def __init__(self, **kwargs):
self._log = logging.getLogger(self.__class__.__name__)
self.evChanged = Subject()
self.evConfigChanged = Subject()
self._displacement = None
self._displacement_px_var = None
self._phi = None
self._theta = None
self._los_factors = None
self.cols = 0
self.rows = 0
self.los = LOSUnitVectors(scene=self)
self._elevation = {}
frame_config = kwargs.pop('frame_config', FrameConfig())
for fattr in ('llLat', 'llLon', 'dLat', 'dLon'):
coord = kwargs.pop(fattr, None)
if coord is not None:
frame_config.__setattr__(fattr, coord)
self.frame = Frame(scene=self, config=frame_config)
for attr in ('displacement', 'displacement_px_var', 'theta', 'phi'):
data = kwargs.pop(attr, None)
if data is not None:
self.__setattr__(attr, data)
def __init__(self, config=SceneConfig(), **kwargs):
self.evChanged = Subject()
self.evConfigChanged = Subject()
self.config = config
self.meta = self.config.meta
BaseScene.__init__(self, frame_config=self.config.frame, **kwargs)
# wiring special methods
self.import_data = self._import_data
self.load = self._load
def __init__(self, scene, config=CovarianceConfig()):
self.evChanged = Subject()
self.evConfigChanged = Subject()
self.frame = scene.frame
self.quadtree = scene.quadtree
self.scene = scene
self.nthreads = 0
self._noise_data = None
self._powerspec1d_cached = None
self._powerspec2d_cached = None
self._powerspec3d_cached = None
self._noise_data_grid = None
self._initialized = False
self._log = scene._log.getChild("Covariance")
self.setConfig(config)
self.quadtree.evChanged.subscribe(self._clear)
self.scene.evConfigChanged.subscribe(self.setConfig)
def __init__(self, config=None, **kwargs):
self.evChanged = Subject()
self.evModelUpdated = Subject()
self.evConfigChanged = Subject()
self.config = config if config else SandboxSceneConfig()
BaseScene.__init__(self, frame_config=self.config.frame, **kwargs)
self.reference = None
self._los_factors = None
for attr in ['theta', 'phi']:
data = kwargs.pop(attr, None)
if data is not None:
self.__setattr__(attr, data)
self.setExtent(self.config.extent_east, self.config.extent_north)
if self.config.reference_scene is not None:
self.loadReferenceScene(self.config.reference_scene)
def __init__(self, scene, **kwargs):
self.evPlotChanged = Subject()
self._scene = scene
self._data = None
self.fig = None
self.ax = None
self._show_plt = False
self._colormap_symmetric = True
self.title = 'unnamed'
self._log = logging.getLogger(self.__class__.__name__)
def __init__(self, scene, config=None):
self.evChanged = Subject()
self._scene = scene
self._log = scene._log.getChild('Frame')
self.N = None
self.E = None
self.llEutm = None
self.llNutm = None
self.utm_zone = None
self.utm_zone_letter = None
self._meter_grid = None
self._updateConfig(config or FrameConfig())
self._scene.evConfigChanged.subscribe(self._updateConfig)
self._scene.evChanged.subscribe(self.updateExtent)
def __init__(self, *args, **kwargs):
Object.__init__(self, *args, **kwargs)
self._cached_result = None
self.evParametersChanged = Subject()
class Covariance(object):
'''Construct the variance-covariance matrix of quadtree subsampled data.
Variance and covariance estimates are used to construct the weighting
matrix to be used later in an optimization.
Two different methods exist to propagate full-resolution data variances
and covariances of :class:`kite.Scene.displacement` to the
covariance matrix of the subsampled dataset:
1. The distance between :py:class:`kite.quadtree.QuadNode`
leaf focal points, :py:class:`kite.covariance.Covariance.matrix_focal`
defines the approximate covariance of the quadtree leaf pair.
2. The _accurate_ propagation of covariances by taking the mean of
every node pair pixel covariances. This process is computational
very expensive and can take a few minutes.
:py:class:`kite.covariance.Covariance.matrix_focal`
:param quadtree: Quadtree to work on
:type quadtree: :class:`~kite.Quadtree`
:param config: Config object
:type config: :class:`~kite.covariance.CovarianceConfig`
'''
evChanged = Subject()
evConfigChanged = Subject()
def __init__(self, scene, config=CovarianceConfig()):
self.frame = scene.frame
self.quadtree = scene.quadtree
self.scene = scene
self._noise_data = None
self._powerspec1d_cached = None
self._powerspec2d_cached = None
self._powerspec3d_cached = None
self._initialized = False
self._nthreads = 0
self._log = scene._log.getChild('Covariance')
self.setConfig(config)
self.quadtree.evChanged.subscribe(self._clear)
self.scene.evConfigChanged.subscribe(self.setConfig)
def __call__(self, *args, **kwargs):
return self.getLeafCovariance(*args, **kwargs)
def setConfig(self, config=None):
''' Sets and updated the config of the instance
:param config: New config instance, defaults to configuration provided
by parent :class:`~kite.Scene`
:type config: :class:`~kite.covariance.CovarianceConfig`, optional
'''
if config is None:
config = self.scene.config.covariance
self.config = config
if config.noise_coord is None\
and (config.a is not None or
config.b is not None or
config.variance is not None):
self.noise_data # init data array
self.config.a = config.a
self.config.b = config.b
self.config.variance = config.variance
self._clear(config=False)
self.evConfigChanged.notify()
def _clear(self, config=True, spectrum=True):
if config:
self.config.a = None
self.config.b = None
self.config.variance = None
self.config.covariance_matrix = None
if spectrum:
self.structure_func = None
self._powerspec1d_cached = None
self._powerspec2d_cached = None
self.covariance_matrix = None
self.covariance_matrix_focal = None
self.covariance_func = None
self.weight_matrix = None
self.weight_matrix_focal = None
self._initialized = False
self.evChanged.notify()
@property
def nthreads(self):
''' Number of threads (CPU cores) to use for full covariance
calculation
Setting ``nthreads`` to ``0`` uses all available cores (default).
:setter: Sets the number of threads
:type: int
'''
return self._nthreads
@nthreads.setter
def nthreads(self, value):
self._nthreads = int(value)
@property
def noise_coord(self):
''' Coordinates of the noise patch in local coordinates.
:setter: Set the noise coordinates
:getter: Get the noise coordinates
:type: :class:`numpy.ndarray`, ``[llE, llN, sizeE, sizeN]``
'''
if self.config.noise_coord is None:
self.noise_data
return self.config.noise_coord
@noise_coord.setter
def noise_coord(self, values):
self.config.noise_coord = num.array(values)
@property
def noise_patch_size_km2(self):
'''
:getter: Noise patch size in :math:`km^2`.
:type: float
'''
if self.noise_coord is None:
return 0.
size = (self.noise_coord[2] * self.noise_coord[3]) * 1e-6
if size < 75:
self._log.warning('Defined noise patch is instably small')
return size
@property
def noise_data(self, data):
''' Noise data we process to estimate the covariance
:setter: Set the noise patch to analyze the covariance.
:getter: If the noise data has not been set manually, we grab data
through :func:`~kite.Covariance.selectNoiseNode`.
:type: :class:`numpy.ndarray`
'''
return self._noise_data
@noise_data.getter
def noise_data(self):
if self._noise_data is not None:
return self._noise_data
elif self.config.noise_coord is not None:
self._log.debug('Selecting noise_data from config...')
llE, llN = self.scene.frame.mapENMatrix(
*self.config.noise_coord[:2])
sE, sN = self.scene.frame.mapENMatrix(*self.config.noise_coord[2:])
slice_E = slice(llE, llE + sE)
slice_N = slice(llN, llN + sN)
self.noise_data = self.scene.displacement[slice_N, slice_E]
else:
self._log.debug('Selecting noise_data from Quadtree...')
node = self.selectNoiseNode()
self.noise_data = node.displacement
self.noise_coord = [node.llE, node.llN, node.sizeE, node.sizeN]
return self.noise_data
@noise_data.setter
def noise_data(self, data):
data = data.copy()
data = derampMatrix(trimMatrix(data))
data = trimMatrix(data)
data[num.isnan(data)] = 0.
self._noise_data = data
self._clear()
def selectNoiseNode(self):
''' Choose noise node from quadtree
the biggest :class:`~kite.quadtree.QuadNode` from
:class:`~kite.Quadtree`.
:returns: A quadnode with the least signal.
:rtype: :class:`~kite.quadtree.QuadNode`
'''
t0 = time.time()
stdmax = max([n.std for n in self.quadtree.nodes]) # noqa
lmax = max([n.std for n in self.quadtree.nodes]) # noqa
def costFunction(n):
nl = num.log2(n.length) / num.log2(lmax)
ns = n.std / stdmax
return nl * (1. - ns) * (1. - n.nan_fraction)
nodes = sorted(self.quadtree.nodes, key=costFunction)
self._log.debug('Fetched noise from Quadtree.nodes [%0.4f s]' %
(time.time() - t0))
return nodes[0]
def _mapLeaves(self, nx, ny):
''' Helper function returning appropriate
:class:`~kite.quadtree.QuadNode` and for maintaining
the internal mapping with the matrices.
:param nx: matrix x position
:type nx: int
:param ny: matrix y position
:type ny: int
:returns: tuple of :class:`~kite.quadtree.QuadNode` s for ``nx``
and ``ny``
:rtype: tuple
'''
leaf1 = self.quadtree.leaves[nx]
leaf2 = self.quadtree.leaves[ny]
self._leaf_mapping[leaf1.id] = nx
self._leaf_mapping[leaf2.id] = ny
return leaf1, leaf2
@property_cached
def covariance_matrix(self):
''' Covariance matrix calculated from mean of all pixel pairs
inside the node pairs (full and accurate propagation).
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves` x
:class:`~kite.Quadtree.nleaves`)
'''
if not isinstance(self.config.covariance_matrix, num.ndarray):
self.config.covariance_matrix =\
self._calcCovarianceMatrix(method='full')
elif self.config.covariance_matrix.ndim == 1:
try:
nl = self.quadtree.nleaves
self.config.covariance_matrix =\
self.config.covariance_matrix.reshape(nl, nl)
except ValueError:
self.config.covariance = None
return self.covariance_matrix
return self.config.covariance_matrix
@property_cached
def covariance_matrix_focal(self):
''' Approximate Covariance matrix from quadtree leaf pair
distance only. Fast, use for intermediate steps only and
finallly use approach :attr:`~kite.Covariance.covariance_matrix`.
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves` x
:class:`~kite.Quadtree.nleaves`)
'''
return self._calcCovarianceMatrix(method='focal')
@property_cached
def weight_matrix(self):
''' Weight matrix from full covariance :math:`\\sqrt{cov^{-1}}`.
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves` x
:class:`~kite.Quadtree.nleaves`)
'''
return num.linalg.inv(self.covariance_matrix)
@property_cached
def weight_matrix_focal(self):
''' Approximated weight matrix from fast focal method
:math:`\\sqrt{cov_{focal}^{-1}}`.
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves` x
:class:`~kite.Quadtree.nleaves`)
'''
return num.linalg.inv(self.covariance_matrix_focal)
@property_cached
def weight_vector(self):
''' Weight vector from full covariance :math:`\\sqrt{cov^{-1}}`.
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves`)
'''
return num.sum(self.weight_matrix, axis=1)
@property_cached
def weight_vector_focal(self):
''' Weight vector from fast focal method
:math:`\\sqrt{cov_{focal}^{-1}}`.
:type: :class:`numpy.ndarray`,
size (:class:`~kite.Quadtree.nleaves`)
'''
return num.sum(self.weight_matrix_focal, axis=1)
def _calcCovarianceMatrix(self, method='focal', nthreads=None):
'''Constructs the covariance matrix.
:param method: Either ``focal`` point distances are used - this is
quick but only an approximation.
Or ``full``, where the full quadtree pixel distances matrices are
calculated , defaults to ``focal``
:type method: str, optional
:returns: Covariance matrix
:rtype: thon:numpy.ndarray
'''
self._initialized = True
if nthreads is None:
nthreads = self.nthreads
nl = len(self.quadtree.leaves)
self._leaf_mapping = {}
t0 = time.time()
ma, mb = self.covariance_model
if method == 'focal':
dist_matrix = num.zeros((nl, nl))
dist_iter = num.nditer(num.triu_indices_from(dist_matrix))
for nx, ny in dist_iter:
leaf1, leaf2 = self._mapLeaves(nx, ny)
dist = self._leafFocalDistance(leaf1, leaf2)
dist_matrix[(nx, ny), (ny, nx)] = dist
cov_matrix = modelCovariance(dist_matrix, ma, mb)
num.fill_diagonal(cov_matrix, self.variance)
elif method == 'full':
leaf_map = num.empty((len(self.quadtree.leaves), 4),
dtype=num.uint32)
for nl, leaf in enumerate(self.quadtree.leaves):
leaf, _ = self._mapLeaves(nl, nl)
leaf_map[nl, 0], leaf_map[nl, 1] = (leaf._slice_rows.start,
leaf._slice_rows.stop)
leaf_map[nl, 2], leaf_map[nl, 3] = (leaf._slice_cols.start,
leaf._slice_cols.stop)
nleaves = self.quadtree.nleaves
cov_matrix = covariance_ext.covariance_matrix(
self.scene.frame.gridE.filled(),
self.scene.frame.gridN.filled(),
leaf_map, ma, mb, self.variance, nthreads,
self.config.adaptive_subsampling)\
.reshape(nleaves, nleaves)
else:
raise TypeError('Covariance calculation %s method not defined!' %
method)
self._log.debug('Created covariance matrix - %s mode [%0.4f s]' %
(method, time.time() - t0))
return cov_matrix
@staticmethod
def _leafFocalDistance(leaf1, leaf2):
return num.sqrt((leaf1.focal_point[0] - leaf2.focal_point[0])**2 +
(leaf1.focal_point[1] - leaf2.focal_point[1])**2)
def _leafMapping(self, leaf1, leaf2):
if not isinstance(leaf1, str):
leaf1 = leaf1.id
if not isinstance(leaf2, str):
leaf2 = leaf2.id
if not self._initialized:
self.covariance_matrix_focal
try:
return self._leaf_mapping[leaf1], self._leaf_mapping[leaf2]
except KeyError as e:
raise KeyError('Unknown quadtree leaf with id %s' % e)
def getLeafCovariance(self, leaf1, leaf2):
'''Get the covariance between ``leaf1`` and ``leaf2`` from
distances.
:param leaf1: Leaf one
:type leaf1: str of `leaf.id` or :class:`~kite.quadtree.QuadNode`
:param leaf2: Leaf two
:type leaf2: str of `leaf.id` or :class:`~kite.quadtree.QuadNode`
:returns: Covariance between ``leaf1`` and ``leaf2``
:rtype: float
'''
return self.covariance_matrix[self._leafMapping(leaf1, leaf2)]
def getLeafWeight(self, leaf, model='focal'):
''' Get the total weight of ``leaf``, which is the summation of
all single pair weights of :attr:`kite.Covariance.weight_matrix`.
.. math ::
w_{x} = \\sum_i W_{x,i}
:param model: ``Focal`` or ``full``, default ``focal``
:type model: str
:param leaf: A leaf from :class:`~kite.Quadtree`
:type leaf: :class:`~kite.quadtree.QuadNode`
:returns: Weight of the leaf
:rtype: float
'''
(nl, _) = self._leafMapping(leaf, leaf)
weight_mat = self.weight_matrix_focal
return num.mean(weight_mat, axis=0)[nl]
def syntheticNoise(self, shape=(1024, 1024), dEdN=None, anisotropic=False):
'''Create random synthetic noise from data noise power spectrum.
This function uses the power spectrum of the data noise
(:attr:`noise_data`) (:func:`powerspecNoise`) to create synthetic
noise, e.g. to use it for data pertubation in optinmizations.
The default sampling distances are taken from
:attr:`kite.scene.Frame.dE` and :attr:`kite.scene.Frame.dN`. They can
be overwritten.
:param shape: shape of the desired noise patch.
Pixels in northing and easting (`nE`, `nN`),
defaults to `(1024, 1024)`.
:type shape: tuple, optional
:param dEdN: The sampling distance in easting, defaults to
(:attr:`kite.scene.Frame.dE`, :attr:`kite.scene.Frame.dN`).
:type dE: tuple, floats
:returns: synthetic noise patch
:rtype: :class:`numpy.ndarray`
'''
if (shape[0] + shape[1]) % 2 != 0:
# self._log.warning('Patch dimensions must be even, '
# 'ceiling dimensions!')
pass
nE = shape[1] + (shape[1] % 2)
nN = shape[0] + (shape[0] % 2)
rfield = num.random.rand(nN, nE)
spec = num.fft.fft2(rfield)
if not dEdN:
dE, dN = (self.scene.frame.dE, self.scene.frame.dN)
kE = num.fft.fftfreq(nE, dE)
kN = num.fft.fftfreq(nN, dN)
k_rad = num.sqrt(kN[:, num.newaxis]**2 + kE[num.newaxis, :]**2)
amp = num.zeros_like(k_rad)
if not anisotropic:
noise_pspec, k, _, _, _, _ = self.powerspecNoise2D()
k_bin = num.insert(k + k[0] / 2, 0, 0)
for i in xrange(k.size):
k_min = k_bin[i]
k_max = k_bin[i + 1]
r = num.logical_and(k_rad > k_min, k_rad <= k_max)
if i == (k.size - 1):
r = k_rad > k_min
if r.sum() == 0:
continue
amp[r] = noise_pspec[i]
amp[k_rad == 0.] = self.variance
amp[k_rad > k.max()] = noise_pspec[num.argmax(k)]
amp = num.sqrt(amp * self.noise_data.size * num.pi * 4)
elif anisotropic:
interp_pspec, _, _, _, skE, skN = self.powerspecNoise3D()
kE = num.fft.fftshift(kE)
kN = num.fft.fftshift(kN)
mkE = num.logical_and(kE >= skE.min(), kE <= skE.max())
mkN = num.logical_and(kN >= skN.min(), kN <= skN.max())
mkRad = num.where( # noqa
k_rad < num.sqrt(kN[mkN].max()**2 + kE[mkE].max()**2))
res = interp_pspec(kN[mkN, num.newaxis],
kE[num.newaxis, mkE],
grid=True)
print amp.shape, res.shape
print kN.size, kE.size
amp = res
amp = num.fft.fftshift(amp)
print amp.min(), amp.max()
spec *= amp
noise = num.abs(num.fft.ifft2(spec))
noise -= num.mean(noise)
return noise
def powerspecNoise1D(self, data=None, ndeg=512, nk=512):
if self._powerspec1d_cached is None:
self._powerspec1d_cached = self._powerspecNoise(data,
norm='1d',
ndeg=ndeg,
nk=nk)
return self._powerspec1d_cached
def powerspecNoise2D(self, data=None, ndeg=512, nk=512):
if self._powerspec2d_cached is None:
self._powerspec2d_cached = self._powerspecNoise(data,
norm='2d',
ndeg=ndeg,
nk=nk)
return self._powerspec2d_cached
def powerspecNoise3D(self, data=None):
if self._powerspec3d_cached is None:
self._powerspec3d_cached = self._powerspecNoise(data, norm='3d')
return self._powerspec3d_cached
def _powerspecNoise(self, data=None, norm='1d', ndeg=512, nk=512):
'''Get the noise power spectrum from
:attr:`kite.Covariance.noise_data`.
:param data: Overwrite Covariance.noise_data, defaults to `None`
:type data: :class:`numpy.ndarray`, optional
:returns: `(power_spec, k, f_spectrum, kN, kE)`
:rtype: tuple
'''
if data is None:
noise = self.noise_data
else:
noise = data.copy()
if norm not in ['1d', '2d', '3d']:
raise AttributeError('norm must be either 1d, 2d or 3d')
# noise = squareMatrix(noise)
shift = num.fft.fftshift
spectrum = shift(num.fft.fft2(noise, axes=(0, 1), norm=None))
power_spec = (num.abs(spectrum) / spectrum.size)**2
kE = shift(
num.fft.fftfreq(power_spec.shape[1], d=self.quadtree.frame.dE))
kN = shift(
num.fft.fftfreq(power_spec.shape[0], d=self.quadtree.frame.dN))
k_rad = num.sqrt(kN[:, num.newaxis]**2 + kE[num.newaxis, :]**2)
power_spec[k_rad == 0.] = 0.
power_interp = sp.interpolate.RectBivariateSpline(kN, kE, power_spec)
# def power1d(k):
# theta = num.linspace(-num.pi, num.pi, ndeg, False)
# power = num.empty_like(k)
# for i in xrange(k.size):
# kE = num.cos(theta) * k[i]
# kN = num.sin(theta) * k[i]
# power[i] = num.median(power_interp.ev(kN, kE)) * k[i]\
# * num.pi * 4
# return power
def power1d(k):
theta = num.linspace(-num.pi, num.pi, ndeg, False)
power = num.empty_like(k)
for i in xrange(k.size):
kE = num.cos(theta) * k[i]
kN = num.sin(theta) * k[i]
power[i] = num.median(power_interp.ev(kN, kE))
return power
def power2d(k):
''' Mean 2D Power works! '''
theta = num.linspace(-num.pi, num.pi, ndeg, False)
power = num.empty_like(k)
for i in xrange(k.size):
kE = num.sin(theta) * k[i]
kN = num.cos(theta) * k[i]
power[i] = num.median(power_interp.ev(kN, kE))
# Median is more stable than the mean here
return power
def power3d(k):
return power_interp
power = power1d
if norm == '2d':
power = power2d
elif norm == '3d':
power = power3d
k_rad = num.sqrt(kN[:, num.newaxis]**2 + kE[num.newaxis, :]**2)
k = num.linspace(k_rad[k_rad > 0].min(), k_rad.max(), nk)
dk = 1. / k.min() / (2. * nk)
return power(k), k, dk, spectrum, kE, kN
# def power1Ddisc():
# self._log.info('Using discrete summation')
# ps = power_spec
# d = num.abs(num.arange(-ps.shape[0]/2,
# ps.shape[0]/2))
# rm = num.sqrt(d[:, num.newaxis]**2 + d[num.newaxis, :]**2)
# axis = num.argmax(ps.shape)
# k_ref = kN if axis == 0 else kE
# p = num.empty(ps.shape[axis]/2)
# k = num.empty(ps.shape[axis]/2)
# for r in xrange(ps.shape[axis]/2):
# mask = num.logical_and(rm >= r-.5, rm < r+.5)
# k[r] = k_ref[(k_ref.size/2)+r]
# p[r] = num.median(ps[mask]) * 4 * num.pi
# return p, k
# power, k = power1Ddisc()
# dk = k[1] - k[0]
# return power, k, dk, spectrum, kE, kN
def _powerspecFit(self, regime=3):
'''Fitting a function to data noise power spectrum.
'''
power_spec, k, _, _, _, _ = self.powerspecNoise1D()
def selectRegime(k, k1, k2):
return num.logical_and(k > k1, k < k2)
regime = selectRegime(k, *noise_regimes[regime])
try:
return sp.optimize.curve_fit(modelPowerspec,
k[regime],
power_spec[regime],
p0=(0.1, 2000))
except RuntimeError as e:
self._log.warning('Could not fit the powerspectrum model. <%s>' %
e)
return (0., 0.), 0.
@property
def powerspec_model(self):
'''Fit function to power spectrum based on the spectral model parameters
:func:`~kite.covariance.modelPowerspec`
:returns: Model parameters ``a`` and ``b``
:rtype: tuple, floats
'''
p, _ = self._powerspecFit()
return p
@property
def powerspec_model_rms(self):
'''
:getter: RMS missfit between :class:`~kite.Covariance.powerspecNoise1D`
and :class:`~kite.Covariance.powerspec_model``
:type: float
'''
power_spec, k, _, _, _, _ = self.powerspecNoise1D()
power_spec_mod = self.powerspecModel(k)
return num.sqrt(num.mean((power_spec - power_spec_mod)**2))
def powerspecModel(self, k):
''' Calculates the model power spectrum based on the fit of
:func:`~kite.covariance.powerspec_model`.
:param k: Wavenumber(s)
:type k: float or :class:`numpy.ndarray`
:returns: Power at wavenumber ``k``
:rtype: float or :class:`numpy.ndarray`
'''
p = self.powerspec_model
return modelPowerspec(k, *p)
def _powerCosineTransform(self, p_spec):
'''Calculating the cosine transform of the power spectrum.
The cosine transform of the power spectrum is an estimate
of the data covariance (see Hanssen, 2001).'''
cos = sp.fftpack.idct(p_spec, type=3)
return cos
@property_cached
def covariance_func(self):
''' Covariance function estimated directly from the power spectrum of
displacement noise patch using the cosine transform.
:type: tuple, :class:`numpy.ndarray` (covariance, distance) '''
power_spec, k, dk, _, _, _ = self.powerspecNoise1D()
# power_spec -= self.variance
d = num.arange(1, power_spec.size + 1) * dk
cov = self._powerCosineTransform(power_spec)
return cov, d
def covarianceFromModel(self, regime=0):
'''Caluclate exponential analytical covariance
Empirical Covariance function based on the power spectral model fit
and not directly on the power spectrum as in
:func:`~kite.covariance.covariance_func`
from :func:`~kite.covariance.modelPowerspec`
.. warning:: Deprecated!
:return: Covariance and corresponding distances.
:rtype: tuple, :class:`numpy.ndarray` (covariance_analytical, distance)
'''
_, k, dk, _, kN, kE = self.powerspecNoise1D()
(a, b) = self.powerspec_model
spec = modelPowerspec(k, a, b)
d = num.arange(1, spec.size + 1) * dk
cos = self._powerCosineTransform(spec)
return cos, d
@property
def covariance_model(self, regime=0):
''' Covariance model parameters for
:func:`~kite.covariance.modelCovariance` retrieved
from :attr:`~kite.Covariance.covarianceFromModel`.
.. note:: using this function implies several model
fits: (1) fit of the spectrum and (2) the cosine transform.
Not sure about the consequences, if this is useful and/or
meaningful.
:getter: Get the parameters.
:type: tuple, ``a`` and ``b``
'''
if self.config.a is None or self.config.b is None:
cov, d = self.covarianceFromModel(regime)
cov, d = self.covariance_func
try:
(a, b), _ =\
sp.optimize.curve_fit(modelCovariance, d, cov,
p0=(.001, 500.))
self.config.a, self.config.b = (float(a), float(b))
except RuntimeError:
self._log.warning('Could not fit the covariance model')
self.config.a, self.config.b = (1., 1000.)
return self.config.a, self.config.b
@property
def covariance_model_rms(self):
'''
:getter: RMS missfit between :class:`~kite.Covariance.covariance_model`
and :class:`~kite.Covariance.covariance_func`
:type: float
'''
cov, d = self.covariance_func
cov_mod = modelCovariance(d, *self.covariance_model)
return num.sqrt(num.mean((cov - cov_mod)**2))
@property_cached
def structure_func(self):
''' Structure function derived from ``noise_patch``
:type: tuple, :class:`numpy.ndarray` (structure_func, distance)
Adapted from
http://clouds.eos.ubc.ca/~phil/courses/atsc500/docs/strfun.pdf
'''
power_spec, k, dk, _, _, _ = self.powerspecNoise1D()
d = num.arange(1, power_spec.size + 1) * dk
def structure_func(power_spec, d, k):
struc_func = num.zeros_like(k)
for i, d in enumerate(d):
for ik, tk in enumerate(k):
# struc_func[i] += (1. - num.cos(tk*d))*power_spec[ik]
struc_func[i] += (1. -
sp.special.j0(tk * d)) * power_spec[ik]
struc_func *= 2. / 1
return struc_func
struc_func = structure_func(power_spec, d, k)
return struc_func, d
@property
def variance(self):
''' Variance of data noise estimated from the
high-frequency end of power spectrum.
:setter: Set the variance manually
:getter: Retrieve the variance
:type: float
'''
return self.config.variance
@variance.setter
def variance(self, value):
self.config.variance = float(value)
self._clear(config=False, spectrum=False)
self.evChanged.notify()
@variance.getter
def variance(self):
if self.config.variance is None:
power_spec, k, dk, spectrum, _, _ = self.powerspecNoise1D()
cov, _ = self.covariance_func
ma, mb = self.covariance_model
# print cov[1]
ps = power_spec * spectrum.size
# print spectrum.size
# print num.mean(ps[-int(ps.size/9.):-1])
var = num.median(ps[-int(ps.size / 9.):]) + ma
self.config.variance = float(var)
return self.config.variance
def export_weight_matrix(self, filename):
''' Export the full :attr:`~kite.Covariance.weight_matrix` to an ASCII
file. The data can be loaded through :func:`numpy.loadtxt`.
:param filename: path to export to
:type filename: str
'''
self._log.debug('Exporting Covariance.weight_matrix to %s' % filename)
header = 'Exported kite.Covariance.weight_matrix, '\
'for more information visit http://pyrocko.com\n'\
'\nThe matrix is symmetric and ordered by QuadNode.id:\n'
header += ', '.join([l.id for l in self.quadtree.leaves])
num.savetxt(filename, self.weight_matrix, header=header)
@property_cached
def plot(self):
''' Simple overview plot to summarize the covariance estimations.
'''
from kite.plot2d import CovariancePlot
return CovariancePlot(self)
class Frame(object):
''' UTM frame holding geographical references for :class:`kite.scene.Scene`
'''
evChanged = Subject()
def __init__(self, scene, config=FrameConfig()):
self._scene = scene
self._log = scene._log.getChild('Frame')
self.extentE = 0.
self.extentN = 0.
self.spherical_distortion = 0.
self.urE = 0.
self.urN = 0.
self.llEutm = None
self.llNutm = None
self.utm_zone = None
self.llN = None
self.llE = None
self.N = None
self.E = None
self.offsetE = 0.
self.offsetN = 0.
self._updateConfig(config)
self._scene.evConfigChanged.subscribe(self._updateConfig)
self._scene.evChanged.subscribe(self.updateExtent)
def _updateConfig(self, config=None):
if config is not None:
self.config = config
elif self.config != self._scene.config.frame:
self.config = self._scene.config.frame
else:
return
self.updateExtent()
self.evChanged.notify()
def updateExtent(self):
if self._scene.cols == 0 or self._scene.rows == 0:
return
self.llEutm, self.llNutm, self.utm_zone, self.utm_zone_letter = \
utm.from_latlon(self.llLat, self.llLon)
self.cols = self._scene.cols
self.rows = self._scene.rows
urlat = self.llLat + self.dLat * self.rows
urlon = self.llLon + self.dLon * self.cols
self.urEutm, self.urNutm, _, _ = utm.from_latlon(
urlat, urlon, self.utm_zone)
# Width at the bottom of the scene
self.extentE = greatCircleDistance(self.llLat, self.llLon, self.llLat,
urlon)
self.extentN = greatCircleDistance(self.llLat, self.llLon, urlat,
self.llLon)
# Width at the N' top of the scene
extentE_top = greatCircleDistance(urlat, self.llLon, urlat, urlon)
self.spherical_distortion = num.abs(self.extentE - extentE_top)
self.dE = (self.extentE + extentE_top) / (2 * self.cols)
self.dN = self.extentN / self.rows
self.E = num.arange(self.cols) * self.dE + self.offsetE
self.N = num.arange(self.rows) * self.dN + self.offsetN
self.llE = 0
self.llN = 0
self.urE = self.E.max()
self.urN = self.N.max()
self.gridE = None
self.gridN = None
self.coordinates = None
self.config.regularize()
return
@property
def llLat(self):
return self.config.llLat
@llLat.setter
def llLat(self, llLat):
self.config.llLat = llLat
self._llLat = llLat
@property
def llLon(self):
return self.config.llLon
@llLon.setter
def llLon(self, llLon):
self.config.llLon = llLon
self._llLon = llLon
@property
def dLat(self):
return self.config.dLat
@dLat.setter
def dLat(self, dLat):
self.config.dLat = dLat
@property
def dLon(self):
return self.config.dLon
@dLon.setter
def dLon(self, dLon):
self.config.dLon = dLon
@property
def dN(self):
return self.config.dN
@dN.setter
def dN(self, dN):
self.config.dN = dN
@property
def dE(self):
return self.config.dE
@dE.setter
def dE(self, dE):
self.config.dE = dE
@property_cached
def gridE(self):
''' UTM grid holding eastings of all pixels in ``NxM`` matrix
of :attr:`~kite.Scene.displacement`.
:type: :class:`numpy.ndarray`, size ``NxM``
'''
valid_data = num.isnan(self._scene.displacement)
gridE = num.repeat(self.E[num.newaxis, :], self.rows, axis=0)
return num.ma.masked_array(gridE, valid_data, fill_value=num.nan)
@property_cached
def gridN(self):
''' UTM grid holding northings of all pixels in ``NxM`` matrix
of :attr:`~kite.Scene.displacement`.
:type: :class:`numpy.ndarray`, size ``NxM``
'''
valid_data = num.isnan(self._scene.displacement)
gridN = num.repeat(self.N[:, num.newaxis], self.cols, axis=1)
return num.ma.masked_array(gridN, valid_data, fill_value=num.nan)
@property_cached
def coordinates(self):
coords = num.empty((self.rows * self.cols, 2))
coords[:, 0] = num.repeat(self.E[num.newaxis, :], self.rows,
axis=0).flatten()
coords[:, 1] = num.repeat(self.N[:, num.newaxis], self.cols,
axis=1).flatten()
return coords
def setENOffset(self, east, north):
'''Set scene offsets in local cartesian coordinates.
:param east: East offset in [m]
:type east: float, :class:`numpy.ndarray` or None
:param north: North offset in [m]
:type north: float, :class:`numpy.ndarray` or None
'''
self.offsetE = east
self.offsetN = north
self.updateExtent()
def setLatLonReference(self, lat, lon):
pass
def mapMatrixEN(self, row, col):
''' Maps matrix row, column to local easting and northing.
:param row: Matrix row number
:type row: int
:param col: Matrix column number
:type col: int
:returns: Easting and northing in local coordinates
:rtype: tuple (float), (easting, northing)
'''
return row * self.dE, col * self.dN
def mapENMatrix(self, E, N):
''' Maps local coordinates (easting and northing) to matrix
row and column
:param E: Easting in local coordinates
:type E: float
:param N: Northing in local coordinates
:type N: float
:returns: Row and column
:rtype: tuple (int), (row, column)
'''
row = int(E / self.dE) if E > 0 else 0
col = int(N / self.dN) if N > 0 else 0
return row, col
def __eq__(self, other):
return self.llLat == other.llLat and\
self.llLon == other.llLon and\
self.dE == other.dE and\
self.dN == other.dN and\
self.rows == other.rows and\
self.cols == other.cols
def __str__(self):
return (
'Lower right latitude: {frame.llLat:.4f} N\n'
'Lower right longitude: {frame.llLon:.4f} E\n'
'\n\n'
'UTM Zone: {frame.utm_zone}{frame.utm_zone_letter}\n'
'Lower right easting: {frame.llE:.4f} m\n'
'Lower right northing: {frame.llN:.4f} m'
'\n\n'
'Pixel spacing east: {frame.dE:.4f} m\n'
'Pixel spacing north: {frame.dN:.4f} m\n'
'Extent east: {frame.extentE:.4f} m\n'
'Extent north: {frame.extentN:.4f} m\n'
'Dimensions: {frame.cols} x {frame.rows} px\n'
'Spherical distortion: {frame.spherical_distortion:.4f} m\n'
).format(frame=self)
class Scene(BaseScene):
'''Scene of unwrapped InSAR ground dispacements measurements
:param config: Configuration object
:type config: :class:`SceneConfig`, optional
Optional parameters
:param displacement: Displacement in [m]
:type displacement: :class:`numpy.ndarray`, NxM, optional
:param theta: Theta look angle, see :attr:`Scene.theta`
:type theta: :class:`numpy.ndarray`, NxM, optional
:param phi: Phi look angle, see :attr:`Scene.phi`
:type phi: :class:`numpy.ndarray`, NxM, optional
:param llLat: Lower left latitude in [deg]
:type llLat: float, optional
:param llLon: Lower left longitude in [deg]
:type llLon: float, optional
:param dLat: Pixel spacing in latitude [deg]
:type dLat: float, optional
:param dLon: Pixel spacing in longitude [deg]
:type dLon: float, optional
'''
evChanged = Subject()
evConfigChanged = Subject()
def __init__(self, config=SceneConfig(), **kwargs):
self.config = config
self.meta = self.config.meta
BaseScene.__init__(self, frame_config=self.config.frame, **kwargs)
# wiring special methods
self.import_data = self._import_data
self.load = self._load
@property_cached
def quadtree(self):
''' Instanciates the scene's quadtree.
:type: :class:`kite.quadtree.Quadtree`
'''
self._log.debug('Creating kite.Quadtree instance')
from kite.quadtree import Quadtree
return Quadtree(scene=self, config=self.config.quadtree)
@property_cached
def covariance(self):
''' Instanciates the scene's covariance attribute.
:type: :class:`kite.covariance.Covariance`
'''
self._log.debug('Creating kite.Covariance instance')
from kite.covariance import Covariance
return Covariance(scene=self, config=self.config.covariance)
@property_cached
def plot(self):
''' Shows a simple plot of the scene's displacement
'''
self._log.debug('Creating kite.ScenePlot instance')
from kite.plot2d import ScenePlot
return ScenePlot(self)
def spool(self):
''' Start the spool user interface :class:`~kite.spool.Spool` to inspect
the scene.
'''
if self.displacement is None:
raise SceneError('Can not display an empty scene.')
from kite.spool import Spool
spool = Spool(scene=self)
return spool.spool_win.buildViews()
def _testImport(self):
try:
self.frame.E
self.frame.N
self.frame.gridE
self.frame.gridN
self.frame.dE
self.frame.dN
self.displacement
self.theta
self.phi
except Exception as e:
print e
raise ImportError('Something went wrong during import - '
'see Exception!')
def save(self, filename=None):
''' Save kite scene to kite file structure
Saves the current scene meta information and UTM frame to a YAML
(``.yml``) file. Numerical data (:attr:`~kite.Scene.displacement`,
:attr:`~kite.Scene.theta` and :attr:`~kite.Scene.phi`)
are saved as binary files from :class:`numpy.ndarray`.
:param filename: Filenames to save scene to, defaults to
' :attr:`~kite.Scene.meta.scene_id` ``_``
:attr:`~kite.Scene.meta.scene_view`
:type filename: str, optional
'''
filename = filename or '%s_%s' % (self.meta.scene_id,
self.meta.scene_view)
_file, ext = op.splitext(filename)
filename = _file if ext in ['.yml', '.npz'] else filename
components = ['displacement', 'theta', 'phi']
self._log.debug('Saving scene data to %s.npz' % filename)
num.savez('%s.npz' % (filename),
*[getattr(self, arr) for arr in components])
self.saveConfig('%s.yml' % filename)
def saveConfig(self, filename):
_file, ext = op.splitext(filename)
filename = filename if ext in ['.yml'] else filename + '.yml'
self._log.debug('Saving scene config to %s' % filename)
self.config.regularize()
self.config.dump(filename='%s' % filename,
header='kite.Scene YAML Config')
@dynamicmethod
def _load(self, filename):
''' Load a kite scene from file ``filename.[npz,yml]``
structure.
:param filename: Filenames the scene data is saved under
:type filename: str
:returns: Scene object from data resources
:rtype: :class:`~kite.Scene`
'''
scene = self
components = ['displacement', 'theta', 'phi']
basename = op.splitext(filename)[0]
scene._log.debug('Loading from %s[.npz,.yml]' % basename)
try:
data = num.load('%s.npz' % basename)
for i, comp in enumerate(components):
scene.__setattr__(comp, data['arr_%d' % i])
except IOError:
raise UserIOWarning('Could not load data from %s.npz' % basename)
try:
scene.load_config('%s.yml' % basename)
except IOError:
raise UserIOWarning('Could not load %s.yml' % basename)
scene.meta.filename = op.basename(filename)
scene._testImport()
return scene
load = staticmethod(_load)
def load_config(self, filename):
self._log.debug('Loading config from %s' % filename)
self.config = guts.load(filename=filename)
self.meta = self.config.meta
self.evConfigChanged.notify()
@dynamicmethod
def _import_data(self, path, **kwargs):
''' Import displacement data from foreign file format.
:param path: Filename of resource to import
:type path: str
:param kwargs: keyword arguments passed to import function
:type kwargs: dict
:returns: Scene from path
:rtype: :class:`~kite.Scene`
:raises: TypeError
'''
scene = self
if not op.isfile(path) or op.isdir(path):
raise ImportError('File %s does not exist!' % path)
data = None
for mod in scene_io.__all__:
module = eval('scene_io.%s(scene)' % mod)
if module.validate(path, **kwargs):
scene._log.debug('Importing %s using %s module' % (path, mod))
data = module.read(path, **kwargs)
break
if data is None:
raise ImportError('Could not recognize format for %s' % path)
scene.meta.filename = op.basename(path)
return scene._import_from_dict(scene, data)
_import_data.__doc__ += \
'\nSupported import modules are **%s**.\n'\
% (', ').join(scene_io.__all__)
for mod in scene_io.__all__:
_import_data.__doc__ += '\n**%s**\n\n' % mod
_import_data.__doc__ += eval('scene_io.%s.__doc__' % mod)
import_data = staticmethod(_import_data)
@staticmethod
def _import_from_dict(scene, data):
for sk in ['theta', 'phi', 'displacement']:
setattr(scene, sk, data[sk])
for fk, fv in data['frame'].iteritems():
setattr(scene.frame, fk, fv)
for mk, mv in data['meta'].iteritems():
if mv is not None:
setattr(scene.meta, mk, mv)
scene.meta.extra.update(data['extra'])
scene.frame.updateExtent()
scene._testImport()
return scene
def __str__(self):
return self.config.__str__()