expansion_monomials, x)

        point = np.sqrt(np.ones(3) / 3.)
        expected_solutions = [point, -point]
        self.assert_solutions_found(result, expected_solutions)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_estimate_orientation_9params(self):
        np.random.seed(0)

        true_s = np.array([.1, .2, -.3])
        true_r = utils.cayley(true_s)
        true_vars = true_r.flatten()

        observed_xs = np.random.rand(8, 3)
        observed_ys = np.dot(observed_xs, true_r.T)

        sym_vars = Polynomial.coordinates(9, ctype=float)
        sym_r = np.reshape(sym_vars, (3, 3))

        residuals = (np.dot(observed_xs, sym_r.T) - observed_ys).flatten()
        constraints = (np.dot(sym_r.T, sym_r) - np.eye(3)).flatten()

        cost = sum(r**2 for r in residuals)
        gradients = cost.partial_derivatives()

        print 'Cost:', cost
        print 'Constraints:'
        for constraint in constraints:
            print '  ', constraint

        print 'At ground truth:'
        print '  Cost = ', cost(*true_vars)
        print '  Constraints = ', utils.evaluate_array(constraints, *true_vars)
        print '  Gradients = ', [p(*true_vars) for p in gradients]
        expansions = [solvers.all_monomials(sym_vars, 2) for _ in range(cost.num_vars)]
        minima = optimize.minimize_globally(cost,
                                            constraints,
                                            expansions=expansions,
                                            #diagnostic_solutions=[true_vars],
                                            )

        estimated_r = np.reshape(minima, (3, 3))
        error = np.linalg.norm(estimated_r - true_r)

        print 'Minima:\n', estimated_r
        print 'Ground truth:\n', true_r
        print 'Error:', error

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_range_optimization_2d(self):
        np.random.seed(0)

        bases = np.round(np.random.randn(2, 3) * 6.)
        true_position = np.array([-2., 1., 3.])
        true_ranges = np.array([np.linalg.norm(base - true_position) for base in bases])
        true_vars = np.hstack((true_position, true_ranges))

        sym_vars = Polynomial.coordinates(3 + len(bases), float)
        sym_position = sym_vars[:3]
        sym_ranges = sym_vars[3:]

        cost = sum((zz - z)**2 for zz, z in zip(sym_ranges, true_ranges))
        constraints = [np.sum(np.square(base - sym_position)) - sym_range*sym_range
                       for base, sym_range in zip(bases, sym_ranges)]

        lagrangian = make_langrangian(-cost, constraints)
        gradients = lagrangian.partial_derivatives()

        print 'Cost:', cost
        print 'Lagrangian:', lagrangian
        print 'Constraints:'
        for constraint in constraints:
            print '  ', constraint
        print 'Gradients:'
        for gradient in gradients:
            print '  ', gradient

        print 'At ground truth:'
        print '  Cost = ', cost(*true_vars)
        print '  Constraints = ', utils.evaluate_array(constraints, *true_vars)
        #print '  Gradients = ', [p(*true_vars) for p in gradients]

        expansions = 3  # solvers.all_monomials(sym_vars, 2) + [sym_vars[2]**3, sym_vars[0]**2*sym_vars[2]]
        minima = optimize.minimize_globally(-lagrangian, [], expansions=expansions, verbosity=2,
                                            diagnostic_solutions=[], include_grobner=False)

        error = np.linalg.norm(minima - true_position)
        print minima
        print true_position
        print 'Error:', error

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
    def test_optimize_2d(self):
        np.random.seed(0)
        x, y = Polynomial.coordinates(2, float)
        fs = [
            x+1,
            y+1,
            x*x,
        ]
        true_solution = np.array([2., 5.])
        cost = sum((f(x, y) - f(*true_solution)) ** 2 for f in fs)

        import matplotlib.pyplot as plt
        dX, dY = np.meshgrid(np.linspace(-.1, .1, 50), np.linspace(-.1, .1, 50))
        X = true_solution[0] + dX
        Y = true_solution[1] + dY * 1j
        Z = np.abs(cost(X, Y))
        print np.min(Z), np.max(Z)
        plt.contourf(dX, dY, Z, levels=np.logspace(np.log10(np.min(Z)), np.log10(np.max(Z)), 16))
        plt.plot(0, 0, 'mx')
        plt.show()
        return

        gradients = cost.partial_derivatives()
        print 'Cost:', cost
        print 'Residuals:'
        for f in fs:
            print f(x, y) - f(*true_solution)
        print 'Gradients:'
        for gradient in gradients:
            print '  ', gradient(*true_solution)
        print 'Jacobian:'
        print polynomial_jacobian(fs)(*true_solution)
        expansions = 3  # solvers.all_monomials(sym_vars, 2) + [sym_vars[2]**3, sym_vars[0]**2*sym_vars[2]]
        minima = optimize.minimize_globally(cost, expansions=expansions, verbosity=2, constraints=fs[:2],
                                            #diagnostic_solutions=[true_vars]
        )
        np.testing.assert_array_almost_equal(minima, true_solution)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]

#.........这里部分代码省略.........

    # Check that lambda * b = action_b * b + action_r * r at solution
    for bi, b_row, r_row in zip(basis, c_action_b, c_action_r):
        assert len(r_row) == len(dependent)
        assert len(b_row) == len(basis)
        lhs = bi*lambda_poly
        rhs = (sum(cj*rj for cj, rj in zip(r_row, x_dependent)) +
               sum(cj*bj for cj, bj in zip(b_row, x_basis)))
        if diagnostic_solutions is not None:
            for solution in diagnostic_solutions:
                lvalue = lhs(*solution)
                rvalue = rhs(*solution)
                if verbosity >= 2:
                    print '    at %s, lhs=%s, rhs=%s' % (solution, lvalue, rvalue)
                if abs(lvalue - rvalue) > 1e-8:
                    raise DiagnosticError('expected action equation lhs (=%f) to match rhs (=%f) at %s:\n'
                                          '  lhs=%s\n  rhs=%s' %
                                          (lvalue, rvalue, solution, lhs, rhs))

    # Compute action matrix form for lambda_poly * basis
    soln = np.linalg.solve(c1, c2)
    c_action = c_action_b - np.dot(c_action_r, soln)

    if verbosity >= 2:
        print 'Solution for required monomials:'
        print soln
        for row, monomial in zip(soln, dependent):
            print '  %s = - %s * basis {at points on variety}' % (as_monomial(monomial), row)

    # Check that basis + soln * required = 0 at diagnostic solutions
    if diagnostic_solutions is not None:
        c_solved = np.hstack((np.eye(len(required) + len(eliminated)), soln))
        for solution in diagnostic_solutions:
            values = np.dot(c_solved, evaluate_poly_vector(x_post, solution))
            if verbosity >= 2:
                print '  Evaluated at %s: %s' % (solution, values)
            if np.abs(values).max() > 1e-8:
                idx = np.abs(values).argmax()
                raise DiagnosticError('expected equation %d to evaluate to zero but received %f' % (idx, values[idx]))

    # Check that lambda * b = action * b at solution (note that lambda is a polynomial, not a matrix here)
    if verbosity >= 2 or diagnostic_solutions is not None:
        if verbosity >= 2:
            print 'Action matrix:'
            print c_action
        for bi, row in zip(basis, c_action):
            assert len(basis) == len(row)
            lhs = bi*lambda_poly
            rhs = sum(cj * bj for cj, bj in zip(row, x_basis))
            if verbosity >= 2:
                print '  %s * (%s) = %s = %s' % (as_term(bi, nvars), lambda_poly, lhs, rhs)
            if diagnostic_solutions is not None:
                for solution in diagnostic_solutions:
                    lvalue = lhs(*solution)
                    rvalue = rhs(*solution)
                    if verbosity >= 2:
                        print '    at %s, lhs=%s, rhs=%s' % (solution, lvalue, rvalue)
                    if abs(lvalue - rvalue) > 1e-8:
                        raise DiagnosticError('expected action equation lhs (=%f) to match rhs (=%f) at %s:\n'
                                              '  lhs=%s\n  rhs=%s' %
                                              (lvalue, rvalue, solution, lhs, rhs))

    # Find indices within basis
    unit_index = basis.index(Polynomial.constant(1, nvars))

    # Compute eigenvalues and eigenvectors
    eigvals, eigvecs = np.linalg.eig(c_action)

    if verbosity >= 2:
        print 'Eigenvectors:'
        print eigvecs

    # Divide out the unit monomial row
    nrm = eigvecs[unit_index]
    mask = np.abs(nrm) > 1e-8
    monomial_values = (eigvecs[:, mask] / eigvecs[unit_index][mask]).T

    if verbosity >= 2:
        print 'Normalized eigenvectors:'
        print monomial_values

    # Test each solution
    solutions = []
    for values in monomial_values:
        #candidate = [eigvec[i]/eigvec[unit_index] for i in var_indices]
        for solution in solve_monomial_equations(basis, values):
            values = [f(*solution) for f in equations]
            if verbosity >= 1:
                print 'Candidate solution: %s -> Values = %s' % (solution, values)
            if np.linalg.norm(values) < 1e-8:
                solutions.append(solution)

    # Report final solutions
    base_vars = Polynomial.coordinates(lambda_poly.num_vars)
    if verbosity >= 1:
        print 'Solutions:'
        for solution in solutions:
            print '  ' + ' '.join('%s=%s' % (var, val) for var, val in zip(base_vars, solution))

    return SolutionSet(solutions)

# 需要导入模块: from polynomial import Polynomial [as 别名]
# 或者: from polynomial.Polynomial import coordinates [as 别名]
def make_langrangian(objective, constraints):
    ex_vars = Polynomial.coordinates(objective.num_vars + len(constraints))
    orig_vars = ex_vars[:objective.num_vars]
    lg_vars = ex_vars[objective.num_vars:]
    return objective(*orig_vars) + sum(lg * c(*orig_vars) for lg, c in zip(lg_vars, constraints))