Python optimize.fmin_ncg方法代码示例

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def _fit_ncg(f, score, start_params, fargs, kwargs, disp=True,
                 maxiter=100, callback=None, retall=False,
                 full_output=True, hess=None):
    fhess_p = kwargs.setdefault('fhess_p', None)
    avextol = kwargs.setdefault('avextol', 1.0000000000000001e-05)
    epsilon = kwargs.setdefault('epsilon', 1.4901161193847656e-08)
    retvals = optimize.fmin_ncg(f, start_params, score, fhess_p=fhess_p,
                                fhess=hess, args=fargs, avextol=avextol,
                                epsilon=epsilon, maxiter=maxiter,
                                full_output=full_output, disp=disp,
                                retall=retall, callback=callback)
    if full_output:
        if not retall:
            xopt, fopt, fcalls, gcalls, hcalls, warnflag = retvals
        else:
            xopt, fopt, fcalls, gcalls, hcalls, warnflag, allvecs =\
                retvals
        converged = not warnflag
        retvals = {'fopt': fopt, 'fcalls': fcalls, 'gcalls': gcalls,
                   'hcalls': hcalls, 'warnflag': warnflag,
                   'converged': converged}
        if retall:
            retvals.update({'allvecs': allvecs})
    else:
        xopt = retvals
        retvals = None

    return xopt, retvals 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def fitgmm_cu(self, start, optim_method='bfgs', optim_args=None):
        '''estimate parameters using continuously updating GMM

        Parameters
        ----------
        start : array_like
            starting values for minimization

        Returns
        -------
        paramest : array
            estimated parameters

        Notes
        -----
        todo: add fixed parameter option, not here ???

        uses scipy.optimize.fmin

        '''
##        if not fixed is None:  #fixed not defined in this version
##            raise NotImplementedError

        if optim_args is None:
            optim_args = {}

        if optim_method == 'nm':
            optimizer = optimize.fmin
        elif optim_method == 'bfgs':
            optimizer = optimize.fmin_bfgs
            optim_args['fprime'] = self.score_cu
        elif optim_method == 'ncg':
            optimizer = optimize.fmin_ncg
        else:
            raise ValueError('optimizer method not available')

        #TODO: add other optimization options and results
        return optimizer(self.gmmobjective_cu, start, args=(), **optim_args) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg(self, use_wrapper=False):
        """ line-search Newton conjugate gradient optimization routine
        """
        if use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': False,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=False,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 22, self.gradcalls)  # 0.13.0
        #assert_(self.gradcalls <= 18, self.gradcalls) # 0.9.0
        #assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        #assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg_hess(self, use_wrapper=False):
        """ Newton conjugate gradient with Hessian """
        if use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': False,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       hess=self.hess,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       fhess=self.hess,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=False,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 18, self.gradcalls)  # 0.9.0
        # assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        # assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg_hessp(self, use_wrapper=False):
        """ Newton conjugate gradient with Hessian times a vector p """
        if use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': False,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       hessp=self.hessp,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       fhess_p=self.hessp,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=False,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 18, self.gradcalls)  # 0.9.0
        # assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        # assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg(self):
        # line-search Newton conjugate gradient optimization routine
        if self.use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': self.disp,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=self.disp,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 22, self.gradcalls)  # 0.13.0
        #assert_(self.gradcalls <= 18, self.gradcalls) # 0.9.0
        #assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        #assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg_hess(self):
        # Newton conjugate gradient with Hessian
        if self.use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': self.disp,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       hess=self.hess,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       fhess=self.hess,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=self.disp,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 18, self.gradcalls)  # 0.9.0
        # assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        # assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_ncg_hessp(self):
        # Newton conjugate gradient with Hessian times a vector p.
        if self.use_wrapper:
            opts = {'maxiter': self.maxiter, 'disp': self.disp,
                    'return_all': False}
            retval = optimize.minimize(self.func, self.startparams,
                                       method='Newton-CG', jac=self.grad,
                                       hessp=self.hessp,
                                       args=(), options=opts)['x']
        else:
            retval = optimize.fmin_ncg(self.func, self.startparams, self.grad,
                                       fhess_p=self.hessp,
                                       args=(), maxiter=self.maxiter,
                                       full_output=False, disp=self.disp,
                                       retall=False)

        params = retval

        assert_allclose(self.func(params), self.func(self.solution),
                        atol=1e-6)

        # Ensure that function call counts are 'known good'; these are from
        # Scipy 0.7.0. Don't allow them to increase.
        assert_(self.funccalls == 7, self.funccalls)
        assert_(self.gradcalls <= 18, self.gradcalls)  # 0.9.0
        # assert_(self.gradcalls == 18, self.gradcalls) # 0.8.0
        # assert_(self.gradcalls == 22, self.gradcalls) # 0.7.0

        # Ensure that the function behaves the same; this is from Scipy 0.7.0
        assert_allclose(self.trace[3:5],
                        [[-4.35700753e-07, -5.24869435e-01, 4.87527480e-01],
                         [-4.35700753e-07, -5.24869401e-01, 4.87527774e-01]],
                        atol=1e-6, rtol=1e-7) 

# 需要导入模块: from scipy import optimize [as 别名]
# 或者: from scipy.optimize import fmin_ncg [as 别名]
def test_minimize_callback_copies_array(self, method):
        # Check that arrays passed to callbacks are not modified
        # inplace by the optimizer afterward

        if method in ('fmin_tnc', 'fmin_l_bfgs_b'):
            func = lambda x: (optimize.rosen(x), optimize.rosen_der(x))
        else:
            func = optimize.rosen
            jac = optimize.rosen_der
            hess = optimize.rosen_hess

        x0 = np.zeros(10)

        # Set options
        kwargs = {}
        if method.startswith('fmin'):
            routine = getattr(optimize, method)
            if method == 'fmin_slsqp':
                kwargs['iter'] = 5
            elif method == 'fmin_tnc':
                kwargs['maxfun'] = 100
            else:
                kwargs['maxiter'] = 5
        else:
            def routine(*a, **kw):
                kw['method'] = method
                return optimize.minimize(*a, **kw)

            if method == 'TNC':
                kwargs['options'] = dict(maxiter=100)
            else:
                kwargs['options'] = dict(maxiter=5)

        if method in ('fmin_ncg',):
            kwargs['fprime'] = jac
        elif method in ('Newton-CG',):
            kwargs['jac'] = jac
        elif method in ('trust-krylov', 'trust-exact', 'trust-ncg', 'dogleg',
                        'trust-constr'):
            kwargs['jac'] = jac
            kwargs['hess'] = hess

        # Run with callback
        results = []

        def callback(x, *args, **kwargs):
            results.append((x, np.copy(x)))

        sol = routine(func, x0, callback=callback, **kwargs)

        # Check returned arrays coincide with their copies and have no memory overlap
        assert_(len(results) > 2)
        assert_(all(np.all(x == y) for x, y in results))
        assert_(not any(np.may_share_memory(x[0], y[0]) for x, y in itertools.combinations(results, 2)))