def w_cache_predict(theta, lam, uvw, src, guv,
wstep=2000,
kernel_cache=None,
kernel_fn=w_kernel,
**kwargs):
"""Predict visibilities using w-kernel cache
:param theta: Field of view (directional cosines)
:param lam: UV grid range (wavelenghts)
:param uvw: UVWs of visibilities (wavelengths)
:param guv: Input uv grid to de-grid from
:param wstep: Size of w-bins (wavelengths)
:param kernel_cache: Kernel cache. If not passed, we fall back
to `kernel_fn`. See `w_cache_imaging` for details.
:param kernel_fn: Function for generating the kernels. Parameters
`(theta, w, **kwargs)`. Default `w_kernel`.
:returns: degridded visibilities
"""
if kernel_cache is None:
kernel_cache = pylru.FunctionCacheManager(kernel_fn, 1)
def kernel_binner(theta, w, **kw):
wbin = wstep * numpy.round(w / wstep)
return kernel_cache(theta, wbin, **kw)
v = numpy.ndarray(nv, dtype=complex)
for p, s in zip(uvw, src):
wbin = wstep * numpy.round(p[2] / wstep)
wg = kernel_cache(theta, wbin, *s, **kwargs)
v[ixs] = convdegrid(wg, guv, numpy.array([p / lam]))
return v
def w_cache_predict(theta,
lam,
p,
guv,
wstep=2000,
kernel_cache=None,
kernel_fn=w_kernel,
**kwargs):
"""Predict visibilities using w-kernel cache
:param theta: Field of view (directional cosines)
:param lam: UV grid range (wavelenghts)
:param p: UVWs of visibilities (wavelengths)
:param guv: Input uv grid to de-grid from
:param wstep: Size of w-bins (wavelengths)
:param kernel_cache: Kernel cache. If not passed, we fall back
to `kernel_fn`. See `w_cache_imaging` for details.
:param kernel_fn: Function for generating the kernels. Parameters
`(theta, w, **kwargs)`. Default `w_kernel`.
:returns: degridded visibilities
"""
if kernel_cache is None:
kernel_cache = pylru.FunctionCacheManager(kernel_fn, 1000)
def kernel_binner(theta, w, **kw):
wbin = wstep * numpy.round(w / wstep)
return kernel_cache(theta, wbin, **kw)
return w_slice_predict(theta, lam, p, guv, 1, kernel_binner, **kwargs)
def w_cache_imaging(theta,
lam,
p,
v,
wstep=2000,
kernel_cache=None,
kernel_fn=w_kernel,
**kwargs):
"""Basic w-projection by caching convolution arl in w
A simple cache can be constructed externally and passed in:
kernel_cache = pylru.FunctionCacheManager(w_kernel, cachesize)
If applicable, consider wrapping in `w_conj_kernel_fn` to improve
effectiveness further.
:param theta: Field of view (directional cosines)
:param lam: UV grid range (wavelenghts)
:param p: UVWs of visibilities (wavelengths)
:param v: Visibilities to be imaged
:param wstep: Size of w-bins (wavelengths)
:param kernel_cache: Kernel cache. If not passed, we fall back
to `kernel_fn`.
:param kernel_fn: Function for generating the kernels. Parameters
`(theta, w, **kwargs)`. Default `w_kernel`.
:returns: UV grid
"""
# Construct default cache, if needed. As visibilities are
# traversed in w-order it only needs to hold the last w-kernel.
if kernel_cache is None:
kernel_cache = pylru.FunctionCacheManager(kernel_fn, 1000)
# Bin w values, then run slice imager with slice size of 1
def kernel_binner(theta, w, **kw):
wbin = wstep * numpy.round(w / wstep)
return kernel_cache(theta, wbin, **kw)
N = int(round(theta * lam))
assert N > 1
guv = numpy.zeros([N, N], dtype=complex)
for ps, vs in slices:
w = numpy.mean(ps[:, 2])
wg = numpy.conj(kernel_fn(theta, w, **kwargs))
convolutional_grid(wg, guv, ps / lam, vs)
return guv
def w_cache_imaging(theta,
lam,
uvw,
src,
vis,
wstep=2000,
kernel_cache=None,
kernel_fn=w_kernel,
**kwargs):
"""Basic w-projection by caching convolution arl in w
A simple cache can be constructed externally and passed in:
kernel_cache = pylru.FunctionCacheManager(w_kernel, cachesize)
If applicable, consider wrapping in `w_conj_kernel_fn` to improve
effectiveness further.
:param theta: Field of view (directional cosines)
:param lam: UV grid range (wavelenghts)
:param uvw: UVWs of visibilities (wavelengths)
:param src: Visibility source information (various)
:param vis: Visibilites to be imaged
:param wstep: Size of w-bins (wavelengths)
:param kernel_cache: Kernel cache. If not passed, we fall back
to `kernel_fn`.
:param kernel_fn: Function for generating the kernels. Parameters
`(theta, w, **kwargs)`. Default `w_kernel`.
:returns: UV grid
"""
# Construct default cache, if needed. As visibilities are
# traversed in w-order it only needs to hold the last w-kernel.
if kernel_cache is None:
kernel_cache = pylru.FunctionCacheManager(kernel_fn, 1)
N = int(round(theta * lam))
guv = numpy.zeros([N, N], dtype=complex)
for p, s, v in zip(uvw, src, vis):
wbin = wstep * numpy.round(p[2] / wstep)
wg = kernel_cache(theta, wbin, *s, **kwargs)
convgrid(wg, guv, numpy.array([p / lam]), [()], numpy.array([v]))
return guv
def invert_visibility(vis: Visibility, params={}):
"""Invert to make dirty Image and PSF
:param vis:
:type Visibility: Visibility to be processed
:returns: (dirty image, psf)
"""
log_parameters(params)
log.debug("invert_visibility: Inverting Visibility to make dirty and psf")
shape, reffrequency, cellsize, w, imagecentre = create_wcs_from_visibility(vis, params=params)
npixel = shape[3]
theta = npixel * cellsize
log.debug("invert_visibility: Specified npixel=%d, cellsize = %f rad, FOV = %f rad" %
(npixel, cellsize, theta))
# Set up the gridding kernel. We try to use a cached version
gridding_algorithm = get_parameter(params, 'gridding_algorithm', 'wprojection')
if gridding_algorithm == 'wprojection':
log.debug("invert_visibility: Gridding by w projection")
wstep = get_parameter(params, "wstep", 10000.0)
wcachesize = w_cache_size(vis, wstep)
log.debug("invert_visibility: Making w-kernel cache of %d kernels" % wcachesize)
cache_fn = w_conj_kernel_fn(pylru.FunctionCacheManager(w_kernel, wcachesize))
imgfn = functools.partial(w_cache_imaging,
wstep=wstep, kernel_cache=cache_fn,
NpixFF=256, NpixKern=15, Qpx=4)
else:
raise NotImplementedError("gridding algorithm %s not supported" % gridding_algorithm)
# Apply a phase rotation from the visibility phase centre to the image phase centre
# visphaserotate = phaserotate(vis, imagecentre)
d = numpy.zeros(shape)
p = numpy.zeros(shape)
spectral_mode = get_parameter(params, 'spectral_mode', 'channel')
log.debug('invert_visibility: spectral mode is %s' % spectral_mode)
if spectral_mode == 'channel':
pmax = 0.0
nchan = shape[0]
npol = shape[1]
for channel in range(nchan):
for pol in range(npol):
log.debug('invert_visibility: Inverting channel %d, polarisation %d' % (channel, pol))
d[channel, pol, :, :], p[channel, 0, :, :], pmax = \
do_imaging(theta, 1.0 / cellsize, vis.uvw_lambda(channel),
vis.vis[:, channel, pol], imgfn=imgfn)
assert pmax > 0.0, ("No data gridded for channel %d" % channel)
else:
raise NotImplementedError("mode %s not supported" % spectral_mode)
dirty = create_image_from_array(d, w)
psf = create_image_from_array(p, w)
log.debug("invert_visibility: Finished making dirty and psf")
return dirty, psf, pmax
def predict_visibility(vis: Visibility, sm: SkyModel, params={}) -> Visibility:
"""Predict the visibility from a SkyModel including both components and images
:param vis:
:type Visibility: Visibility to be processed
:param sm:
:type SkyModel:
:returns: Visibility
"""
shape, reffrequency, cellsize, w, imagecentre = create_wcs_from_visibility(vis, params=params)
# Create copy of visibilities
vis = copy.copy(vis)
vis.data = copy.copy(vis.data)
vis.data['vis'] = numpy.zeros(vis.vis.shape)
spectral_mode = get_parameter(params, 'spectral_mode', 'channel')
log.debug('predict_visibility: spectral mode is %s' % spectral_mode)
if len(sm.images):
log.debug("predict_visibility: Predicting Visibility from sky model images")
for im in sm.images:
assert_same_chan_pol(vis, im)
# Determine image size
cellsize = abs(im.wcs.wcs.cdelt[0]) * numpy.pi / 180.0
theta = im.npixel * cellsize
log.debug("predict_visibility: Image cellsize %f radians" % cellsize)
log.debug("predict_visibility: Field of view %f radians" % theta)
assert (theta / numpy.sqrt(2) < 1.0), "Field of view larger than celestial sphere"
# Parameterise imaging
wstep = get_parameter(params, "wstep", 10000.0)
wcachesize = w_cache_size(vis, wstep)
log.debug("predict_visibility: Making w-kernel cache of %d kernels" % wcachesize)
cache_fn = w_conj_kernel_fn(pylru.FunctionCacheManager(w_kernel, wcachesize))
predfn = functools.partial(w_cache_predict,
wstep=wstep, kernel_cache=cache_fn,
NpixFF=256, NpixKern=15, Qpx=4)
spectral_mode = get_parameter(params, 'spectral_mode', 'channel')
log.debug('predict_visibility: spectral mode is %s' % spectral_mode)
if spectral_mode == 'channel':
for channel in range(im.nchan):
uvw = vis.uvw_lambda(channel)
for pol in range(im.npol):
log.debug('predict_visibility: Predicting from image channel %d, polarisation %d' % (
channel, pol))
img = sm.images[0].data[channel, pol, :, :]
dv = do_predict(theta, 1.0 / cellsize, numpy.array(uvw), img, predfn)
vis.vis[:, channel, pol] += dv
else:
raise NotImplementedError("mode %s not supported" % spectral_mode)
log.debug("predict_visibility: Finished predicting Visibility from sky model images")
if len(sm.components):
log.debug("predict_visibility: Predicting Visibility from sky model components")
for icomp, comp in enumerate(sm.components):
log.debug("predict_visibility: visibility shape = %s" % str(vis.vis.shape))
assert_same_chan_pol(vis, comp)
l,m,n = skycoord_to_lmn(comp.direction, vis.phasecentre)
log.debug('fourier_transforms.predict_visibility: Cartesian representation of component %d = (%f, %f, %f)'
% (icomp, l,m,n))
if spectral_mode =='channel':
for channel in range(comp.nchan):
uvw = vis.uvw_lambda(channel)
phasor = simulate_point(uvw, l, m)
for pol in range(comp.npol):
log.debug(
'predict_visibility: Predicting from component %d channel %d, polarisation %d' % (
icomp, channel,
pol))
vis.vis[:, channel, pol] += comp.flux[channel, pol] * phasor
else:
raise NotImplementedError("mode %s not supported" % spectral_mode)
log.debug("predict_visibility: Finished predicting Visibility from sky model components")
return vis
# calculating: 1 + 2
# 3
# 3
# calculating: 2 + 3
# 5
# 方法2 pylru
# https://pypi.org/project/pylru/
def square(x):
print('not hit cache')
return x * x
size = 100
cached = pylru.FunctionCacheManager(square, size)
print(cached(7))
print(cached(6))
print(cached(7))
# The results of cached are the same as square, but automatically cached
# to speed up future calls.
cached.size() # Returns the size of the cache
cached.size(3) # Changes the size of the cache. x MUST be greater than
# zero. Returns the new size x.
cached.clear() # Remove all items from the cache.
wsteps = numpy.array(
sorted(map(float, wkern_file['wkern/%s' % args.theta].keys())))
wstep = wsteps[1] - wsteps[0]
print("w kernels: %.2f - %.2f lambda (step %.2f lambda)" %
(min(wsteps), max(wsteps), wstep))
# Make a custom kernel cache that reads from the hdf5 file
def closest(xs, x):
return xs[numpy.argmin(numpy.abs(numpy.array(xs) - x))]
def w_kernel_fn(theta, w):
kernw = closest(wsteps, w)
#print("w=", kernw)
return wkern_file['wkern/%s/%s/kern' % (theta, kernw)]
w_cache = pylru.FunctionCacheManager(w_kernel_fn, len(wsteps))
# A-kernels?
if args.akern is None:
# Just pure w-projection, also no source dependency.
print("Gridder: W-projection")
grid_fn = w_cache_imaging
grid_pars = {'wstep': wstep, 'kernel_cache': w_cache}
src = numpy.zeros((src.shape[0], 0))
else:
# Open A-kernel file
akern_file = h5py.File(args.akern.name, "r", driver='core')
times = list(map(float, akern_file['akern/%s/0' % args.theta]))
