def _select_function(sort, typ):
if typ in ['F','D']:
if callable(sort):
# assume the user knows what they're doing
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x,y: (np.real(x/y) < 0.0)
elif sort == 'rhp':
sfunction = lambda x,y: (np.real(x/y) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x,y: (abs(x/y) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x,y: (abs(x/y) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or "
"one of ('lhp','rhp','iuc','ouc')")
elif typ in ['f','d']:
if callable(sort):
# assume the user knows what they're doing
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x,y,z: (np.real((x+y*1j)/z) < 0.0)
elif sort == 'rhp':
sfunction = lambda x,y,z: (np.real((x+y*1j)/z) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x,y,z: (abs((x+y*1j)/z) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x,y,z: (abs((x+y*1j)/z) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or "
"one of ('lhp','rhp','iuc','ouc')")
else: # to avoid an error later
raise ValueError("dtype %s not understood" % typ)
return sfunction
def _select_function(sort, typ):
if typ in ['F','D']:
if callable(sort):
# assume the user knows what they're doing
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x,y: (np.real(x/y) < 0.0)
elif sort == 'rhp':
sfunction = lambda x,y: (np.real(x/y) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x,y: (abs(x/y) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x,y: (abs(x/y) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or "
"one of ('lhp','rhp','iuc','ouc')")
elif typ in ['f','d']:
if callable(sort):
# assume the user knows what they're doing
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x,y,z: (np.real((x+y*1j)/z) < 0.0)
elif sort == 'rhp':
sfunction = lambda x,y,z: (np.real((x+y*1j)/z) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x,y,z: (abs((x+y*1j)/z) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x,y,z: (abs((x+y*1j)/z) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or "
"one of ('lhp','rhp','iuc','ouc')")
else: # to avoid an error later
raise ValueError("dtype %s not understood" % typ)
return sfunction
def set_bandwidth(self, bw_method=None):
"""Compute the estimator bandwidth with given method.
The new bandwidth calculated after a call to `set_bandwidth` is used
for subsequent evaluations of the estimated density.
Parameters
----------
bw_method : str, scalar or callable, optional
The method used to calculate the estimator bandwidth. This can be
'scott', 'silverman', a scalar constant or a callable. If a
scalar, this will be used directly as `kde.factor`. If a callable,
it should take a `gaussian_kde` instance as only parameter and
return a scalar. If None (default), nothing happens; the current
`kde.covariance_factor` method is kept.
Notes
-----
.. versionadded:: 0.11
Examples
--------
>>> x1 = np.array([-7, -5, 1, 4, 5.])
>>> kde = stats.gaussian_kde(x1)
>>> xs = np.linspace(-10, 10, num=50)
>>> y1 = kde(xs)
>>> kde.set_bandwidth(bw_method='silverman')
>>> y2 = kde(xs)
>>> kde.set_bandwidth(bw_method=kde.factor / 3.)
>>> y3 = kde(xs)
>>> fig = plt.figure()
>>> ax = fig.add_subplot(111)
>>> ax.plot(x1, np.ones(x1.shape) / (4. * x1.size), 'bo',
... label='Data points (rescaled)')
>>> ax.plot(xs, y1, label='Scott (default)')
>>> ax.plot(xs, y2, label='Silverman')
>>> ax.plot(xs, y3, label='Const (1/3 * Silverman)')
>>> ax.legend()
>>> plt.show()
"""
if bw_method is None:
pass
elif bw_method == 'scott':
self.covariance_factor = self.scotts_factor
elif bw_method == 'silverman':
self.covariance_factor = self.silverman_factor
elif np.isscalar(bw_method) and not isinstance(bw_method, string_types):
self._bw_method = 'use constant'
self.covariance_factor = lambda: bw_method
elif callable(bw_method):
self._bw_method = bw_method
self.covariance_factor = lambda: self._bw_method(self)
else:
msg = "`bw_method` should be 'scott', 'silverman', a scalar " \
"or a callable."
raise ValueError(msg)
self._compute_covariance()
def set_bandwidth(self, bw_method=None):
"""Compute the estimator bandwidth with given method.
The new bandwidth calculated after a call to `set_bandwidth` is used
for subsequent evaluations of the estimated density.
Parameters
----------
bw_method : str, scalar or callable, optional
The method used to calculate the estimator bandwidth. This can be
'scott', 'silverman', a scalar constant or a callable. If a
scalar, this will be used directly as `kde.factor`. If a callable,
it should take a `gaussian_kde` instance as only parameter and
return a scalar. If None (default), nothing happens; the current
`kde.covariance_factor` method is kept.
Notes
-----
.. versionadded:: 0.11
Examples
--------
>>> x1 = np.array([-7, -5, 1, 4, 5.])
>>> kde = stats.gaussian_kde(x1)
>>> xs = np.linspace(-10, 10, num=50)
>>> y1 = kde(xs)
>>> kde.set_bandwidth(bw_method='silverman')
>>> y2 = kde(xs)
>>> kde.set_bandwidth(bw_method=kde.factor / 3.)
>>> y3 = kde(xs)
>>> fig = plt.figure()
>>> ax = fig.add_subplot(111)
>>> ax.plot(x1, np.ones(x1.shape) / (4. * x1.size), 'bo',
... label='Data points (rescaled)')
>>> ax.plot(xs, y1, label='Scott (default)')
>>> ax.plot(xs, y2, label='Silverman')
>>> ax.plot(xs, y3, label='Const (1/3 * Silverman)')
>>> ax.legend()
>>> plt.show()
"""
if bw_method is None:
pass
elif bw_method == 'scott':
self.covariance_factor = self.scotts_factor
elif bw_method == 'silverman':
self.covariance_factor = self.silverman_factor
elif np.isscalar(bw_method) and not isinstance(bw_method, string_types):
self._bw_method = 'use constant'
self.covariance_factor = lambda: bw_method
elif callable(bw_method):
self._bw_method = bw_method
self.covariance_factor = lambda: self._bw_method(self)
else:
msg = "`bw_method` should be 'scott', 'silverman', a scalar " \
"or a callable."
raise ValueError(msg)
self._compute_covariance()
def _init_function(self, r): # 选择径向基函数使用的方程，最终返回径向基函数用r计算后的距离
if isinstance(self.function,
str): # 选择已提供的径向基函数形式，若self.function是一个字符串
self.function = self.function.lower() #字符串调整为小写
_mapped = {
'inverse': 'inverse_multiquadric',
'inverse multiquadric': 'inverse_multiquadric',
'thin-plate': 'thin_plate'
} #函数字典
if self.function in _mapped: #若self.function在函数字典内，变为函数名
self.function = _mapped[self.function]
func_name = "_h_" + self.function #调整函数真名
if hasattr(self, func_name): #若类中有该函数
self._function = getattr(self, func_name) #调用该函数
else: #若类中无函数真名
functionlist = [
x[3:] for x in dir(self) if x.startswith('_h_')
] #指向所有函数名
raise ValueError("function must be a callable or one of "
+ #抛出赋值异常，函数必须为一可调用对象
", ".join(functionlist))
self._function = getattr(self, "_h_" + self.function)
elif callable(self.function): # 若方程为一个可调用的对象：反正这段看不懂我特么就不看了吧算了我特么还是看看吧
allow_one = False #
if hasattr(self.function, 'func_code') or \
hasattr(self.function, '__code__'): #若方程对象有func_code属性或__code__属性
val = self.function # val+函数对象
allow_one = True #allow_one设为1
elif hasattr(self.function, "im_func"):
val = get_method_function(self.function)
elif hasattr(self.function, "__call__"):
val = get_method_function(self.function.__call__)
else:
raise ValueError(
"Cannot determine number of arguments to function"
) #抛出赋值异常，不能确定有函数的赋值变量数
argcount = get_function_code(val).co_argcount #获取变量数
if allow_one and argcount == 1:
self._function = self.function
elif argcount == 2:
if sys.version_info[0] >= 3:
self._function = self.function.__get__(self, Rbf)
else:
import new
self._function = new.instancemethod(
self.function, self, Rbf)
else:
raise ValueError(
"Function argument must take 1 or 2 arguments.")
a0 = self._function(r)
if a0.shape != r.shape:
raise ValueError(
"Callable must take array and return array of the same shape")
return a0
def _init_function(self, r):
if isinstance(self.function, str):
self.function = self.function.lower()
_mapped = {
'inverse': 'inverse_multiquadric',
'inverse multiquadric': 'inverse_multiquadric',
'thin-plate': 'thin_plate'
}
if self.function in _mapped:
self.function = _mapped[self.function]
func_name = "_h_" + self.function
if hasattr(self, func_name):
self._function = getattr(self, func_name)
else:
functionlist = [
x[3:] for x in dir(self) if x.startswith('_h_')
]
raise ValueError("function must be a callable or one of " +
", ".join(functionlist))
self._function = getattr(self, "_h_" + self.function)
elif callable(self.function):
allow_one = False
if hasattr(self.function, 'func_code') or \
hasattr(self.function, '__code__'):
val = self.function
allow_one = True
elif hasattr(self.function, "im_func"):
val = get_method_function(self.function)
elif hasattr(self.function, "__call__"):
val = get_method_function(self.function.__call__)
else:
raise ValueError(
"Cannot determine number of arguments to function")
argcount = get_function_code(val).co_argcount
if allow_one and argcount == 1:
self._function = self.function
elif argcount == 2:
if sys.version_info[0] >= 3:
self._function = self.function.__get__(self, Rbf)
else:
import new
self._function = new.instancemethod(
self.function, self, Rbf)
else:
raise ValueError(
"Function argument must take 1 or 2 arguments.")
a0 = self._function(r)
if a0.shape != r.shape:
raise ValueError(
"Callable must take array and return array of the same shape")
return a0
def _init_function(self, r):
if isinstance(self.function, str):
self.function = self.function.lower()
_mapped = {'inverse': 'inverse_multiquadric',
'inverse multiquadric': 'inverse_multiquadric',
'thin-plate': 'thin_plate'}
if self.function in _mapped:
self.function = _mapped[self.function]
func_name = "_h_" + self.function
if hasattr(self, func_name):
self._function = getattr(self, func_name)
else:
functionlist = [x[3:] for x in dir(self) if x.startswith('_h_')]
raise ValueError("function must be a callable or one of " +
", ".join(functionlist))
self._function = getattr(self, "_h_"+self.function)
elif callable(self.function):
allow_one = False
if hasattr(self.function, 'func_code') or \
hasattr(self.function, '__code__'):
val = self.function
allow_one = True
elif hasattr(self.function, "im_func"):
val = get_method_function(self.function)
elif hasattr(self.function, "__call__"):
val = get_method_function(self.function.__call__)
else:
raise ValueError("Cannot determine number of arguments to function")
argcount = get_function_code(val).co_argcount
if allow_one and argcount == 1:
self._function = self.function
elif argcount == 2:
if sys.version_info[0] >= 3:
self._function = self.function.__get__(self, Rbf)
else:
import new
self._function = new.instancemethod(self.function, self,
Rbf)
else:
raise ValueError("Function argument must take 1 or 2 arguments.")
a0 = self._function(r)
if a0.shape != r.shape:
raise ValueError("Callable must take array and return array of the same shape")
return a0
def binned_statistic_dd(sample, values, statistic='mean', bins=10, range=None):
"""
Compute a multidimensional binned statistic for a set of data.
This is a generalization of a histogramdd function. A histogram divides
the space into bins, and returns the count of the number of points in
each bin. This function allows the computation of the sum, mean, median,
or other statistic of the values within each bin.
.. versionadded:: 0.11.0
Parameters
----------
sample : array_like
Data to histogram passed as a sequence of D arrays of length N, or
as an (N,D) array.
values : array_like
The values on which the statistic will be computed. This must be
the same shape as x.
statistic : string or callable, optional
The statistic to compute (default is 'mean').
The following statistics are available:
* 'mean' : compute the mean of values for points within each bin.
Empty bins will be represented by NaN.
* 'median' : compute the median of values for points within each
bin. Empty bins will be represented by NaN.
* 'count' : compute the count of points within each bin. This is
identical to an unweighted histogram. `values` array is not
referenced.
* 'sum' : compute the sum of values for points within each bin.
This is identical to a weighted histogram.
* function : a user-defined function which takes a 1D array of
values, and outputs a single numerical statistic. This function
will be called on the values in each bin. Empty bins will be
represented by function([]), or NaN if this returns an error.
bins : sequence or int, optional
The bin specification:
* A sequence of arrays describing the bin edges along each dimension.
* The number of bins for each dimension (nx, ny, ... =bins)
* The number of bins for all dimensions (nx=ny=...=bins).
range : sequence, optional
A sequence of lower and upper bin edges to be used if the edges are
not given explicitely in `bins`. Defaults to the minimum and maximum
values along each dimension.
Returns
-------
statistic : ndarray, shape(nx1, nx2, nx3,...)
The values of the selected statistic in each two-dimensional bin
edges : list of ndarrays
A list of D arrays describing the (nxi + 1) bin edges for each
dimension
binnumber : 1-D ndarray of ints
This assigns to each observation an integer that represents the bin
in which this observation falls. Array has the same length as values.
See Also
--------
np.histogramdd, binned_statistic, binned_statistic_2d
"""
if type(statistic) == str:
if statistic not in ['mean', 'median', 'count', 'sum', 'std']:
raise ValueError('unrecognized statistic "%s"' % statistic)
elif callable(statistic):
pass
else:
raise ValueError("statistic not understood")
# This code is based on np.histogramdd
try:
# Sample is an ND-array.
N, D = sample.shape
except (AttributeError, ValueError):
# Sample is a sequence of 1D arrays.
sample = np.atleast_2d(sample).T
N, D = sample.shape
nbin = np.empty(D, int)
edges = D * [None]
dedges = D * [None]
try:
M = len(bins)
if M != D:
raise AttributeError('The dimension of bins must be equal '
'to the dimension of the sample x.')
except TypeError:
bins = D * [bins]
# Select range for each dimension
# Used only if number of bins is given.
if range is None:
smin = np.atleast_1d(np.array(sample.min(0), float))
smax = np.atleast_1d(np.array(sample.max(0), float))
else:
smin = np.zeros(D)
smax = np.zeros(D)
for i in np.arange(D):
smin[i], smax[i] = range[i]
# Make sure the bins have a finite width.
for i in np.arange(len(smin)):
if smin[i] == smax[i]:
smin[i] = smin[i] - .5
smax[i] = smax[i] + .5
# Create edge arrays
for i in np.arange(D):
if np.isscalar(bins[i]):
nbin[i] = bins[i] + 2 # +2 for outlier bins
edges[i] = np.linspace(smin[i], smax[i], nbin[i] - 1)
else:
edges[i] = np.asarray(bins[i], float)
nbin[i] = len(edges[i]) + 1 # +1 for outlier bins
dedges[i] = np.diff(edges[i])
nbin = np.asarray(nbin)
# Compute the bin number each sample falls into.
Ncount = {}
for i in np.arange(D):
Ncount[i] = np.digitize(sample[:, i], edges[i])
# Using digitize, values that fall on an edge are put in the right bin.
# For the rightmost bin, we want values equal to the right
# edge to be counted in the last bin, and not as an outlier.
for i in np.arange(D):
# Rounding precision
decimal = int(-np.log10(dedges[i].min())) + 6
# Find which points are on the rightmost edge.
on_edge = np.where(
np.around(sample[:, i], decimal) == np.around(
edges[i][-1], decimal))[0]
# Shift these points one bin to the left.
Ncount[i][on_edge] -= 1
# Compute the sample indices in the flattened statistic matrix.
ni = nbin.argsort()
xy = np.zeros(N, int)
for i in np.arange(0, D - 1):
xy += Ncount[ni[i]] * nbin[ni[i + 1:]].prod()
xy += Ncount[ni[-1]]
result = np.empty(nbin.prod(), float)
if statistic == 'mean':
result.fill(np.nan)
flatcount = np.bincount(xy, None)
flatsum = np.bincount(xy, values)
a = flatcount.nonzero()
result[a] = flatsum[a] / flatcount[a]
elif statistic == 'std':
result.fill(0)
flatcount = np.bincount(xy, None)
flatsum = np.bincount(xy, values)
flatsum2 = np.bincount(xy, values**2)
a = flatcount.nonzero()
result[a] = np.sqrt(flatsum2[a] / flatcount[a] -
(flatsum[a] / flatcount[a])**2)
elif statistic == 'count':
result.fill(0)
flatcount = np.bincount(xy, None)
a = np.arange(len(flatcount))
result[a] = flatcount
elif statistic == 'sum':
result.fill(0)
flatsum = np.bincount(xy, values)
a = np.arange(len(flatsum))
result[a] = flatsum
elif statistic == 'median':
result.fill(np.nan)
for i in np.unique(xy):
result[i] = np.median(values[xy == i])
elif callable(statistic):
old = np.seterr(invalid='ignore')
try:
null = statistic([])
except:
null = np.nan
np.seterr(**old)
result.fill(null)
for i in np.unique(xy):
result[i] = statistic(values[xy == i])
# Shape into a proper matrix
result = result.reshape(np.sort(nbin))
for i in np.arange(nbin.size):
j = ni.argsort()[i]
result = result.swapaxes(i, j)
ni[i], ni[j] = ni[j], ni[i]
# Remove outliers (indices 0 and -1 for each dimension).
core = D * [slice(1, -1)]
result = result[core]
if (result.shape != nbin - 2).any():
raise RuntimeError('Internal Shape Error')
return result, edges, xy
def fmin_cobyla(func, x0, cons, args=(), consargs=None, rhobeg=1.0,
rhoend=1e-4, iprint=1, maxfun=1000, disp=None):
"""
Minimize a function using the Constrained Optimization BY Linear
Approximation (COBYLA) method. This method wraps a FORTRAN
implentation of the algorithm.
Parameters
----------
func : callable
Function to minimize. In the form func(x, \\*args).
x0 : ndarray
Initial guess.
cons : sequence
Constraint functions; must all be ``>=0`` (a single function
if only 1 constraint). Each function takes the parameters `x`
as its first argument.
args : tuple
Extra arguments to pass to function.
consargs : tuple
Extra arguments to pass to constraint functions (default of None means
use same extra arguments as those passed to func).
Use ``()`` for no extra arguments.
rhobeg :
Reasonable initial changes to the variables.
rhoend :
Final accuracy in the optimization (not precisely guaranteed). This
is a lower bound on the size of the trust region.
iprint : {0, 1, 2, 3}
Controls the frequency of output; 0 implies no output. Deprecated.
disp : {0, 1, 2, 3}
Over-rides the iprint interface. Preferred.
maxfun : int
Maximum number of function evaluations.
Returns
-------
x : ndarray
The argument that minimises `f`.
See also
--------
minimize: Interface to minimization algorithms for multivariate
functions. See the 'COBYLA' `method` in particular.
Notes
-----
This algorithm is based on linear approximations to the objective
function and each constraint. We briefly describe the algorithm.
Suppose the function is being minimized over k variables. At the
jth iteration the algorithm has k+1 points v_1, ..., v_(k+1),
an approximate solution x_j, and a radius RHO_j.
(i.e. linear plus a constant) approximations to the objective
function and constraint functions such that their function values
agree with the linear approximation on the k+1 points v_1,.., v_(k+1).
This gives a linear program to solve (where the linear approximations
of the constraint functions are constrained to be non-negative).
However the linear approximations are likely only good
approximations near the current simplex, so the linear program is
given the further requirement that the solution, which
will become x_(j+1), must be within RHO_j from x_j. RHO_j only
decreases, never increases. The initial RHO_j is rhobeg and the
final RHO_j is rhoend. In this way COBYLA's iterations behave
like a trust region algorithm.
Additionally, the linear program may be inconsistent, or the
approximation may give poor improvement. For details about
how these issues are resolved, as well as how the points v_i are
updated, refer to the source code or the references below.
References
----------
Powell M.J.D. (1994), "A direct search optimization method that models
the objective and constraint functions by linear interpolation.", in
Advances in Optimization and Numerical Analysis, eds. S. Gomez and
J-P Hennart, Kluwer Academic (Dordrecht), pp. 51-67
Powell M.J.D. (1998), "Direct search algorithms for optimization
calculations", Acta Numerica 7, 287-336
Powell M.J.D. (2007), "A view of algorithms for optimization without
derivatives", Cambridge University Technical Report DAMTP 2007/NA03
Examples
--------
Minimize the objective function f(x,y) = x*y subject
to the constraints x**2 + y**2 < 1 and y > 0::
>>> def objective(x):
... return x[0]*x[1]
...
>>> def constr1(x):
... return 1 - (x[0]**2 + x[1]**2)
...
>>> def constr2(x):
... return x[1]
...
>>> fmin_cobyla(objective, [0.0, 0.1], [constr1, constr2], rhoend=1e-7)
Normal return from subroutine COBYLA
NFVALS = 64 F =-5.000000E-01 MAXCV = 1.998401E-14
X =-7.071069E-01 7.071067E-01
array([-0.70710685, 0.70710671])
The exact solution is (-sqrt(2)/2, sqrt(2)/2).
"""
err = "cons must be a sequence of callable functions or a single"\
" callable function."
try:
len(cons)
except TypeError:
if callable(cons):
cons = [cons]
else:
raise TypeError(err)
else:
for thisfunc in cons:
if not callable(thisfunc):
raise TypeError(err)
if consargs is None:
consargs = args
# build constraints
con = tuple({'type': 'ineq', 'fun': c, 'args': consargs} for c in cons)
# options
if disp is not None:
iprint = disp
opts = {'rhobeg': rhobeg,
'tol': rhoend,
'iprint': iprint,
'disp': iprint != 0,
'maxiter': maxfun}
return _minimize_cobyla(func, x0, args, constraints=con,
**opts)['x']
def binned_statistic_dd(sample, values, statistic='mean',
bins=10, range=None):
"""
Compute a multidimensional binned statistic for a set of data.
This is a generalization of a histogramdd function. A histogram divides
the space into bins, and returns the count of the number of points in
each bin. This function allows the computation of the sum, mean, median,
or other statistic of the values within each bin.
.. versionadded:: 0.11.0
Parameters
----------
sample : array_like
Data to histogram passed as a sequence of D arrays of length N, or
as an (N,D) array.
values : array_like
The values on which the statistic will be computed. This must be
the same shape as x.
statistic : string or callable, optional
The statistic to compute (default is 'mean').
The following statistics are available:
* 'mean' : compute the mean of values for points within each bin.
Empty bins will be represented by NaN.
* 'median' : compute the median of values for points within each
bin. Empty bins will be represented by NaN.
* 'count' : compute the count of points within each bin. This is
identical to an unweighted histogram. `values` array is not
referenced.
* 'sum' : compute the sum of values for points within each bin.
This is identical to a weighted histogram.
* function : a user-defined function which takes a 1D array of
values, and outputs a single numerical statistic. This function
will be called on the values in each bin. Empty bins will be
represented by function([]), or NaN if this returns an error.
bins : sequence or int, optional
The bin specification:
* A sequence of arrays describing the bin edges along each dimension.
* The number of bins for each dimension (nx, ny, ... =bins)
* The number of bins for all dimensions (nx=ny=...=bins).
range : sequence, optional
A sequence of lower and upper bin edges to be used if the edges are
not given explicitely in `bins`. Defaults to the minimum and maximum
values along each dimension.
Returns
-------
statistic : ndarray, shape(nx1, nx2, nx3,...)
The values of the selected statistic in each two-dimensional bin
edges : list of ndarrays
A list of D arrays describing the (nxi + 1) bin edges for each
dimension
binnumber : 1-D ndarray of ints
This assigns to each observation an integer that represents the bin
in which this observation falls. Array has the same length as values.
See Also
--------
np.histogramdd, binned_statistic, binned_statistic_2d
"""
if type(statistic) == str:
if statistic not in ['mean', 'median', 'count', 'sum', 'std']:
raise ValueError('unrecognized statistic "%s"' % statistic)
elif callable(statistic):
pass
else:
raise ValueError("statistic not understood")
# This code is based on np.histogramdd
try:
# Sample is an ND-array.
N, D = sample.shape
except (AttributeError, ValueError):
# Sample is a sequence of 1D arrays.
sample = np.atleast_2d(sample).T
N, D = sample.shape
nbin = np.empty(D, int)
edges = D * [None]
dedges = D * [None]
try:
M = len(bins)
if M != D:
raise AttributeError('The dimension of bins must be equal '
'to the dimension of the sample x.')
except TypeError:
bins = D * [bins]
# Select range for each dimension
# Used only if number of bins is given.
if range is None:
smin = np.atleast_1d(np.array(sample.min(0), float))
smax = np.atleast_1d(np.array(sample.max(0), float))
else:
smin = np.zeros(D)
smax = np.zeros(D)
for i in np.arange(D):
smin[i], smax[i] = range[i]
# Make sure the bins have a finite width.
for i in np.arange(len(smin)):
if smin[i] == smax[i]:
smin[i] = smin[i] - .5
smax[i] = smax[i] + .5
# Create edge arrays
for i in np.arange(D):
if np.isscalar(bins[i]):
nbin[i] = bins[i] + 2 # +2 for outlier bins
edges[i] = np.linspace(smin[i], smax[i], nbin[i] - 1)
else:
edges[i] = np.asarray(bins[i], float)
nbin[i] = len(edges[i]) + 1 # +1 for outlier bins
dedges[i] = np.diff(edges[i])
nbin = np.asarray(nbin)
# Compute the bin number each sample falls into.
Ncount = {}
for i in np.arange(D):
Ncount[i] = np.digitize(sample[:, i], edges[i])
# Using digitize, values that fall on an edge are put in the right bin.
# For the rightmost bin, we want values equal to the right
# edge to be counted in the last bin, and not as an outlier.
for i in np.arange(D):
# Rounding precision
decimal = int(-np.log10(dedges[i].min())) + 6
# Find which points are on the rightmost edge.
on_edge = np.where(np.around(sample[:, i], decimal)
== np.around(edges[i][-1], decimal))[0]
# Shift these points one bin to the left.
Ncount[i][on_edge] -= 1
# Compute the sample indices in the flattened statistic matrix.
ni = nbin.argsort()
xy = np.zeros(N, int)
for i in np.arange(0, D - 1):
xy += Ncount[ni[i]] * nbin[ni[i + 1:]].prod()
xy += Ncount[ni[-1]]
result = np.empty(nbin.prod(), float)
if statistic == 'mean':
result.fill(np.nan)
flatcount = np.bincount(xy, None)
flatsum = np.bincount(xy, values)
a = flatcount.nonzero()
result[a] = flatsum[a] / flatcount[a]
elif statistic == 'std':
result.fill(0)
flatcount = np.bincount(xy, None)
flatsum = np.bincount(xy, values)
flatsum2 = np.bincount(xy, values ** 2)
a = flatcount.nonzero()
result[a] = np.sqrt(flatsum2[a] / flatcount[a]
- (flatsum[a] / flatcount[a]) ** 2)
elif statistic == 'count':
result.fill(0)
flatcount = np.bincount(xy, None)
a = np.arange(len(flatcount))
result[a] = flatcount
elif statistic == 'sum':
result.fill(0)
flatsum = np.bincount(xy, values)
a = np.arange(len(flatsum))
result[a] = flatsum
elif statistic == 'median':
result.fill(np.nan)
for i in np.unique(xy):
result[i] = np.median(values[xy == i])
elif callable(statistic):
old = np.seterr(invalid='ignore')
try:
null = statistic([])
except:
null = np.nan
np.seterr(**old)
result.fill(null)
for i in np.unique(xy):
result[i] = statistic(values[xy == i])
# Shape into a proper matrix
result = result.reshape(np.sort(nbin))
for i in np.arange(nbin.size):
j = ni.argsort()[i]
result = result.swapaxes(i, j)
ni[i], ni[j] = ni[j], ni[i]
# Remove outliers (indices 0 and -1 for each dimension).
core = D * [slice(1, -1)]
result = result[core]
if (result.shape != nbin - 2).any():
raise RuntimeError('Internal Shape Error')
return result, edges, xy
def resample(x, num, t=None, axis=0, window=None):
"""
Resample `x` to `num` samples using Fourier method along the given axis.
The resampled signal starts at the same value as `x` but is sampled
with a spacing of ``len(x) / num * (spacing of x)``. Because a
Fourier method is used, the signal is assumed to be periodic.
Parameters
----------
x : array_like
The data to be resampled.
num : int
The number of samples in the resampled signal.
t : array_like, optional
If `t` is given, it is assumed to be the sample positions
associated with the signal data in `x`.
axis : int, optional
The axis of `x` that is resampled. Default is 0.
window : array_like, callable, string, float, or tuple, optional
Specifies the window applied to the signal in the Fourier
domain. See below for details.
Returns
-------
resampled_x or (resampled_x, resampled_t)
Either the resampled array, or, if `t` was given, a tuple
containing the resampled array and the corresponding resampled
positions.
Notes
-----
The argument `window` controls a Fourier-domain window that tapers
the Fourier spectrum before zero-padding to alleviate ringing in
the resampled values for sampled signals you didn't intend to be
interpreted as band-limited.
If `window` is a function, then it is called with a vector of inputs
indicating the frequency bins (i.e. fftfreq(x.shape[axis]) ).
If `window` is an array of the same length as `x.shape[axis]` it is
assumed to be the window to be applied directly in the Fourier
domain (with dc and low-frequency first).
For any other type of `window`, the function `scipy.signal.get_window`
is called to generate the window.
The first sample of the returned vector is the same as the first
sample of the input vector. The spacing between samples is changed
from dx to:
dx * len(x) / num
If `t` is not None, then it represents the old sample positions,
and the new sample positions will be returned as well as the new
samples.
"""
x = asarray(x)
X = fft(x, axis=axis)
Nx = x.shape[axis]
if window is not None:
if callable(window):
W = window(fftfreq(Nx))
elif isinstance(window, ndarray) and window.shape == (Nx, ):
W = window
else:
W = ifftshift(get_window(window, Nx))
newshape = ones(len(x.shape))
newshape[axis] = len(W)
W.shape = newshape
X = X * W
sl = [slice(None)] * len(x.shape)
newshape = list(x.shape)
newshape[axis] = num
N = int(np.minimum(num, Nx))
Y = zeros(newshape, 'D')
sl[axis] = slice(0, (N + 1) // 2)
Y[sl] = X[sl]
sl[axis] = slice(-(N - 1) // 2, None)
Y[sl] = X[sl]
y = ifft(Y, axis=axis) * (float(num) / float(Nx))
if x.dtype.char not in ['F', 'D']:
y = y.real
if t is None:
return y
else:
new_t = arange(0, num) * (t[1] - t[0]) * Nx / float(num) + t[0]
return y, new_t
def asjacobian(J):
"""
Convert given object to one suitable for use as a Jacobian.
"""
spsolve = scipy.sparse.linalg.spsolve
if isinstance(J, Jacobian):
return J
elif inspect.isclass(J) and issubclass(J, Jacobian):
return J()
elif isinstance(J, np.ndarray):
if J.ndim > 2:
raise ValueError('array must have rank <= 2')
J = np.atleast_2d(np.asarray(J))
if J.shape[0] != J.shape[1]:
raise ValueError('array must be square')
return Jacobian(matvec=lambda v: dot(J, v),
rmatvec=lambda v: dot(J.conj().T, v),
solve=lambda v: solve(J, v),
rsolve=lambda v: solve(J.conj().T, v),
dtype=J.dtype, shape=J.shape)
elif scipy.sparse.isspmatrix(J):
if J.shape[0] != J.shape[1]:
raise ValueError('matrix must be square')
return Jacobian(matvec=lambda v: J*v,
rmatvec=lambda v: J.conj().T * v,
solve=lambda v: spsolve(J, v),
rsolve=lambda v: spsolve(J.conj().T, v),
dtype=J.dtype, shape=J.shape)
elif hasattr(J, 'shape') and hasattr(J, 'dtype') and hasattr(J, 'solve'):
return Jacobian(matvec=getattr(J, 'matvec'),
rmatvec=getattr(J, 'rmatvec'),
solve=J.solve,
rsolve=getattr(J, 'rsolve'),
update=getattr(J, 'update'),
setup=getattr(J, 'setup'),
dtype=J.dtype,
shape=J.shape)
elif callable(J):
# Assume it's a function J(x) that returns the Jacobian
class Jac(Jacobian):
def update(self, x, F):
self.x = x
def solve(self, v, tol=0):
m = J(self.x)
if isinstance(m, np.ndarray):
return solve(m, v)
elif scipy.sparse.isspmatrix(m):
return spsolve(m, v)
else:
raise ValueError("Unknown matrix type")
def matvec(self, v):
m = J(self.x)
if isinstance(m, np.ndarray):
return dot(m, v)
elif scipy.sparse.isspmatrix(m):
return m*v
else:
raise ValueError("Unknown matrix type")
def rsolve(self, v, tol=0):
m = J(self.x)
if isinstance(m, np.ndarray):
return solve(m.conj().T, v)
elif scipy.sparse.isspmatrix(m):
return spsolve(m.conj().T, v)
else:
raise ValueError("Unknown matrix type")
def rmatvec(self, v):
m = J(self.x)
if isinstance(m, np.ndarray):
return dot(m.conj().T, v)
elif scipy.sparse.isspmatrix(m):
return m.conj().T * v
else:
raise ValueError("Unknown matrix type")
return Jac()
elif isinstance(J, str):
return dict(broyden1=BroydenFirst,
broyden2=BroydenSecond,
anderson=Anderson,
diagbroyden=DiagBroyden,
linearmixing=LinearMixing,
excitingmixing=ExcitingMixing,
krylov=KrylovJacobian)[J]()
else:
raise TypeError('Cannot convert object to a Jacobian')
def asjacobian(J):
"""
Convert given object to one suitable for use as a Jacobian.
"""
spsolve = scipy.sparse.linalg.spsolve
if isinstance(J, Jacobian):
return J
elif inspect.isclass(J) and issubclass(J, Jacobian):
return J()
elif isinstance(J, np.ndarray):
if J.ndim > 2:
raise ValueError('array must have rank <= 2')
J = np.atleast_2d(np.asarray(J))
if J.shape[0] != J.shape[1]:
raise ValueError('array must be square')
return Jacobian(matvec=lambda v: dot(J, v),
rmatvec=lambda v: dot(J.conj().T, v),
solve=lambda v: solve(J, v),
rsolve=lambda v: solve(J.conj().T, v),
dtype=J.dtype,
shape=J.shape)
elif scipy.sparse.isspmatrix(J):
if J.shape[0] != J.shape[1]:
raise ValueError('matrix must be square')
return Jacobian(matvec=lambda v: J * v,
rmatvec=lambda v: J.conj().T * v,
solve=lambda v: spsolve(J, v),
rsolve=lambda v: spsolve(J.conj().T, v),
dtype=J.dtype,
shape=J.shape)
elif hasattr(J, 'shape') and hasattr(J, 'dtype') and hasattr(J, 'solve'):
return Jacobian(matvec=getattr(J, 'matvec'),
rmatvec=getattr(J, 'rmatvec'),
solve=J.solve,
rsolve=getattr(J, 'rsolve'),
update=getattr(J, 'update'),
setup=getattr(J, 'setup'),
dtype=J.dtype,
shape=J.shape)
elif callable(J):
# Assume it's a function J(x) that returns the Jacobian
class Jac(Jacobian):
def update(self, x, F):
self.x = x
def solve(self, v, tol=0):
m = J(self.x)
if isinstance(m, np.ndarray):
return solve(m, v)
elif scipy.sparse.isspmatrix(m):
return spsolve(m, v)
else:
raise ValueError("Unknown matrix type")
def matvec(self, v):
m = J(self.x)
if isinstance(m, np.ndarray):
return dot(m, v)
elif scipy.sparse.isspmatrix(m):
return m * v
else:
raise ValueError("Unknown matrix type")
def rsolve(self, v, tol=0):
m = J(self.x)
if isinstance(m, np.ndarray):
return solve(m.conj().T, v)
elif scipy.sparse.isspmatrix(m):
return spsolve(m.conj().T, v)
else:
raise ValueError("Unknown matrix type")
def rmatvec(self, v):
m = J(self.x)
if isinstance(m, np.ndarray):
return dot(m.conj().T, v)
elif scipy.sparse.isspmatrix(m):
return m.conj().T * v
else:
raise ValueError("Unknown matrix type")
return Jac()
elif isinstance(J, str):
return dict(broyden1=BroydenFirst,
broyden2=BroydenSecond,
anderson=Anderson,
diagbroyden=DiagBroyden,
linearmixing=LinearMixing,
excitingmixing=ExcitingMixing,
krylov=KrylovJacobian)[J]()
else:
raise TypeError('Cannot convert object to a Jacobian')
def nquad(func, ranges, args=None, opts=None):
"""
Integration over multiple variables.
Wraps `quad` to enable integration over multiple variables.
Various options allow improved integration of discontinuous functions, as
well as the use of weighted integration, and generally finer control of the
integration process.
Parameters
----------
func : callable
The function to be integrated. Has arguments of ``x0, ... xn``,
``t0, tm``, where integration is carried out over ``x0, ... xn``, which
must be floats. Function signature should be
``func(x0, x1, ..., xn, t0, t1, ..., tm)``. Integration is carried out
in order. That is, integration over ``x0`` is the innermost integral,
and ``xn`` is the outermost.
ranges : iterable object
Each element of ranges may be either a sequence of 2 numbers, or else
a callable that returns such a sequence. ``ranges[0]`` corresponds to
integration over x0, and so on. If an element of ranges is a callable,
then it will be called with all of the integration arguments available.
e.g. if ``func = f(x0, x1, x2)``, then ``ranges[0]`` may be defined as
either ``(a, b)`` or else as ``(a, b) = range0(x1, x2)``.
args : iterable object, optional
Additional arguments ``t0, ..., tn``, required by `func`.
opts : iterable object or dict, optional
Options to be passed to `quad`. May be empty, a dict, or
a sequence of dicts or functions that return a dict. If empty, the
default options from scipy.integrate.quadare used. If a dict, the same
options are used for all levels of integraion. If a sequence, then each
element of the sequence corresponds to a particular integration. e.g.
opts[0] corresponds to integration over x0, and so on. The available
options together with their default values are:
- epsabs = 1.49e-08
- epsrel = 1.49e-08
- limit = 50
- points = None
- weight = None
- wvar = None
- wopts = None
The ``full_output`` option from `quad` is unavailable, due to the
complexity of handling the large amount of data such an option would
return for this kind of nested integration. For more information on
these options, see `quad` and `quad_explain`.
Returns
-------
result : float
The result of the integration.
abserr : float
The maximum of the estimates of the absolute error in the various
integration results.
See Also
--------
quad : 1-dimensional numerical integration
dblquad, tplquad : double and triple integrals
fixed_quad : fixed-order Gaussian quadrature
quadrature : adaptive Gaussian quadrature
Examples
--------
>>> from scipy import integrate
>>> func = lambda x0,x1,x2,x3 : x0**2 + x1*x2 - x3**3 + np.sin(x0) + (
... 1 if (x0-.2*x3-.5-.25*x1>0) else 0)
>>> points = [[lambda (x1,x2,x3) : 0.2*x3 + 0.5 + 0.25*x1], [], [], []]
>>> def opts0(*args, **kwargs):
... return {'points':[0.2*args[2] + 0.5 + 0.25*args[0]]}
>>> integrate.nquad(func, [[0,1], [-1,1], [.13,.8], [-.15,1]],
... opts=[opts0,{},{},{}])
(1.5267454070738633, 2.9437360001402324e-14)
>>> scale = .1
>>> def func2(x0, x1, x2, x3, t0, t1):
... return x0*x1*x3**2 + np.sin(x2) + 1 + (1 if x0+t1*x1-t0>0 else 0)
>>> def lim0(x1, x2, x3, t0, t1):
... return [scale * (x1**2 + x2 + np.cos(x3)*t0*t1 + 1) - 1,
... scale * (x1**2 + x2 + np.cos(x3)*t0*t1 + 1) + 1]
>>> def lim1(x2, x3, t0, t1):
... return [scale * (t0*x2 + t1*x3) - 1,
... scale * (t0*x2 + t1*x3) + 1]
>>> def lim2(x3, t0, t1):
... return [scale * (x3 + t0**2*t1**3) - 1,
... scale * (x3 + t0**2*t1**3) + 1]
>>> def lim3(t0, t1):
... return [scale * (t0+t1) - 1, scale * (t0+t1) + 1]
>>> def opts0(x1, x2, x3, t0, t1):
... return {'points' : [t0 - t1*x1]}
>>> def opts1(x2, x3, t0, t1):
... return {}
>>> def opts2(x3, t0, t1):
... return {}
>>> def opts3(t0, t1):
... return {}
>>> integrate.nquad(func2, [lim0, lim1, lim2, lim3], args=(0,0),
opts=[opts0, opts1, opts2, opts3])
(25.066666666666666, 2.7829590483937256e-13)
"""
depth = len(ranges)
ranges = [rng if callable(rng) else _RangeFunc(rng) for rng in ranges]
if args is None:
args = ()
if opts is None:
opts = [dict([])] * depth
if isinstance(opts, dict):
opts = [opts] * depth
else:
opts = [opt if callable(opt) else _OptFunc(opt) for opt in opts]
return _NQuad(func, ranges, opts).integrate(*args)
def nquad(func, ranges, args=None, opts=None):
"""
Integration over multiple variables.
Wraps `quad` to enable integration over multiple variables.
Various options allow improved integration of discontinuous functions, as
well as the use of weighted integration, and generally finer control of the
integration process.
Parameters
----------
func : callable
The function to be integrated. Has arguments of ``x0, ... xn``,
``t0, tm``, where integration is carried out over ``x0, ... xn``, which
must be floats. Function signature should be
``func(x0, x1, ..., xn, t0, t1, ..., tm)``. Integration is carried out
in order. That is, integration over ``x0`` is the innermost integral,
and ``xn`` is the outermost.
ranges : iterable object
Each element of ranges may be either a sequence of 2 numbers, or else
a callable that returns such a sequence. ``ranges[0]`` corresponds to
integration over x0, and so on. If an element of ranges is a callable,
then it will be called with all of the integration arguments available.
e.g. if ``func = f(x0, x1, x2)``, then ``ranges[0]`` may be defined as
either ``(a, b)`` or else as ``(a, b) = range0(x1, x2)``.
args : iterable object, optional
Additional arguments ``t0, ..., tn``, required by `func`.
opts : iterable object or dict, optional
Options to be passed to `quad`. May be empty, a dict, or
a sequence of dicts or functions that return a dict. If empty, the
default options from scipy.integrate.quadare used. If a dict, the same
options are used for all levels of integraion. If a sequence, then each
element of the sequence corresponds to a particular integration. e.g.
opts[0] corresponds to integration over x0, and so on. The available
options together with their default values are:
- epsabs = 1.49e-08
- epsrel = 1.49e-08
- limit = 50
- points = None
- weight = None
- wvar = None
- wopts = None
The ``full_output`` option from `quad` is unavailable, due to the
complexity of handling the large amount of data such an option would
return for this kind of nested integration. For more information on
these options, see `quad` and `quad_explain`.
Returns
-------
result : float
The result of the integration.
abserr : float
The maximum of the estimates of the absolute error in the various
integration results.
See Also
--------
quad : 1-dimensional numerical integration
dblquad, tplquad : double and triple integrals
fixed_quad : fixed-order Gaussian quadrature
quadrature : adaptive Gaussian quadrature
Examples
--------
>>> from scipy import integrate
>>> func = lambda x0,x1,x2,x3 : x0**2 + x1*x2 - x3**3 + np.sin(x0) + (
... 1 if (x0-.2*x3-.5-.25*x1>0) else 0)
>>> points = [[lambda (x1,x2,x3) : 0.2*x3 + 0.5 + 0.25*x1], [], [], []]
>>> def opts0(*args, **kwargs):
... return {'points':[0.2*args[2] + 0.5 + 0.25*args[0]]}
>>> integrate.nquad(func, [[0,1], [-1,1], [.13,.8], [-.15,1]],
... opts=[opts0,{},{},{}])
(1.5267454070738633, 2.9437360001402324e-14)
>>> scale = .1
>>> def func2(x0, x1, x2, x3, t0, t1):
... return x0*x1*x3**2 + np.sin(x2) + 1 + (1 if x0+t1*x1-t0>0 else 0)
>>> def lim0(x1, x2, x3, t0, t1):
... return [scale * (x1**2 + x2 + np.cos(x3)*t0*t1 + 1) - 1,
... scale * (x1**2 + x2 + np.cos(x3)*t0*t1 + 1) + 1]
>>> def lim1(x2, x3, t0, t1):
... return [scale * (t0*x2 + t1*x3) - 1,
... scale * (t0*x2 + t1*x3) + 1]
>>> def lim2(x3, t0, t1):
... return [scale * (x3 + t0**2*t1**3) - 1,
... scale * (x3 + t0**2*t1**3) + 1]
>>> def lim3(t0, t1):
... return [scale * (t0+t1) - 1, scale * (t0+t1) + 1]
>>> def opts0(x1, x2, x3, t0, t1):
... return {'points' : [t0 - t1*x1]}
>>> def opts1(x2, x3, t0, t1):
... return {}
>>> def opts2(x3, t0, t1):
... return {}
>>> def opts3(t0, t1):
... return {}
>>> integrate.nquad(func2, [lim0, lim1, lim2, lim3], args=(0,0),
opts=[opts0, opts1, opts2, opts3])
(25.066666666666666, 2.7829590483937256e-13)
"""
depth = len(ranges)
ranges = [rng if callable(rng) else _RangeFunc(rng) for rng in ranges]
if args is None:
args = ()
if opts is None:
opts = [dict([])] * depth
if isinstance(opts, dict):
opts = [opts] * depth
else:
opts = [opt if callable(opt) else _OptFunc(opt) for opt in opts]
return _NQuad(func, ranges, opts).integrate(*args)
def schur(a, output='real', lwork=None, overwrite_a=False, sort=None,
check_finite=True):
"""
Compute Schur decomposition of a matrix.
The Schur decomposition is::
A = Z T Z^H
where Z is unitary and T is either upper-triangular, or for real
Schur decomposition (output='real'), quasi-upper triangular. In
the quasi-triangular form, 2x2 blocks describing complex-valued
eigenvalue pairs may extrude from the diagonal.
Parameters
----------
a : (M, M) array_like
Matrix to decompose
output : {'real', 'complex'}, optional
Construct the real or complex Schur decomposition (for real matrices).
lwork : int, optional
Work array size. If None or -1, it is automatically computed.
overwrite_a : bool, optional
Whether to overwrite data in a (may improve performance).
sort : {None, callable, 'lhp', 'rhp', 'iuc', 'ouc'}, optional
Specifies whether the upper eigenvalues should be sorted. A callable
may be passed that, given a eigenvalue, returns a boolean denoting
whether the eigenvalue should be sorted to the top-left (True).
Alternatively, string parameters may be used::
'lhp' Left-hand plane (x.real < 0.0)
'rhp' Right-hand plane (x.real > 0.0)
'iuc' Inside the unit circle (x*x.conjugate() <= 1.0)
'ouc' Outside the unit circle (x*x.conjugate() > 1.0)
Defaults to None (no sorting).
check_finite : boolean, optional
Whether to check that the input matrix contains only finite numbers.
Disabling may give a performance gain, but may result in problems
(crashes, non-termination) if the inputs do contain infinities or NaNs.
Returns
-------
T : (M, M) ndarray
Schur form of A. It is real-valued for the real Schur decomposition.
Z : (M, M) ndarray
An unitary Schur transformation matrix for A.
It is real-valued for the real Schur decomposition.
sdim : int
If and only if sorting was requested, a third return value will
contain the number of eigenvalues satisfying the sort condition.
Raises
------
LinAlgError
Error raised under three conditions:
1. The algorithm failed due to a failure of the QR algorithm to
compute all eigenvalues
2. If eigenvalue sorting was requested, the eigenvalues could not be
reordered due to a failure to separate eigenvalues, usually because
of poor conditioning
3. If eigenvalue sorting was requested, roundoff errors caused the
leading eigenvalues to no longer satisfy the sorting condition
See also
--------
rsf2csf : Convert real Schur form to complex Schur form
"""
if not output in ['real','complex','r','c']:
raise ValueError("argument must be 'real', or 'complex'")
if check_finite:
a1 = asarray_chkfinite(a)
else:
a1 = asarray(a)
if len(a1.shape) != 2 or (a1.shape[0] != a1.shape[1]):
raise ValueError('expected square matrix')
typ = a1.dtype.char
if output in ['complex','c'] and typ not in ['F','D']:
if typ in _double_precision:
a1 = a1.astype('D')
typ = 'D'
else:
a1 = a1.astype('F')
typ = 'F'
overwrite_a = overwrite_a or (_datacopied(a1, a))
gees, = get_lapack_funcs(('gees',), (a1,))
if lwork is None or lwork == -1:
# get optimal work array
result = gees(lambda x: None, a1, lwork=-1)
lwork = result[-2][0].real.astype(numpy.int)
if sort is None:
sort_t = 0
sfunction = lambda x: None
else:
sort_t = 1
if callable(sort):
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x: (numpy.real(x) < 0.0)
elif sort == 'rhp':
sfunction = lambda x: (numpy.real(x) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x: (abs(x) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x: (abs(x) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or " +
"one of ('lhp','rhp','iuc','ouc')")
result = gees(sfunction, a1, lwork=lwork, overwrite_a=overwrite_a,
sort_t=sort_t)
info = result[-1]
if info < 0:
raise ValueError('illegal value in %d-th argument of internal gees'
% -info)
elif info == a1.shape[0] + 1:
raise LinAlgError('Eigenvalues could not be separated for reordering.')
elif info == a1.shape[0] + 2:
raise LinAlgError('Leading eigenvalues do not satisfy sort condition.')
elif info > 0:
raise LinAlgError("Schur form not found. Possibly ill-conditioned.")
if sort_t == 0:
return result[0], result[-3]
else:
return result[0], result[-3], result[1]
def minimize(
fun,
x0,
args=(),
method="BFGS",
jac=None,
hess=None,
hessp=None,
bounds=None,
constraints=(),
tol=None,
callback=None,
options=None,
):
"""
Minimization of scalar function of one or more variables.
.. versionadded:: 0.11.0
Parameters
----------
fun : callable
Objective function.
x0 : ndarray
Initial guess.
args : tuple, optional
Extra arguments passed to the objective function and its
derivatives (Jacobian, Hessian).
method : str, optional
Type of solver. Should be one of
- 'Nelder-Mead'
- 'Powell'
- 'CG'
- 'BFGS'
- 'Newton-CG'
- 'Anneal'
- 'L-BFGS-B'
- 'TNC'
- 'COBYLA'
- 'SLSQP'
- 'dogleg'
- 'trust-ncg'
jac : bool or callable, optional
Jacobian (gradient) of objective function. Only for CG, BFGS,
Newton-CG, L-BFGS-B, TNC, SLSQP, dogleg, trust-ncg.
If `jac` is a Boolean and is True, `fun` is assumed to return the
gradient along with the objective function. If False, the
gradient will be estimated numerically.
`jac` can also be a callable returning the gradient of the
objective. In this case, it must accept the same arguments as `fun`.
hess, hessp : callable, optional
Hessian (matrix of second-order derivatives) of objective function or
Hessian of objective function times an arbitrary vector p. Only for
Newton-CG, dogleg, trust-ncg.
Only one of `hessp` or `hess` needs to be given. If `hess` is
provided, then `hessp` will be ignored. If neither `hess` nor
`hessp` is provided, then the Hessian product will be approximated
using finite differences on `jac`. `hessp` must compute the Hessian
times an arbitrary vector.
bounds : sequence, optional
Bounds for variables (only for L-BFGS-B, TNC and SLSQP).
``(min, max)`` pairs for each element in ``x``, defining
the bounds on that parameter. Use None for one of ``min`` or
``max`` when there is no bound in that direction.
constraints : dict or sequence of dict, optional
Constraints definition (only for COBYLA and SLSQP).
Each constraint is defined in a dictionary with fields:
type : str
Constraint type: 'eq' for equality, 'ineq' for inequality.
fun : callable
The function defining the constraint.
jac : callable, optional
The Jacobian of `fun` (only for SLSQP).
args : sequence, optional
Extra arguments to be passed to the function and Jacobian.
Equality constraint means that the constraint function result is to
be zero whereas inequality means that it is to be non-negative.
Note that COBYLA only supports inequality constraints.
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
options : dict, optional
A dictionary of solver options. All methods accept the following
generic options:
maxiter : int
Maximum number of iterations to perform.
disp : bool
Set to True to print convergence messages.
For method-specific options, see :func:`show_options()`.
callback : callable, optional
Called after each iteration, as ``callback(xk)``, where ``xk`` is the
current parameter vector.
Returns
-------
res : Result
The optimization result represented as a ``Result`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the optimizer exited successfully and
``message`` which describes the cause of the termination. See
`Result` for a description of other attributes.
See also
--------
minimize_scalar : Interface to minimization algorithms for scalar
univariate functions
show_options : Additional options accepted by the solvers
Notes
-----
This section describes the available solvers that can be selected by the
'method' parameter. The default method is *BFGS*.
**Unconstrained minimization**
Method *Nelder-Mead* uses the Simplex algorithm [1]_, [2]_. This
algorithm has been successful in many applications but other algorithms
using the first and/or second derivatives information might be preferred
for their better performances and robustness in general.
Method *Powell* is a modification of Powell's method [3]_, [4]_ which
is a conjugate direction method. It performs sequential one-dimensional
minimizations along each vector of the directions set (`direc` field in
`options` and `info`), which is updated at each iteration of the main
minimization loop. The function need not be differentiable, and no
derivatives are taken.
Method *CG* uses a nonlinear conjugate gradient algorithm by Polak and
Ribiere, a variant of the Fletcher-Reeves method described in [5]_ pp.
120-122. Only the first derivatives are used.
Method *BFGS* uses the quasi-Newton method of Broyden, Fletcher,
Goldfarb, and Shanno (BFGS) [5]_ pp. 136. It uses the first derivatives
only. BFGS has proven good performance even for non-smooth
optimizations. This method also returns an approximation of the Hessian
inverse, stored as `hess_inv` in the Result object.
Method *Newton-CG* uses a Newton-CG algorithm [5]_ pp. 168 (also known
as the truncated Newton method). It uses a CG method to the compute the
search direction. See also *TNC* method for a box-constrained
minimization with a similar algorithm.
Method *Anneal* uses simulated annealing, which is a probabilistic
metaheuristic algorithm for global optimization. It uses no derivative
information from the function being optimized.
Method *dogleg* uses the dog-leg trust-region algorithm [5]_
for unconstrained minimization. This algorithm requires the gradient
and Hessian; furthermore the Hessian is required to be positive definite.
Method *trust-ncg* uses the Newton conjugate gradient trust-region
algorithm [5]_ for unconstrained minimization. This algorithm requires
the gradient and either the Hessian or a function that computes the
product of the Hessian with a given vector.
**Constrained minimization**
Method *L-BFGS-B* uses the L-BFGS-B algorithm [6]_, [7]_ for bound
constrained minimization.
Method *TNC* uses a truncated Newton algorithm [5]_, [8]_ to minimize a
function with variables subject to bounds. This algorithm uses
gradient information; it is also called Newton Conjugate-Gradient. It
differs from the *Newton-CG* method described above as it wraps a C
implementation and allows each variable to be given upper and lower
bounds.
Method *COBYLA* uses the Constrained Optimization BY Linear
Approximation (COBYLA) method [9]_, [10]_, [11]_. The algorithm is
based on linear approximations to the objective function and each
constraint. The method wraps a FORTRAN implementation of the algorithm.
Method *SLSQP* uses Sequential Least SQuares Programming to minimize a
function of several variables with any combination of bounds, equality
and inequality constraints. The method wraps the SLSQP Optimization
subroutine originally implemented by Dieter Kraft [12]_. Note that the
wrapper handles infinite values in bounds by converting them into large
floating values.
References
----------
.. [1] Nelder, J A, and R Mead. 1965. A Simplex Method for Function
Minimization. The Computer Journal 7: 308-13.
.. [2] Wright M H. 1996. Direct search methods: Once scorned, now
respectable, in Numerical Analysis 1995: Proceedings of the 1995
Dundee Biennial Conference in Numerical Analysis (Eds. D F
Griffiths and G A Watson). Addison Wesley Longman, Harlow, UK.
191-208.
.. [3] Powell, M J D. 1964. An efficient method for finding the minimum of
a function of several variables without calculating derivatives. The
Computer Journal 7: 155-162.
.. [4] Press W, S A Teukolsky, W T Vetterling and B P Flannery.
Numerical Recipes (any edition), Cambridge University Press.
.. [5] Nocedal, J, and S J Wright. 2006. Numerical Optimization.
Springer New York.
.. [6] Byrd, R H and P Lu and J. Nocedal. 1995. A Limited Memory
Algorithm for Bound Constrained Optimization. SIAM Journal on
Scientific and Statistical Computing 16 (5): 1190-1208.
.. [7] Zhu, C and R H Byrd and J Nocedal. 1997. L-BFGS-B: Algorithm
778: L-BFGS-B, FORTRAN routines for large scale bound constrained
optimization. ACM Transactions on Mathematical Software 23 (4):
550-560.
.. [8] Nash, S G. Newton-Type Minimization Via the Lanczos Method.
1984. SIAM Journal of Numerical Analysis 21: 770-778.
.. [9] Powell, M J D. A direct search optimization method that models
the objective and constraint functions by linear interpolation.
1994. Advances in Optimization and Numerical Analysis, eds. S. Gomez
and J-P Hennart, Kluwer Academic (Dordrecht), 51-67.
.. [10] Powell M J D. Direct search algorithms for optimization
calculations. 1998. Acta Numerica 7: 287-336.
.. [11] Powell M J D. A view of algorithms for optimization without
derivatives. 2007.Cambridge University Technical Report DAMTP
2007/NA03
.. [12] Kraft, D. A software package for sequential quadratic
programming. 1988. Tech. Rep. DFVLR-FB 88-28, DLR German Aerospace
Center -- Institute for Flight Mechanics, Koln, Germany.
Examples
--------
Let us consider the problem of minimizing the Rosenbrock function. This
function (and its respective derivatives) is implemented in `rosen`
(resp. `rosen_der`, `rosen_hess`) in the `scipy.optimize`.
>>> from scipy.optimize import minimize, rosen, rosen_der
A simple application of the *Nelder-Mead* method is:
>>> x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
>>> res = minimize(rosen, x0, method='Nelder-Mead')
>>> res.x
[ 1. 1. 1. 1. 1.]
Now using the *BFGS* algorithm, using the first derivative and a few
options:
>>> res = minimize(rosen, x0, method='BFGS', jac=rosen_der,
... options={'gtol': 1e-6, 'disp': True})
Optimization terminated successfully.
Current function value: 0.000000
Iterations: 52
Function evaluations: 64
Gradient evaluations: 64
>>> res.x
[ 1. 1. 1. 1. 1.]
>>> print res.message
Optimization terminated successfully.
>>> res.hess
[[ 0.00749589 0.01255155 0.02396251 0.04750988 0.09495377]
[ 0.01255155 0.02510441 0.04794055 0.09502834 0.18996269]
[ 0.02396251 0.04794055 0.09631614 0.19092151 0.38165151]
[ 0.04750988 0.09502834 0.19092151 0.38341252 0.7664427 ]
[ 0.09495377 0.18996269 0.38165151 0.7664427 1.53713523]]
Next, consider a minimization problem with several constraints (namely
Example 16.4 from [5]_). The objective function is:
>>> fun = lambda x: (x[0] - 1)**2 + (x[1] - 2.5)**2
There are three constraints defined as:
>>> cons = ({'type': 'ineq', 'fun': lambda x: x[0] - 2 * x[1] + 2},
... {'type': 'ineq', 'fun': lambda x: -x[0] - 2 * x[1] + 6},
... {'type': 'ineq', 'fun': lambda x: -x[0] + 2 * x[1] + 2})
And variables must be positive, hence the following bounds:
>>> bnds = ((0, None), (0, None))
The optimization problem is solved using the SLSQP method as:
>>> res = minimize(fun, (2, 0), method='SLSQP', bounds=bnds,
... constraints=cons)
It should converge to the theoretical solution (1.4 ,1.7).
"""
meth = method.lower()
if options is None:
options = {}
# check if optional parameters are supported by the selected method
# - jac
if meth in ["nelder-mead", "powell", "anneal", "cobyla"] and bool(jac):
warn("Method %s does not use gradient information (jac)." % method, RuntimeWarning)
# - hess
if meth not in ("newton-cg", "dogleg", "trust-ncg") and hess is not None:
warn("Method %s does not use Hessian information (hess)." % method, RuntimeWarning)
# - hessp
if meth not in ("newton-cg", "dogleg", "trust-ncg") and hessp is not None:
warn("Method %s does not use Hessian-vector product " "information (hessp)." % method, RuntimeWarning)
# - constraints or bounds
if meth in ["nelder-mead", "powell", "cg", "bfgs", "newton-cg", "dogleg", "trust-ncg"] and (
bounds is not None or any(constraints)
):
warn("Method %s cannot handle constraints nor bounds." % method, RuntimeWarning)
if meth in ["l-bfgs-b", "tnc"] and any(constraints):
warn("Method %s cannot handle constraints." % method, RuntimeWarning)
if meth == "cobyla" and bounds is not None:
warn("Method %s cannot handle bounds." % method, RuntimeWarning)
# - callback
if meth in ["anneal", "cobyla", "slsqp"] and callback is not None:
warn("Method %s does not support callback." % method, RuntimeWarning)
# - return_all
if meth in ["anneal", "l-bfgs-b", "tnc", "cobyla", "slsqp"] and options.get("return_all", False):
warn("Method %s does not support the return_all option." % method, RuntimeWarning)
# fun also returns the jacobian
if not callable(jac):
if bool(jac):
fun = MemoizeJac(fun)
jac = fun.derivative
else:
jac = None
# set default tolerances
if tol is not None:
options = dict(options)
if meth in ["nelder-mead", "newton-cg", "powell", "tnc"]:
options.setdefault("xtol", tol)
if meth in ["nelder-mead", "powell", "anneal", "l-bfgs-b", "tnc", "slsqp"]:
options.setdefault("ftol", tol)
if meth in ["bfgs", "cg", "l-bfgs-b", "tnc", "dogleg", "trust-ncg"]:
options.setdefault("gtol", tol)
if meth in ["cobyla"]:
options.setdefault("tol", tol)
if meth == "nelder-mead":
return _minimize_neldermead(fun, x0, args, callback, **options)
elif meth == "powell":
return _minimize_powell(fun, x0, args, callback, **options)
elif meth == "cg":
return _minimize_cg(fun, x0, args, jac, callback, **options)
elif meth == "bfgs":
return _minimize_bfgs(fun, x0, args, jac, callback, **options)
elif meth == "newton-cg":
return _minimize_newtoncg(fun, x0, args, jac, hess, hessp, callback, **options)
elif meth == "anneal":
return _minimize_anneal(fun, x0, args, **options)
elif meth == "l-bfgs-b":
return _minimize_lbfgsb(fun, x0, args, jac, bounds, callback=callback, **options)
elif meth == "tnc":
return _minimize_tnc(fun, x0, args, jac, bounds, callback=callback, **options)
elif meth == "cobyla":
return _minimize_cobyla(fun, x0, args, constraints, **options)
elif meth == "slsqp":
return _minimize_slsqp(fun, x0, args, jac, bounds, constraints, **options)
elif meth == "dogleg":
return _minimize_dogleg(fun, x0, args, jac, hess, callback=callback, **options)
elif meth == "trust-ncg":
return _minimize_trust_ncg(fun, x0, args, jac, hess, hessp, callback=callback, **options)
else:
raise ValueError("Unknown solver %s" % method)
def minimize(fun, x0, args=(), method=None, jac=None, hess=None,
hessp=None, bounds=None, constraints=(), tol=None,
callback=None, options=None):
"""
Minimization of scalar function of one or more variables.
.. versionadded:: 0.11.0
Parameters
----------
fun : callable
Objective function.
x0 : ndarray
Initial guess.
args : tuple, optional
Extra arguments passed to the objective function and its
derivatives (Jacobian, Hessian).
method : str or callable, optional
Type of solver. Should be one of
- 'Nelder-Mead'
- 'Powell'
- 'CG'
- 'BFGS'
- 'Newton-CG'
- 'Anneal (deprecated as of scipy version 0.14.0)'
- 'L-BFGS-B'
- 'TNC'
- 'COBYLA'
- 'SLSQP'
- 'dogleg'
- 'trust-ncg'
- custom - a callable object (added in version 0.14.0)
If not given, chosen to be one of ``BFGS``, ``L-BFGS-B``, ``SLSQP``,
depending if the problem has constraints or bounds.
jac : bool or callable, optional
Jacobian (gradient) of objective function. Only for CG, BFGS,
Newton-CG, L-BFGS-B, TNC, SLSQP, dogleg, trust-ncg.
If `jac` is a Boolean and is True, `fun` is assumed to return the
gradient along with the objective function. If False, the
gradient will be estimated numerically.
`jac` can also be a callable returning the gradient of the
objective. In this case, it must accept the same arguments as `fun`.
hess, hessp : callable, optional
Hessian (matrix of second-order derivatives) of objective function or
Hessian of objective function times an arbitrary vector p. Only for
Newton-CG, dogleg, trust-ncg.
Only one of `hessp` or `hess` needs to be given. If `hess` is
provided, then `hessp` will be ignored. If neither `hess` nor
`hessp` is provided, then the Hessian product will be approximated
using finite differences on `jac`. `hessp` must compute the Hessian
times an arbitrary vector.
bounds : sequence, optional
Bounds for variables (only for L-BFGS-B, TNC and SLSQP).
``(min, max)`` pairs for each element in ``x``, defining
the bounds on that parameter. Use None for one of ``min`` or
``max`` when there is no bound in that direction.
constraints : dict or sequence of dict, optional
Constraints definition (only for COBYLA and SLSQP).
Each constraint is defined in a dictionary with fields:
type : str
Constraint type: 'eq' for equality, 'ineq' for inequality.
fun : callable
The function defining the constraint.
jac : callable, optional
The Jacobian of `fun` (only for SLSQP).
args : sequence, optional
Extra arguments to be passed to the function and Jacobian.
Equality constraint means that the constraint function result is to
be zero whereas inequality means that it is to be non-negative.
Note that COBYLA only supports inequality constraints.
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
options : dict, optional
A dictionary of solver options. All methods accept the following
generic options:
maxiter : int
Maximum number of iterations to perform.
disp : bool
Set to True to print convergence messages.
For method-specific options, see :func:`show_options()`.
callback : callable, optional
Called after each iteration, as ``callback(xk)``, where ``xk`` is the
current parameter vector.
Returns
-------
res : OptimizeResult
The optimization result represented as a ``OptimizeResult`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the optimizer exited successfully and
``message`` which describes the cause of the termination. See
`OptimizeResult` for a description of other attributes.
See also
--------
minimize_scalar : Interface to minimization algorithms for scalar
univariate functions
show_options : Additional options accepted by the solvers
Notes
-----
This section describes the available solvers that can be selected by the
'method' parameter. The default method is *BFGS*.
**Unconstrained minimization**
Method *Nelder-Mead* uses the Simplex algorithm [1]_, [2]_. This
algorithm has been successful in many applications but other algorithms
using the first and/or second derivatives information might be preferred
for their better performances and robustness in general.
Method *Powell* is a modification of Powell's method [3]_, [4]_ which
is a conjugate direction method. It performs sequential one-dimensional
minimizations along each vector of the directions set (`direc` field in
`options` and `info`), which is updated at each iteration of the main
minimization loop. The function need not be differentiable, and no
derivatives are taken.
Method *CG* uses a nonlinear conjugate gradient algorithm by Polak and
Ribiere, a variant of the Fletcher-Reeves method described in [5]_ pp.
120-122. Only the first derivatives are used.
Method *BFGS* uses the quasi-Newton method of Broyden, Fletcher,
Goldfarb, and Shanno (BFGS) [5]_ pp. 136. It uses the first derivatives
only. BFGS has proven good performance even for non-smooth
optimizations. This method also returns an approximation of the Hessian
inverse, stored as `hess_inv` in the OptimizeResult object.
Method *Newton-CG* uses a Newton-CG algorithm [5]_ pp. 168 (also known
as the truncated Newton method). It uses a CG method to the compute the
search direction. See also *TNC* method for a box-constrained
minimization with a similar algorithm.
Method *Anneal* uses simulated annealing, which is a probabilistic
metaheuristic algorithm for global optimization. It uses no derivative
information from the function being optimized.
Method *dogleg* uses the dog-leg trust-region algorithm [5]_
for unconstrained minimization. This algorithm requires the gradient
and Hessian; furthermore the Hessian is required to be positive definite.
Method *trust-ncg* uses the Newton conjugate gradient trust-region
algorithm [5]_ for unconstrained minimization. This algorithm requires
the gradient and either the Hessian or a function that computes the
product of the Hessian with a given vector.
**Constrained minimization**
Method *L-BFGS-B* uses the L-BFGS-B algorithm [6]_, [7]_ for bound
constrained minimization.
Method *TNC* uses a truncated Newton algorithm [5]_, [8]_ to minimize a
function with variables subject to bounds. This algorithm uses
gradient information; it is also called Newton Conjugate-Gradient. It
differs from the *Newton-CG* method described above as it wraps a C
implementation and allows each variable to be given upper and lower
bounds.
Method *COBYLA* uses the Constrained Optimization BY Linear
Approximation (COBYLA) method [9]_, [10]_, [11]_. The algorithm is
based on linear approximations to the objective function and each
constraint. The method wraps a FORTRAN implementation of the algorithm.
Method *SLSQP* uses Sequential Least SQuares Programming to minimize a
function of several variables with any combination of bounds, equality
and inequality constraints. The method wraps the SLSQP Optimization
subroutine originally implemented by Dieter Kraft [12]_. Note that the
wrapper handles infinite values in bounds by converting them into large
floating values.
**Custom minimizers**
It may be useful to pass a custom minimization method, for example
when using a frontend to this method such as `scipy.optimize.basinhopping`
or a different library. You can simply pass a callable as the ``method``
parameter.
The callable is called as ``method(fun, x0, args, **kwargs, **options)``
where ``kwargs`` corresponds to any other parameters passed to `minimize`
(such as `callback`, `hess`, etc.), except the `options` dict, which has
its contents also passed as `method` parameters pair by pair. Also, if
`jac` has been passed as a bool type, `jac` and `fun` are mangled so that
`fun` returns just the function values and `jac` is converted to a function
returning the Jacobian. The method shall return an ``OptimizeResult``
object.
The provided `method` callable must be able to accept (and possibly ignore)
arbitrary parameters; the set of parameters accepted by `minimize` may
expand in future versions and then these parameters will be passed to
the method. You can find an example in the scipy.optimize tutorial.
References
----------
.. [1] Nelder, J A, and R Mead. 1965. A Simplex Method for Function
Minimization. The Computer Journal 7: 308-13.
.. [2] Wright M H. 1996. Direct search methods: Once scorned, now
respectable, in Numerical Analysis 1995: Proceedings of the 1995
Dundee Biennial Conference in Numerical Analysis (Eds. D F
Griffiths and G A Watson). Addison Wesley Longman, Harlow, UK.
191-208.
.. [3] Powell, M J D. 1964. An efficient method for finding the minimum of
a function of several variables without calculating derivatives. The
Computer Journal 7: 155-162.
.. [4] Press W, S A Teukolsky, W T Vetterling and B P Flannery.
Numerical Recipes (any edition), Cambridge University Press.
.. [5] Nocedal, J, and S J Wright. 2006. Numerical Optimization.
Springer New York.
.. [6] Byrd, R H and P Lu and J. Nocedal. 1995. A Limited Memory
Algorithm for Bound Constrained Optimization. SIAM Journal on
Scientific and Statistical Computing 16 (5): 1190-1208.
.. [7] Zhu, C and R H Byrd and J Nocedal. 1997. L-BFGS-B: Algorithm
778: L-BFGS-B, FORTRAN routines for large scale bound constrained
optimization. ACM Transactions on Mathematical Software 23 (4):
550-560.
.. [8] Nash, S G. Newton-Type Minimization Via the Lanczos Method.
1984. SIAM Journal of Numerical Analysis 21: 770-778.
.. [9] Powell, M J D. A direct search optimization method that models
the objective and constraint functions by linear interpolation.
1994. Advances in Optimization and Numerical Analysis, eds. S. Gomez
and J-P Hennart, Kluwer Academic (Dordrecht), 51-67.
.. [10] Powell M J D. Direct search algorithms for optimization
calculations. 1998. Acta Numerica 7: 287-336.
.. [11] Powell M J D. A view of algorithms for optimization without
derivatives. 2007.Cambridge University Technical Report DAMTP
2007/NA03
.. [12] Kraft, D. A software package for sequential quadratic
programming. 1988. Tech. Rep. DFVLR-FB 88-28, DLR German Aerospace
Center -- Institute for Flight Mechanics, Koln, Germany.
Examples
--------
Let us consider the problem of minimizing the Rosenbrock function. This
function (and its respective derivatives) is implemented in `rosen`
(resp. `rosen_der`, `rosen_hess`) in the `scipy.optimize`.
>>> from scipy.optimize import minimize, rosen, rosen_der
A simple application of the *Nelder-Mead* method is:
>>> x0 = [1.3, 0.7, 0.8, 1.9, 1.2]
>>> res = minimize(rosen, x0, method='Nelder-Mead')
>>> res.x
[ 1. 1. 1. 1. 1.]
Now using the *BFGS* algorithm, using the first derivative and a few
options:
>>> res = minimize(rosen, x0, method='BFGS', jac=rosen_der,
... options={'gtol': 1e-6, 'disp': True})
Optimization terminated successfully.
Current function value: 0.000000
Iterations: 52
Function evaluations: 64
Gradient evaluations: 64
>>> res.x
[ 1. 1. 1. 1. 1.]
>>> print res.message
Optimization terminated successfully.
>>> res.hess
[[ 0.00749589 0.01255155 0.02396251 0.04750988 0.09495377]
[ 0.01255155 0.02510441 0.04794055 0.09502834 0.18996269]
[ 0.02396251 0.04794055 0.09631614 0.19092151 0.38165151]
[ 0.04750988 0.09502834 0.19092151 0.38341252 0.7664427 ]
[ 0.09495377 0.18996269 0.38165151 0.7664427 1.53713523]]
Next, consider a minimization problem with several constraints (namely
Example 16.4 from [5]_). The objective function is:
>>> fun = lambda x: (x[0] - 1)**2 + (x[1] - 2.5)**2
There are three constraints defined as:
>>> cons = ({'type': 'ineq', 'fun': lambda x: x[0] - 2 * x[1] + 2},
... {'type': 'ineq', 'fun': lambda x: -x[0] - 2 * x[1] + 6},
... {'type': 'ineq', 'fun': lambda x: -x[0] + 2 * x[1] + 2})
And variables must be positive, hence the following bounds:
>>> bnds = ((0, None), (0, None))
The optimization problem is solved using the SLSQP method as:
>>> res = minimize(fun, (2, 0), method='SLSQP', bounds=bnds,
... constraints=cons)
It should converge to the theoretical solution (1.4 ,1.7).
"""
if method is None:
# Select automatically
if constraints:
method = 'SLSQP'
elif bounds is not None:
method = 'L-BFGS-B'
else:
method = 'BFGS'
if callable(method):
meth = "_custom"
else:
meth = method.lower()
# deprecated methods
if meth == 'anneal':
warn('Method %s is deprecated in scipy 0.14.0' % method,
DeprecationWarning)
if options is None:
options = {}
# check if optional parameters are supported by the selected method
# - jac
if meth in ['nelder-mead', 'powell', 'anneal', 'cobyla'] and bool(jac):
warn('Method %s does not use gradient information (jac).' % method,
RuntimeWarning)
# - hess
if meth not in ('newton-cg', 'dogleg', 'trust-ncg', '_custom') and hess is not None:
warn('Method %s does not use Hessian information (hess).' % method,
RuntimeWarning)
# - hessp
if meth not in ('newton-cg', 'dogleg', 'trust-ncg', '_custom') and hessp is not None:
warn('Method %s does not use Hessian-vector product '
'information (hessp).' % method, RuntimeWarning)
# - constraints or bounds
if (meth in ['nelder-mead', 'powell', 'cg', 'bfgs', 'newton-cg', 'dogleg',
'trust-ncg'] and (bounds is not None or any(constraints))):
warn('Method %s cannot handle constraints nor bounds.' % method,
RuntimeWarning)
if meth in ['l-bfgs-b', 'tnc'] and any(constraints):
warn('Method %s cannot handle constraints.' % method,
RuntimeWarning)
if meth == 'cobyla' and bounds is not None:
warn('Method %s cannot handle bounds.' % method,
RuntimeWarning)
# - callback
if (meth in ['anneal', 'cobyla'] and callback is not None):
warn('Method %s does not support callback.' % method, RuntimeWarning)
# - return_all
if (meth in ['anneal', 'l-bfgs-b', 'tnc', 'cobyla', 'slsqp'] and
options.get('return_all', False)):
warn('Method %s does not support the return_all option.' % method,
RuntimeWarning)
# fun also returns the jacobian
if not callable(jac):
if bool(jac):
fun = MemoizeJac(fun)
jac = fun.derivative
else:
jac = None
# set default tolerances
if tol is not None:
options = dict(options)
if meth in ['nelder-mead', 'newton-cg', 'powell', 'tnc']:
options.setdefault('xtol', tol)
if meth in ['nelder-mead', 'powell', 'anneal', 'l-bfgs-b', 'tnc',
'slsqp']:
options.setdefault('ftol', tol)
if meth in ['bfgs', 'cg', 'l-bfgs-b', 'tnc', 'dogleg', 'trust-ncg']:
options.setdefault('gtol', tol)
if meth in ['cobyla', '_custom']:
options.setdefault('tol', tol)
if meth == '_custom':
return method(fun, x0, args=args, jac=jac, hess=hess, hessp=hessp,
bounds=bounds, constraints=constraints,
callback=callback, **options)
elif meth == 'nelder-mead':
return _minimize_neldermead(fun, x0, args, callback, **options)
elif meth == 'powell':
return _minimize_powell(fun, x0, args, callback, **options)
elif meth == 'cg':
return _minimize_cg(fun, x0, args, jac, callback, **options)
elif meth == 'bfgs':
return _minimize_bfgs(fun, x0, args, jac, callback, **options)
elif meth == 'newton-cg':
return _minimize_newtoncg(fun, x0, args, jac, hess, hessp, callback,
**options)
elif meth == 'anneal':
return _minimize_anneal(fun, x0, args, **options)
elif meth == 'l-bfgs-b':
return _minimize_lbfgsb(fun, x0, args, jac, bounds,
callback=callback, **options)
elif meth == 'tnc':
return _minimize_tnc(fun, x0, args, jac, bounds, callback=callback,
**options)
elif meth == 'cobyla':
return _minimize_cobyla(fun, x0, args, constraints, **options)
elif meth == 'slsqp':
return _minimize_slsqp(fun, x0, args, jac, bounds,
constraints, callback=callback, **options)
elif meth == 'dogleg':
return _minimize_dogleg(fun, x0, args, jac, hess,
callback=callback, **options)
elif meth == 'trust-ncg':
return _minimize_trust_ncg(fun, x0, args, jac, hess, hessp,
callback=callback, **options)
else:
raise ValueError('Unknown solver %s' % method)
def root(fun, x0, args=(), method="hybr", jac=None, tol=None, callback=None, options=None):
"""
Find a root of a vector function.
Parameters
----------
fun : callable
A vector function to find a root of.
x0 : ndarray
Initial guess.
args : tuple, optional
Extra arguments passed to the objective function and its Jacobian.
method : str, optional
Type of solver. Should be one of
- 'hybr'
- 'lm'
- 'broyden1'
- 'broyden2'
- 'anderson'
- 'linearmixing'
- 'diagbroyden'
- 'excitingmixing'
- 'krylov'
jac : bool or callable, optional
If `jac` is a Boolean and is True, `fun` is assumed to return the
value of Jacobian along with the objective function. If False, the
Jacobian will be estimated numerically.
`jac` can also be a callable returning the Jacobian of `fun`. In
this case, it must accept the same arguments as `fun`.
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
callback : function, optional
Optional callback function. It is called on every iteration as
``callback(x, f)`` where `x` is the current solution and `f`
the corresponding residual. For all methods but 'hybr' and 'lm'.
options : dict, optional
A dictionary of solver options. E.g. `xtol` or `maxiter`, see
:obj:`show_options()` for details.
Returns
-------
sol : OptimizeResult
The solution represented as a ``OptimizeResult`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the algorithm exited successfully and
``message`` which describes the cause of the termination. See
`OptimizeResult` for a description of other attributes.
See also
--------
show_options : Additional options accepted by the solvers
Notes
-----
This section describes the available solvers that can be selected by the
'method' parameter. The default method is *hybr*.
Method *hybr* uses a modification of the Powell hybrid method as
implemented in MINPACK [1]_.
Method *lm* solves the system of nonlinear equations in a least squares
sense using a modification of the Levenberg-Marquardt algorithm as
implemented in MINPACK [1]_.
Methods *broyden1*, *broyden2*, *anderson*, *linearmixing*,
*diagbroyden*, *excitingmixing*, *krylov* are inexact Newton methods,
with backtracking or full line searches [2]_. Each method corresponds
to a particular Jacobian approximations. See `nonlin` for details.
- Method *broyden1* uses Broyden's first Jacobian approximation, it is
known as Broyden's good method.
- Method *broyden2* uses Broyden's second Jacobian approximation, it
is known as Broyden's bad method.
- Method *anderson* uses (extended) Anderson mixing.
- Method *Krylov* uses Krylov approximation for inverse Jacobian. It
is suitable for large-scale problem.
- Method *diagbroyden* uses diagonal Broyden Jacobian approximation.
- Method *linearmixing* uses a scalar Jacobian approximation.
- Method *excitingmixing* uses a tuned diagonal Jacobian
approximation.
.. warning::
The algorithms implemented for methods *diagbroyden*,
*linearmixing* and *excitingmixing* may be useful for specific
problems, but whether they will work may depend strongly on the
problem.
.. versionadded:: 0.11.0
References
----------
.. [1] More, Jorge J., Burton S. Garbow, and Kenneth E. Hillstrom.
1980. User Guide for MINPACK-1.
.. [2] C. T. Kelley. 1995. Iterative Methods for Linear and Nonlinear
Equations. Society for Industrial and Applied Mathematics.
<http://www.siam.org/books/kelley/>
Examples
--------
The following functions define a system of nonlinear equations and its
jacobian.
>>> def fun(x):
... return [x[0] + 0.5 * (x[0] - x[1])**3 - 1.0,
... 0.5 * (x[1] - x[0])**3 + x[1]]
>>> def jac(x):
... return np.array([[1 + 1.5 * (x[0] - x[1])**2,
... -1.5 * (x[0] - x[1])**2],
... [-1.5 * (x[1] - x[0])**2,
... 1 + 1.5 * (x[1] - x[0])**2]])
A solution can be obtained as follows.
>>> from scipy import optimize
>>> sol = optimize.root(fun, [0, 0], jac=jac, method='hybr')
>>> sol.x
array([ 0.8411639, 0.1588361])
"""
if not isinstance(args, tuple):
args = (args,)
meth = method.lower()
if options is None:
options = {}
if callback is not None and meth in ("hybr", "lm"):
warn("Method %s does not accept callback." % method, RuntimeWarning)
# fun also returns the jacobian
if not callable(jac) and meth in ("hybr", "lm"):
if bool(jac):
fun = MemoizeJac(fun)
jac = fun.derivative
else:
jac = None
# set default tolerances
if tol is not None:
options = dict(options)
if meth in ("hybr", "lm"):
options.setdefault("xtol", tol)
elif meth in ("broyden1", "broyden2", "anderson", "linearmixing", "diagbroyden", "excitingmixing", "krylov"):
options.setdefault("xtol", tol)
options.setdefault("xatol", np.inf)
options.setdefault("ftol", np.inf)
options.setdefault("fatol", np.inf)
if meth == "hybr":
sol = _root_hybr(fun, x0, args=args, jac=jac, **options)
elif meth == "lm":
sol = _root_leastsq(fun, x0, args=args, jac=jac, **options)
elif meth in ("broyden1", "broyden2", "anderson", "linearmixing", "diagbroyden", "excitingmixing", "krylov"):
if jac is not None:
warn("Method %s does not use the jacobian (jac)." % method, RuntimeWarning)
sol = _root_nonlin_solve(fun, x0, args=args, jac=jac, _method=meth, _callback=callback, **options)
else:
raise ValueError("Unknown solver %s" % method)
return sol
def resample(x, num, t=None, axis=0, window=None):
"""
Resample `x` to `num` samples using Fourier method along the given axis.
The resampled signal starts at the same value as `x` but is sampled
with a spacing of ``len(x) / num * (spacing of x)``. Because a
Fourier method is used, the signal is assumed to be periodic.
Parameters
----------
x : array_like
The data to be resampled.
num : int
The number of samples in the resampled signal.
t : array_like, optional
If `t` is given, it is assumed to be the sample positions
associated with the signal data in `x`.
axis : int, optional
The axis of `x` that is resampled. Default is 0.
window : array_like, callable, string, float, or tuple, optional
Specifies the window applied to the signal in the Fourier
domain. See below for details.
Returns
-------
resampled_x or (resampled_x, resampled_t)
Either the resampled array, or, if `t` was given, a tuple
containing the resampled array and the corresponding resampled
positions.
Notes
-----
The argument `window` controls a Fourier-domain window that tapers
the Fourier spectrum before zero-padding to alleviate ringing in
the resampled values for sampled signals you didn't intend to be
interpreted as band-limited.
If `window` is a function, then it is called with a vector of inputs
indicating the frequency bins (i.e. fftfreq(x.shape[axis]) ).
If `window` is an array of the same length as `x.shape[axis]` it is
assumed to be the window to be applied directly in the Fourier
domain (with dc and low-frequency first).
For any other type of `window`, the function `scipy.signal.get_window`
is called to generate the window.
The first sample of the returned vector is the same as the first
sample of the input vector. The spacing between samples is changed
from dx to:
dx * len(x) / num
If `t` is not None, then it represents the old sample positions,
and the new sample positions will be returned as well as the new
samples.
"""
x = asarray(x)
X = fft(x, axis=axis)
Nx = x.shape[axis]
if window is not None:
if callable(window):
W = window(fftfreq(Nx))
elif isinstance(window, ndarray) and window.shape == (Nx,):
W = window
else:
W = ifftshift(get_window(window, Nx))
newshape = ones(len(x.shape))
newshape[axis] = len(W)
W.shape = newshape
X = X * W
sl = [slice(None)] * len(x.shape)
newshape = list(x.shape)
newshape[axis] = num
N = int(np.minimum(num, Nx))
Y = zeros(newshape, 'D')
sl[axis] = slice(0, (N + 1) / 2)
Y[sl] = X[sl]
sl[axis] = slice(-(N - 1) / 2, None)
Y[sl] = X[sl]
y = ifft(Y, axis=axis) * (float(num) / float(Nx))
if x.dtype.char not in ['F', 'D']:
y = y.real
if t is None:
return y
else:
new_t = arange(0, num) * (t[1] - t[0]) * Nx / float(num) + t[0]
return y, new_t
def minimize(fun, x0, args=(), method='BFGS', jac=None, hess=None,
hessp=None, bounds=None, constraints=(), tol=None,
callback=None, options=None):
"""
Minimization of scalar function of one or more variables.
Parameters
----------
fun : callable
Objective function.
x0 : ndarray
Initial guess.
args : tuple, optional
Extra arguments passed to the objective function and its
derivatives (Jacobian, Hessian).
method : str, optional
- 'BFGS'
jac : bool or callable, optional
Jacobian of objective function. Only for CG, BFGS, Newton-CG.
If `jac` is a Boolean and is True, `fun` is assumed to return the
value of Jacobian along with the objective function. If False, the
Jacobian will be estimated numerically.
`jac` can also be a callable returning the Jacobian of the
objective. In this case, it must accept the same arguments as `fun`.
hess, hessp : callable, optional
Hessian of objective function or Hessian of objective function
times an arbitrary vector p. Only for Newton-CG.
Only one of `hessp` or `hess` needs to be given. If `hess` is
provided, then `hessp` will be ignored. If neither `hess` nor
`hessp` is provided, then the hessian product will be approximated
using finite differences on `jac`. `hessp` must compute the Hessian
times an arbitrary vector.
bounds : sequence, optional
Bounds for variables (only for L-BFGS-B, TNC, COBYLA and SLSQP).
``(min, max)`` pairs for each element in ``x``, defining
the bounds on that parameter. Use None for one of ``min`` or
``max`` when there is no bound in that direction.
constraints : dict or sequence of dict, optional
Constraints definition (only for COBYLA and SLSQP).
Each constraint is defined in a dictionary with fields:
type : str
Constraint type: 'eq' for equality, 'ineq' for inequality.
fun : callable
The function defining the constraint.
jac : callable, optional
The Jacobian of `fun` (only for SLSQP).
args : sequence, optional
Extra arguments to be passed to the function and Jacobian.
Equality constraint means that the constraint function result is to
be zero whereas inequality means that it is to be non-negative.
Note that COBYLA only supports inequality constraints.
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
options : dict, optional
A dictionary of solver options. All methods accept the following
generic options:
maxiter : int
Maximum number of iterations to perform.
disp : bool
Set to True to print convergence messages.
For method-specific options, see `show_options('minimize', method)`.
callback : callable, optional
Called after each iteration, as ``callback(xk)``, where ``xk`` is the
current parameter vector.
Returns
-------
res : Result
The optimization result represented as a ``Result`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the optimizer exited successfully and
``message`` which describes the cause of the termination. See
`Result` for a description of other attributes.
Method *BFGS* uses the quasi-Newton method of Broyden, Fletcher,
Goldfarb, and Shanno (BFGS) [5]_ pp. 136. It uses the first derivatives
only. BFGS has proven good performance even for non-smooth
optimizations
"""
meth = method.lower()
if options is None:
options = {}
# fun also returns the jacobian
if not callable(jac):
if bool(jac):
fun = MemoizeJac(fun)
jac = fun.derivative
else:
jac = None
# set default tolerances
if tol is not None:
options = dict(options)
if meth in ['bfgs']:
options.setdefault('gtol', tol)
#call _minimize_bfgs in optimize_joe
if meth == 'bfgs':
return _minimize_bfgs(fun, x0, args, jac, callback, **options)
else:
raise ValueError('Unknown solver %s' % method)
def minimize_scalar(fun, bracket=None, bounds=None, args=(),
method='brent', tol=None, options=None):
"""
Minimization of scalar function of one variable.
.. versionadded:: 0.11.0
Parameters
----------
fun : callable
Objective function.
Scalar function, must return a scalar.
bracket : sequence, optional
For methods 'brent' and 'golden', `bracket` defines the bracketing
interval and can either have three items `(a, b, c)` so that `a < b
< c` and `fun(b) < fun(a), fun(c)` or two items `a` and `c` which
are assumed to be a starting interval for a downhill bracket search
(see `bracket`); it doesn't always mean that the obtained solution
will satisfy `a <= x <= c`.
bounds : sequence, optional
For method 'bounded', `bounds` is mandatory and must have two items
corresponding to the optimization bounds.
args : tuple, optional
Extra arguments passed to the objective function.
method : str or callable, optional
Type of solver. Should be one of
- 'Brent'
- 'Bounded'
- 'Golden'
- custom - a callable object (added in version 0.14.0)
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
options : dict, optional
A dictionary of solver options.
maxiter : int
Maximum number of iterations to perform.
disp : bool
Set to True to print convergence messages.
See :func:`show_options()` for solver-specific options.
Returns
-------
res : OptimizeResult
The optimization result represented as a ``OptimizeResult`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the optimizer exited successfully and
``message`` which describes the cause of the termination. See
`OptimizeResult` for a description of other attributes.
See also
--------
minimize : Interface to minimization algorithms for scalar multivariate
functions
show_options : Additional options accepted by the solvers
Notes
-----
This section describes the available solvers that can be selected by the
'method' parameter. The default method is *Brent*.
Method *Brent* uses Brent's algorithm to find a local minimum.
The algorithm uses inverse parabolic interpolation when possible to
speed up convergence of the golden section method.
Method *Golden* uses the golden section search technique. It uses
analog of the bisection method to decrease the bracketed interval. It
is usually preferable to use the *Brent* method.
Method *Bounded* can perform bounded minimization. It uses the Brent
method to find a local minimum in the interval x1 < xopt < x2.
**Custom minimizers**
It may be useful to pass a custom minimization method, for example
when using some library frontend to minimize_scalar. You can simply
pass a callable as the ``method`` parameter.
The callable is called as ``method(fun, args, **kwargs, **options)``
where ``kwargs`` corresponds to any other parameters passed to `minimize`
(such as `bracket`, `tol`, etc.), except the `options` dict, which has
its contents also passed as `method` parameters pair by pair. The method
shall return an ``OptimizeResult`` object.
The provided `method` callable must be able to accept (and possibly ignore)
arbitrary parameters; the set of parameters accepted by `minimize` may
expand in future versions and then these parameters will be passed to
the method. You can find an example in the scipy.optimize tutorial.
Examples
--------
Consider the problem of minimizing the following function.
>>> def f(x):
... return (x - 2) * x * (x + 2)**2
Using the *Brent* method, we find the local minimum as:
>>> from scipy.optimize import minimize_scalar
>>> res = minimize_scalar(f)
>>> res.x
1.28077640403
Using the *Bounded* method, we find a local minimum with specified
bounds as:
>>> res = minimize_scalar(f, bounds=(-3, -1), method='bounded')
>>> res.x
-2.0000002026
"""
if callable(method):
meth = "_custom"
else:
meth = method.lower()
if options is None:
options = {}
if tol is not None:
options = dict(options)
if meth == 'bounded' and 'xatol' not in options:
warn("Method 'bounded' does not support relative tolerance in x; "
"defaulting to absolute tolerance.", RuntimeWarning)
options['xatol'] = tol
elif meth == '_custom':
options.setdefault('tol', tol)
else:
options.setdefault('xtol', tol)
if meth == '_custom':
return method(fun, args=args, bracket=bracket, bounds=bounds, **options)
elif meth == 'brent':
return _minimize_scalar_brent(fun, bracket, args, **options)
elif meth == 'bounded':
if bounds is None:
raise ValueError('The `bounds` parameter is mandatory for '
'method `bounded`.')
return _minimize_scalar_bounded(fun, bounds, args, **options)
elif meth == 'golden':
return _minimize_scalar_golden(fun, bracket, args, **options)
else:
raise ValueError('Unknown solver %s' % method)
def fmin_cobyla(func, x0, cons, args=(), consargs=None, rhobeg=1.0,
rhoend=1e-4, iprint=1, maxfun=1000, disp=None, catol=1e-6):
"""
Minimize a function using the Constrained Optimization BY Linear
Approximation (COBYLA) method. This method wraps a FORTRAN
implentation of the algorithm.
Parameters
----------
func : callable
Function to minimize. In the form func(x, \\*args).
x0 : ndarray
Initial guess.
cons : sequence
Constraint functions; must all be ``>=0`` (a single function
if only 1 constraint). Each function takes the parameters `x`
as its first argument.
args : tuple
Extra arguments to pass to function.
consargs : tuple
Extra arguments to pass to constraint functions (default of None means
use same extra arguments as those passed to func).
Use ``()`` for no extra arguments.
rhobeg :
Reasonable initial changes to the variables.
rhoend :
Final accuracy in the optimization (not precisely guaranteed). This
is a lower bound on the size of the trust region.
iprint : {0, 1, 2, 3}
Controls the frequency of output; 0 implies no output. Deprecated.
disp : {0, 1, 2, 3}
Over-rides the iprint interface. Preferred.
maxfun : int
Maximum number of function evaluations.
catol : float
Absolute tolerance for constraint violations.
Returns
-------
x : ndarray
The argument that minimises `f`.
See also
--------
minimize: Interface to minimization algorithms for multivariate
functions. See the 'COBYLA' `method` in particular.
Notes
-----
This algorithm is based on linear approximations to the objective
function and each constraint. We briefly describe the algorithm.
Suppose the function is being minimized over k variables. At the
jth iteration the algorithm has k+1 points v_1, ..., v_(k+1),
an approximate solution x_j, and a radius RHO_j.
(i.e. linear plus a constant) approximations to the objective
function and constraint functions such that their function values
agree with the linear approximation on the k+1 points v_1,.., v_(k+1).
This gives a linear program to solve (where the linear approximations
of the constraint functions are constrained to be non-negative).
However the linear approximations are likely only good
approximations near the current simplex, so the linear program is
given the further requirement that the solution, which
will become x_(j+1), must be within RHO_j from x_j. RHO_j only
decreases, never increases. The initial RHO_j is rhobeg and the
final RHO_j is rhoend. In this way COBYLA's iterations behave
like a trust region algorithm.
Additionally, the linear program may be inconsistent, or the
approximation may give poor improvement. For details about
how these issues are resolved, as well as how the points v_i are
updated, refer to the source code or the references below.
References
----------
Powell M.J.D. (1994), "A direct search optimization method that models
the objective and constraint functions by linear interpolation.", in
Advances in Optimization and Numerical Analysis, eds. S. Gomez and
J-P Hennart, Kluwer Academic (Dordrecht), pp. 51-67
Powell M.J.D. (1998), "Direct search algorithms for optimization
calculations", Acta Numerica 7, 287-336
Powell M.J.D. (2007), "A view of algorithms for optimization without
derivatives", Cambridge University Technical Report DAMTP 2007/NA03
Examples
--------
Minimize the objective function f(x,y) = x*y subject
to the constraints x**2 + y**2 < 1 and y > 0::
>>> def objective(x):
... return x[0]*x[1]
...
>>> def constr1(x):
... return 1 - (x[0]**2 + x[1]**2)
...
>>> def constr2(x):
... return x[1]
...
>>> fmin_cobyla(objective, [0.0, 0.1], [constr1, constr2], rhoend=1e-7)
Normal return from subroutine COBYLA
NFVALS = 64 F =-5.000000E-01 MAXCV = 1.998401E-14
X =-7.071069E-01 7.071067E-01
array([-0.70710685, 0.70710671])
The exact solution is (-sqrt(2)/2, sqrt(2)/2).
"""
err = "cons must be a sequence of callable functions or a single"\
" callable function."
try:
len(cons)
except TypeError:
if callable(cons):
cons = [cons]
else:
raise TypeError(err)
else:
for thisfunc in cons:
if not callable(thisfunc):
raise TypeError(err)
if consargs is None:
consargs = args
# build constraints
con = tuple({'type': 'ineq', 'fun': c, 'args': consargs} for c in cons)
# options
if disp is not None:
iprint = disp
opts = {'rhobeg': rhobeg,
'tol': rhoend,
'iprint': iprint,
'disp': iprint != 0,
'maxiter': maxfun,
'catol': catol}
sol = _minimize_cobyla(func, x0, args, constraints=con,
**opts)
if iprint > 0 and not sol['success']:
print("COBYLA failed to find a solution: %s" % (sol.message,))
return sol['x']
def schur(a, output='real', lwork=None, overwrite_a=False, sort=None,
check_finite=True):
"""Compute Schur decomposition of a matrix.
The Schur decomposition is::
A = Z T Z^H
where Z is unitary and T is either upper-triangular, or for real
Schur decomposition (output='real'), quasi-upper triangular. In
the quasi-triangular form, 2x2 blocks describing complex-valued
eigenvalue pairs may extrude from the diagonal.
Parameters
----------
a : ndarray, shape (M, M)
Matrix to decompose
output : {'real', 'complex'}, optional
Construct the real or complex Schur decomposition (for real matrices).
lwork : int, optional
Work array size. If None or -1, it is automatically computed.
overwrite_a : bool, optional
Whether to overwrite data in a (may improve performance).
sort : {None, callable, 'lhp', 'rhp', 'iuc', 'ouc'}, optional
Specifies whether the upper eigenvalues should be sorted. A callable
may be passed that, given a eigenvalue, returns a boolean denoting
whether the eigenvalue should be sorted to the top-left (True).
Alternatively, string parameters may be used::
'lhp' Left-hand plane (x.real < 0.0)
'rhp' Right-hand plane (x.real > 0.0)
'iuc' Inside the unit circle (x*x.conjugate() <= 1.0)
'ouc' Outside the unit circle (x*x.conjugate() > 1.0)
Defaults to None (no sorting).
check_finite : boolean, optional
Whether to check the input matrixes contain only finite numbers.
Disabling may give a performance gain, but may result to problems
(crashes, non-termination) if the inputs do contain infinities or NaNs.
Returns
-------
T : ndarray, shape (M, M)
Schur form of A. It is real-valued for the real Schur decomposition.
Z : ndarray, shape (M, M)
An unitary Schur transformation matrix for A.
It is real-valued for the real Schur decomposition.
sdim : int
If and only if sorting was requested, a third return value will
contain the number of eigenvalues satisfying the sort condition.
Raises
------
LinAlgError
Error raised under three conditions:
1. The algorithm failed due to a failure of the QR algorithm to
compute all eigenvalues
2. If eigenvalue sorting was requested, the eigenvalues could not be
reordered due to a failure to separate eigenvalues, usually because
of poor conditioning
3. If eigenvalue sorting was requested, roundoff errors caused the
leading eigenvalues to no longer satisfy the sorting condition
See also
--------
rsf2csf : Convert real Schur form to complex Schur form
"""
if not output in ['real','complex','r','c']:
raise ValueError("argument must be 'real', or 'complex'")
if check_finite:
a1 = asarray_chkfinite(a)
else:
a1 = asarray(a)
if len(a1.shape) != 2 or (a1.shape[0] != a1.shape[1]):
raise ValueError('expected square matrix')
typ = a1.dtype.char
if output in ['complex','c'] and typ not in ['F','D']:
if typ in _double_precision:
a1 = a1.astype('D')
typ = 'D'
else:
a1 = a1.astype('F')
typ = 'F'
overwrite_a = overwrite_a or (_datacopied(a1, a))
gees, = get_lapack_funcs(('gees',), (a1,))
if lwork is None or lwork == -1:
# get optimal work array
result = gees(lambda x: None, a1, lwork=-1)
lwork = result[-2][0].real.astype(numpy.int)
if sort is None:
sort_t = 0
sfunction = lambda x: None
else:
sort_t = 1
if callable(sort):
sfunction = sort
elif sort == 'lhp':
sfunction = lambda x: (numpy.real(x) < 0.0)
elif sort == 'rhp':
sfunction = lambda x: (numpy.real(x) >= 0.0)
elif sort == 'iuc':
sfunction = lambda x: (abs(x) <= 1.0)
elif sort == 'ouc':
sfunction = lambda x: (abs(x) > 1.0)
else:
raise ValueError("sort parameter must be None, a callable, or " +
"one of ('lhp','rhp','iuc','ouc')")
result = gees(sfunction, a1, lwork=lwork, overwrite_a=overwrite_a,
sort_t=sort_t)
info = result[-1]
if info < 0:
raise ValueError('illegal value in %d-th argument of internal gees'
% -info)
elif info == a1.shape[0] + 1:
raise LinAlgError('Eigenvalues could not be separated for reordering.')
elif info == a1.shape[0] + 2:
raise LinAlgError('Leading eigenvalues do not satisfy sort condition.')
elif info > 0:
raise LinAlgError("Schur form not found. Possibly ill-conditioned.")
if sort_t == 0:
return result[0], result[-3]
else:
return result[0], result[-3], result[1]
def root(fun,
x0,
args=(),
method='hybr',
jac=None,
tol=None,
callback=None,
options=None):
"""
Find a root of a vector function.
Parameters
----------
fun : callable
A vector function to find a root of.
x0 : ndarray
Initial guess.
args : tuple, optional
Extra arguments passed to the objective function and its Jacobian.
method : str, optional
Type of solver. Should be one of
- 'hybr'
- 'lm'
- 'broyden1'
- 'broyden2'
- 'anderson'
- 'linearmixing'
- 'diagbroyden'
- 'excitingmixing'
- 'krylov'
jac : bool or callable, optional
If `jac` is a Boolean and is True, `fun` is assumed to return the
value of Jacobian along with the objective function. If False, the
Jacobian will be estimated numerically.
`jac` can also be a callable returning the Jacobian of `fun`. In
this case, it must accept the same arguments as `fun`.
tol : float, optional
Tolerance for termination. For detailed control, use solver-specific
options.
callback : function, optional
Optional callback function. It is called on every iteration as
``callback(x, f)`` where `x` is the current solution and `f`
the corresponding residual. For all methods but 'hybr' and 'lm'.
options : dict, optional
A dictionary of solver options. E.g. `xtol` or `maxiter`, see
:obj:`show_options()` for details.
Returns
-------
sol : OptimizeResult
The solution represented as a ``OptimizeResult`` object.
Important attributes are: ``x`` the solution array, ``success`` a
Boolean flag indicating if the algorithm exited successfully and
``message`` which describes the cause of the termination. See
`OptimizeResult` for a description of other attributes.
See also
--------
show_options : Additional options accepted by the solvers
Notes
-----
This section describes the available solvers that can be selected by the
'method' parameter. The default method is *hybr*.
Method *hybr* uses a modification of the Powell hybrid method as
implemented in MINPACK [1]_.
Method *lm* solves the system of nonlinear equations in a least squares
sense using a modification of the Levenberg-Marquardt algorithm as
implemented in MINPACK [1]_.
Methods *broyden1*, *broyden2*, *anderson*, *linearmixing*,
*diagbroyden*, *excitingmixing*, *krylov* are inexact Newton methods,
with backtracking or full line searches [2]_. Each method corresponds
to a particular Jacobian approximations. See `nonlin` for details.
- Method *broyden1* uses Broyden's first Jacobian approximation, it is
known as Broyden's good method.
- Method *broyden2* uses Broyden's second Jacobian approximation, it
is known as Broyden's bad method.
- Method *anderson* uses (extended) Anderson mixing.
- Method *Krylov* uses Krylov approximation for inverse Jacobian. It
is suitable for large-scale problem.
- Method *diagbroyden* uses diagonal Broyden Jacobian approximation.
- Method *linearmixing* uses a scalar Jacobian approximation.
- Method *excitingmixing* uses a tuned diagonal Jacobian
approximation.
.. warning::
The algorithms implemented for methods *diagbroyden*,
*linearmixing* and *excitingmixing* may be useful for specific
problems, but whether they will work may depend strongly on the
problem.
.. versionadded:: 0.11.0
References
----------
.. [1] More, Jorge J., Burton S. Garbow, and Kenneth E. Hillstrom.
1980. User Guide for MINPACK-1.
.. [2] C. T. Kelley. 1995. Iterative Methods for Linear and Nonlinear
Equations. Society for Industrial and Applied Mathematics.
<http://www.siam.org/books/kelley/>
Examples
--------
The following functions define a system of nonlinear equations and its
jacobian.
>>> def fun(x):
... return [x[0] + 0.5 * (x[0] - x[1])**3 - 1.0,
... 0.5 * (x[1] - x[0])**3 + x[1]]
>>> def jac(x):
... return np.array([[1 + 1.5 * (x[0] - x[1])**2,
... -1.5 * (x[0] - x[1])**2],
... [-1.5 * (x[1] - x[0])**2,
... 1 + 1.5 * (x[1] - x[0])**2]])
A solution can be obtained as follows.
>>> from scipy import optimize
>>> sol = optimize.root(fun, [0, 0], jac=jac, method='hybr')
>>> sol.x
array([ 0.8411639, 0.1588361])
"""
if not isinstance(args, tuple):
args = (args, )
meth = method.lower()
if options is None:
options = {}
if callback is not None and meth in ('hybr', 'lm'):
warn('Method %s does not accept callback.' % method, RuntimeWarning)
# fun also returns the jacobian
if not callable(jac) and meth in ('hybr', 'lm'):
if bool(jac):
fun = MemoizeJac(fun)
jac = fun.derivative
else:
jac = None
# set default tolerances
if tol is not None:
options = dict(options)
if meth in ('hybr', 'lm'):
options.setdefault('xtol', tol)
elif meth in ('broyden1', 'broyden2', 'anderson', 'linearmixing',
'diagbroyden', 'excitingmixing', 'krylov'):
options.setdefault('xtol', tol)
options.setdefault('xatol', np.inf)
options.setdefault('ftol', np.inf)
options.setdefault('fatol', np.inf)
if meth == 'hybr':
sol = _root_hybr(fun, x0, args=args, jac=jac, **options)
elif meth == 'lm':
sol = _root_leastsq(fun, x0, args=args, jac=jac, **options)
elif meth in ('broyden1', 'broyden2', 'anderson', 'linearmixing',
'diagbroyden', 'excitingmixing', 'krylov'):
if jac is not None:
warn('Method %s does not use the jacobian (jac).' % method,
RuntimeWarning)
sol = _root_nonlin_solve(fun,
x0,
args=args,
jac=jac,
_method=meth,
_callback=callback,
**options)
else:
raise ValueError('Unknown solver %s' % method)
return sol