def interp(self, xo, yo, geterr=False, blockr=None):
"""Interpolate to positions xo,yo
:Params:
- **xo/yo**: Output positions.
- **geterr**, optional: Also return errors.
:Return: ``zo`` or ``zo,eo``
"""
# Inits
xo = N.asarray(xo, 'd')
yo = N.asarray(yo, 'd')
npo = xo.shape[0]
vgf = self.variogram_func
so = (self.nt, npo) if self.nt else npo
zo = N.zeros(so, 'd')
if geterr:
eo = N.zeros(npo, 'd')
if self.ncloud>1 or geterr:
wo = N.zeros(npo, 'd')
# Loop on clouds
Ainv = self.Ainv
next = int(not self._simple)
for ic in xrange(self.ncloud): # TODO: multiproc here?
# Distances to output points
# dd = cdist(N.transpose([xi,yi]),N.transpose([xo,yo])) # TODO: test cdist
dd = get_distances(xo, yo, self.xc[ic], self.yc[ic], mode=self.distfunc)
# Form B
np = self.npc[ic]
B = N.empty((np+next, npo))
B[:self.npc[ic]] = vgf(dd)
if not self._simple:
B[-1] = 1
if self.exact:
B[:np][isclose(B[:np], 0.)] = 0.
del dd
# Block kriging
if blockr:
tree = cKDTree(N.transpose([xo, yo]))
Bb = B.copy()
for i, iineigh in enumerate(tree.query_ball_tree(tree, blockr)):
Bb[:, i] = B[:, iineigh].mean()
B = Bb
# Compute weights
W = N.ascontiguousarray(symm(Ainv[ic], N.asfortranarray(B, 'd')))
# Simple kriging with adjusted mean for long distance values
if self._simple and self.farvalue is not None:
s = self.get_sill()
Ais = self.get_sill() * Ainv[ic].sum(axis=0)
mean = self.farvalue - (self.zc[ic] * Ais).sum()
mean /= (1 - Ais.sum())
else:
mean = self.mean
# Interpolate
z = N.ascontiguousarray(dgemv(N.asfortranarray(W[:np].T, 'd'),
N.asfortranarray(self.zc[ic]-mean, 'd')))
if self._simple:
z += mean
# Simplest case
if not geterr and self.ncloud<2:
zo[:] = z.T
continue
# Get error
# e = (W[:-1]*B[:-1]).sum(axis=0)
e = (W*B).sum(axis=0)
del W, B
# Weigthed contribution based on errors
w = 1/e**2
if self.ncloud>1:
z[:] *= w
wo += w
del w
zo[:] += z.T ; del z
# Error
if geterr:
eo = 1/N.sqrt(wo)
# Normalization
if self.ncloud>1:
zo[:] /= wo
gc.collect()
if geterr: return zo, eo
return zo
def interp(self, xo, yo, geterr=False, blockr=None):
"""Interpolate to positions xo,yo
:Params:
- **xo/yo**: Output positions.
- **geterr**, optional: Also return errors.
:Return: ``zo`` or ``zo,eo``
"""
# Inits
xo = N.asarray(xo, 'd')
yo = N.asarray(yo, 'd')
npo = xo.shape[0]
vgf = self.variogram_func
so = (self.nt, npo) if self.nt else npo
zo = N.zeros(so, 'd')
if geterr:
eo = N.zeros(npo, 'd')
if self.ncloud > 1 or geterr:
wo = N.zeros(npo, 'd')
# Loop on clouds
Ainv = self.Ainv
next = int(not self._simple)
for ic in xrange(self.ncloud): # TODO: multiproc here?
# Distances to output points
# dd = cdist(N.transpose([xi,yi]),N.transpose([xo,yo])) # TODO: test cdist
dd = get_distances(xo,
yo,
self.xc[ic],
self.yc[ic],
mode=self.distfunc)
# Form B
np = self.npc[ic]
B = N.empty((np + next, npo))
B[:self.npc[ic]] = vgf(dd)
if not self._simple:
B[-1] = 1
if self.exact:
B[:np][isclose(B[:np], 0.)] = 0.
del dd
# Block kriging
if blockr:
tree = cKDTree(N.transpose([xo, yo]))
Bb = B.copy()
for i, iineigh in enumerate(tree.query_ball_tree(tree,
blockr)):
Bb[:, i] = B[:, iineigh].mean()
B = Bb
# Compute weights
W = N.ascontiguousarray(symm(Ainv[ic], N.asfortranarray(B, 'd')))
# Simple kriging with adjusted mean for long distance values
if self._simple and self.farvalue is not None:
s = self.get_sill()
Ais = self.get_sill() * Ainv[ic].sum(axis=0)
mean = self.farvalue - (self.zc[ic] * Ais).sum()
mean /= (1 - Ais.sum())
else:
mean = self.mean
# Interpolate
z = N.ascontiguousarray(
dgemv(N.asfortranarray(W[:np].T, 'd'),
N.asfortranarray(self.zc[ic] - mean, 'd')))
if self._simple:
z += mean
# Simplest case
if not geterr and self.ncloud < 2:
zo[:] = z.T
continue
# Get error
# e = (W[:-1]*B[:-1]).sum(axis=0)
e = (W * B).sum(axis=0)
del W, B
# Weigthed contribution based on errors
w = 1 / e**2
if self.ncloud > 1:
z[:] *= w
wo += w
del w
zo[:] += z.T
del z
# Error
if geterr:
eo = 1 / N.sqrt(wo)
# Normalization
if self.ncloud > 1:
zo[:] /= wo
gc.collect()
if geterr: return zo, eo
return zo
def interp(self, xo, yo, geterr=False, blockr=None):
"""Interpolate to positions xo,yo
:Params:
- **xo/yo**: Output positions.
- **geterr**, optional: Also return errors.
:Return: ``zo`` or ``zo,eo``
"""
# Inits
xo = N.asarray(xo, 'd')
yo = N.asarray(yo, 'd')
npo = xo.shape[0]
vgf = self.variogram_func
so = (self.nt, npo) if self.nt else npo
zo = N.zeros(so, 'd')
if geterr:
eo = N.zeros(npo, 'd')
if self.ncloud>1 or geterr:
wo = N.zeros(npo, 'd')
# Loop on clouds
Ainv = self.Ainv
for ic in xrange(self.ncloud): # TODO: multiproc here?
# Distances to output points
# dd = cdist(N.transpose([xi,yi]),N.transpose([xo,yo])) # TODO: test cdist
xxo, xxi = N.meshgrid(xo, self.xc[ic])
dd = (xxo-xxi)**2 ; del xxi, xxo
yyo, yyi = N.meshgrid(yo, self.yc[ic])
dd += (yyo-yyi)**2 ; del yyi, yyo
dd = N.sqrt(dd)
# Form B
B = N.empty((self.npc[ic]+1, npo))
B[-1] = 1
B[:-1] = vgf(dd)
if self.exact:
B[:-1][isclose(B[:-1], 0.)] = 0.
del dd
# Block kriging
if blockr:
tree = cKDTree(N.transpose([xo, yo]))
Bb = B.copy()
for i, iineigh in enumerate(tree.query_ball_tree(tree, blockr)):
Bb[:, i] = B[:, iineigh].mean()
B = Bb
# Compute weights
W = N.ascontiguousarray(symm(Ainv[ic], N.asfortranarray(B, 'd')))
# Interpolate
z = N.ascontiguousarray(dgemv(N.asfortranarray(W[:-1].T, 'd'),
N.asfortranarray(self.zc[ic], 'd')))
# Simplest case
if not geterr and self.ncloud<2:
zo[:] = z.T
continue
# Get error
# e = (W[:-1]*B[:-1]).sum(axis=0)
e = (W*B).sum(axis=0)
del W, B
# Weigthed contribution based on errors
w = 1/e**2
if self.ncloud>1:
z[:] *= w
wo += w
del w
zo[:] += z.T ; del z
# Error
if geterr:
eo = 1/N.sqrt(wo)
# Normalization
if self.ncloud>1:
zo[:] /= wo
gc.collect()
if geterr: return zo, eo
return zo