def DD(autocorr, nthreads, binfile, X1, Y1, Z1, weights1=None, periodic=True,
X2=None, Y2=None, Z2=None, weights2=None, verbose=False, boxsize=0.0,
output_ravg=False, xbin_refine_factor=2, ybin_refine_factor=2,
zbin_refine_factor=1, max_cells_per_dim=100,
c_api_timer=False, isa=r'fastest', weight_type=None):
"""
Calculate the 3-D pair-counts corresponding to the real-space correlation
function, :math:`\\xi(r)`.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
.. note:: This module only returns pair counts and not the actual
correlation function :math:`\\xi(r)`. See
:py:mod:`Corrfunc.utils.convert_3d_counts_to_cf` for computing
for computing :math:`\\xi(r)` from the pair counts returned.
Parameters
-----------
autocorr: boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
nthreads: integer
The number of OpenMP threads to use. Has no effect if OpenMP was not
enabled during library compilation.
binfile: string or an list/array of floats
For string input: filename specifying the ``r`` bins for
``DD``. The file should contain white-space separated values
of (rmin, rmax) for each ``r`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``r`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
X1/Y1/Z1: array_like, real (float/double)
The array of X/Y/Z positions for the first set of points.
Calculations are done in the precision of the supplied arrays.
weights1: array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or (n_positions,).
`weight_type` specifies how these weights are used; results are returned
in the `weightavg` field. If only one of weights1 and weights2 is
specified, the other will be set to uniform weights.
periodic: boolean
Boolean flag to indicate periodic boundary conditions.
X2/Y2/Z2: array-like, real (float/double)
Array of XYZ positions for the second set of points. *Must* be the same
precision as the X1/Y1/Z1 arrays. Only required when ``autocorr==0``.
weights2: array-like, real (float/double), optional
Same as weights1, but for the second set of positions
verbose: boolean (default false)
Boolean flag to control output of informational messages
boxsize: double
The side-length of the cube in the cosmological simulation.
Present to facilitate exact calculations for periodic wrapping.
If boxsize is not supplied, then the wrapping is done based on
the maximum difference within each dimension of the X/Y/Z arrays.
output_ravg: boolean (default false)
Boolean flag to output the average ``r`` for each bin. Code will
run slower if you set this flag.
Note: If you are calculating in single-precision, ``ravg`` will
suffer from numerical loss of precision and can not be trusted.
If you need accurate ``ravg`` values, then pass in double precision
arrays for the particle positions.
(xyz)bin_refine_factor: integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer: boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code will
revert to using ``fallback`` (even though ``sse42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type: string, optional
The type of weighting to apply. One of ["pair_product", None]. Default: None.
Returns
--------
results: Numpy structured array
A numpy structured array containing [rmin, rmax, ravg, npairs, weightavg]
for each radial bin specified in the ``binfile``. If ``output_ravg`` is
not set, then ``ravg`` will be set to 0.0 for all bins; similarly for
``weightavg``. ``npairs`` contains the number of pairs in that bin and can
be used to compute the actual :math:`\\xi(r)` by combining with (DR, RR) counts.
api_time: float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only the time
spent within the C library and ignores all python overhead.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.theory.DD import DD
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> N = 10000
>>> boxsize = 420.0
>>> nthreads = 4
>>> autocorr = 1
>>> seed = 42
>>> np.random.seed(seed)
>>> X = np.random.uniform(0, boxsize, N)
>>> Y = np.random.uniform(0, boxsize, N)
>>> Z = np.random.uniform(0, boxsize, N)
>>> weights = np.ones_like(X)
>>> results = DD(autocorr, nthreads, binfile, X, Y, Z, weights1=weights, weight_type='pair_product', output_ravg=True)
>>> for r in results: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10d} {4:10.6f}".
... format(r['rmin'], r['rmax'], r['ravg'],
... r['npairs'], r['weightavg'])) # doctest: +NORMALIZE_WHITESPACE
0.167536 0.238755 0.000000 0 0.000000
0.238755 0.340251 0.000000 0 0.000000
0.340251 0.484892 0.000000 0 0.000000
0.484892 0.691021 0.000000 0 0.000000
0.691021 0.984777 0.945372 2 1.000000
0.984777 1.403410 1.340525 10 1.000000
1.403410 2.000000 1.732968 36 1.000000
2.000000 2.850200 2.558878 54 1.000000
2.850200 4.061840 3.564959 208 1.000000
4.061840 5.788530 4.999278 674 1.000000
5.788530 8.249250 7.126673 2154 1.000000
8.249250 11.756000 10.201834 5996 1.000000
11.756000 16.753600 14.517830 17746 1.000000
16.753600 23.875500 20.716017 50252 1.000000
"""
try:
from Corrfunc._countpairs import countpairs as DD_extn
except ImportError:
msg = "Could not import the C extension for the 3-D "\
"real-space pair counter."
raise ImportError(msg)
import numpy as np
from warnings import warn
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins, convert_to_native_endian,\
is_native_endian
from future.utils import bytes_to_native_str
# Broadcast scalar weights to arrays
if weights1 is not None:
weights1 = np.atleast_1d(weights1)
if weights2 is not None:
weights2 = np.atleast_1d(weights2)
if not autocorr:
if X2 is None or Y2 is None or Z2 is None:
msg = "Must pass valid arrays for X2/Y2/Z2 for "\
"computing cross-correlation"
raise ValueError(msg)
# If only one set of points has weights, set the other to uniform weights
if weights1 is None and weights2 is not None:
weights1 = np.ones_like(weights2)
if weights2 is None and weights1 is not None:
weights2 = np.ones_like(weights1)
# Warn about non-native endian arrays
if not all(is_native_endian(arr) for arr in [X1, Y1, Z1, weights1, X2, Y2, Z2, weights2]):
warn('One or more input array has non-native endianness! A copy will be made with the correct endianness.')
X1, Y1, Z1, weights1, X2, Y2, Z2, weights2 = [convert_to_native_endian(arr) for arr in [X1, Y1, Z1, weights1, X2, Y2, Z2, weights2]]
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'X2', 'Y2', 'Z2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time = DD_extn(autocorr, nthreads, rbinfile,
X1, Y1, Z1,
periodic=periodic,
verbose=verbose,
boxsize=boxsize,
output_ravg=output_ravg,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa, **kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'rmin'), np.float),
(bytes_to_native_str(b'rmax'), np.float),
(bytes_to_native_str(b'ravg'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def wp(boxsize,
pimax,
nthreads,
binfile,
X,
Y,
Z,
weights=None,
weight_type=None,
verbose=False,
output_rpavg=False,
xbin_refine_factor=2,
ybin_refine_factor=2,
zbin_refine_factor=1,
max_cells_per_dim=100,
c_api_timer=False,
c_cell_timer=False,
isa='fastest'):
"""
Function to compute the projected correlation function in a
periodic cosmological box. Pairs which are separated by less
than the ``rp`` bins (specified in ``binfile``) in the
X-Y plane, and less than ``pimax`` in the Z-dimension are
counted.
If ``weights`` are provided, the resulting correlation function
is weighted. The weighting scheme depends on ``weight_type``.
.. note:: Pairs are double-counted. And if ``rpmin`` is set to
0.0, then all the self-pairs (i'th particle with itself) are
added to the first bin => minimum number of pairs in the first bin
is the total number of particles.
Parameters
-----------
boxsize: double
A double-precision value for the boxsize of the simulation
in same units as the particle positions and the ``rp`` bins.
pimax: double
A double-precision value for the maximum separation along
the Z-dimension.
.. note:: Only pairs with ``0 <= dz < pimax`` are counted (no equality).
nthreads: integer
Number of threads to use.
binfile: string or an list/array of floats
For string input: filename specifying the ``rp`` bins for
``DDrppi_mocks``. The file should contain white-space separated values
of (rpmin, rpmax) for each ``rp`` wanted. The bins do not need to be
contiguous but must be in increasing order (smallest bins come first).
For array-like input: A sequence of ``rp`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input, specifying 15 (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
X/Y/Z: arraytype, real (float/double)
Particle positions in the 3 axes. Must be within [0, boxsize]
and specified in the same units as ``rp_bins`` and boxsize. All
3 arrays must be of the same floating-point type.
Calculations will be done in the same precision as these arrays,
i.e., calculations will be in floating point if XYZ are single
precision arrays (C float type); or in double-precision if XYZ
are double precision arrays (C double type).
weights: array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or (n_positions,).
`weight_type` specifies how these weights are used; results are returned
in the `weightavg` field.
verbose: boolean (default false)
Boolean flag to control output of informational messages
output_rpavg: boolean (default false)
Boolean flag to output the average ``rp`` for each bin. Code will
run slower if you set this flag.
.. note:: If you are calculating in single-precision, ``rpavg`` will
suffer from numerical loss of precision and can not be trusted. If
you need accurate ``rpavg`` values, then pass in double precision
arrays for the particle positions.
(xyz)bin_refine_factor: integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer: boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
c_cell_timer : boolean (default false)
Boolean flag to measure actual time spent **per cell-pair** within the
C libraries. A very detailed timer that stores information about the
number of particles in each cell, the thread id that processed that
cell-pair and the amount of time in nano-seconds taken to process that
cell pair. This timer can be used to study the instruction set
efficiency, and load-balancing of the code.
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code will
revert to using ``fallback`` (even though ``sse42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type: string, optional
The type of weighting to apply. One of ["pair_product", None]. Default: None.
Returns
--------
results: Numpy structured array
A numpy structured array containing [rpmin, rpmax, rpavg, wp, npairs, weightavg]
for each radial specified in the ``binfile``. If ``output_rpavg`` is not
set then ``rpavg`` will be set to 0.0 for all bins; similarly for ``weightavg``.
``wp`` contains the projected correlation function while ``npairs`` contains the
number of unique pairs in that bin. If using weights, ``wp`` will be weighted
while ``npairs`` will not be.
api_time: float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only the time spent
within the C library and ignores all python overhead.
cell_time: list, optional
Only returned if ``c_cell_timer`` is set. Contains
detailed stats about each cell-pair visited during pair-counting,
viz., number of particles in each of the cells in the pair, 1-D
cell-indices for each cell in the pair, time (in nano-seconds) to
process the pair and the thread-id for the thread that processed that
cell-pair.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.theory.wp import wp
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> N = 10000
>>> boxsize = 420.0
>>> pimax = 40.0
>>> nthreads = 4
>>> seed = 42
>>> np.random.seed(seed)
>>> X = np.random.uniform(0, boxsize, N)
>>> Y = np.random.uniform(0, boxsize, N)
>>> Z = np.random.uniform(0, boxsize, N)
>>> results = wp(boxsize, pimax, nthreads, binfile, X, Y, Z, weights=np.ones_like(X), weight_type='pair_product')
>>> for r in results:
... print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10.6f} {4:10d} {5:10.6f}".
... format(r['rmin'], r['rmax'],
... r['rpavg'], r['wp'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
0.167536 0.238755 0.000000 66.717143 18 1.000000
0.238755 0.340251 0.000000 -15.786045 16 1.000000
0.340251 0.484892 0.000000 2.998470 42 1.000000
0.484892 0.691021 0.000000 -15.779885 66 1.000000
0.691021 0.984777 0.000000 -11.966728 142 1.000000
0.984777 1.403410 0.000000 -9.699906 298 1.000000
1.403410 2.000000 0.000000 -11.698771 588 1.000000
2.000000 2.850200 0.000000 3.848375 1466 1.000000
2.850200 4.061840 0.000000 -0.921452 2808 1.000000
4.061840 5.788530 0.000000 0.454851 5802 1.000000
5.788530 8.249250 0.000000 1.428344 11926 1.000000
8.249250 11.756000 0.000000 -1.067885 23478 1.000000
11.756000 16.753600 0.000000 -0.553319 47994 1.000000
16.753600 23.875500 0.000000 -0.086433 98042 1.000000
"""
try:
from Corrfunc._countpairs import countpairs_wp as wp_extn
except ImportError:
msg = "Could not import the C extension for the projected "\
"correlation function."
raise ImportError(msg)
import numpy as np
from warnings import warn
from future.utils import bytes_to_native_str
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins, convert_to_native_endian,\
is_native_endian
# Broadcast scalar weights to arrays
if weights is not None:
weights = np.atleast_1d(weights)
# Warn about non-native endian arrays
if not all(is_native_endian(arr) for arr in [X, Y, Z, weights]):
warn(
'One or more input array has non-native endianness! A copy will be made with the correct endianness.'
)
X, Y, Z, weights = [
convert_to_native_endian(arr) for arr in X, Y, Z, weights
]
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights', 'weight_type']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time, cell_time = wp_extn(
boxsize,
pimax,
nthreads,
rbinfile,
X,
Y,
Z,
verbose=verbose,
output_rpavg=output_rpavg,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
c_cell_timer=c_cell_timer,
isa=integer_isa,
**kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'rmin'), np.float),
(bytes_to_native_str(b'rmax'), np.float),
(bytes_to_native_str(b'rpavg'), np.float),
(bytes_to_native_str(b'wp'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
results = np.array(extn_results, dtype=results_dtype)
# A better solution for returning multiple values based on
# input parameter. Lifted straight from numpy.unique -- MS 10/26/2016
optional_returns = c_api_timer or c_cell_timer
if not optional_returns:
ret = results
else:
ret = (results, )
if c_api_timer:
ret += (api_time, )
if c_cell_timer:
# Convert to numpy structured array
np_cell_time = _convert_cell_timer(cell_time)
ret += (np_cell_time, )
return ret
def DDsmu(autocorr, nthreads, binfile, mu_max, nmu_bins, X1, Y1, Z1, weights1=None,
periodic=True, X2=None, Y2=None, Z2=None, weights2=None,
verbose=False, boxsize=0.0, output_savg=False,
xbin_refine_factor=2, ybin_refine_factor=2,
zbin_refine_factor=1, max_cells_per_dim=100,
c_api_timer=False, isa=r'fastest', weight_type=None):
"""
Calculate the 2-D pair-counts corresponding to the redshift-space
correlation function, :math:`\\xi(s, \mu)` Pairs which are separated
by less than the ``s`` bins (specified in ``binfile``) in 3-D, and
less than ``s*mu_max`` in the Z-dimension are counted.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
.. note:: This module only returns pair counts and not the actual
correlation function :math:`\\xi(s, \mu)`. See the
utilities :py:mod:`Corrfunc.utils.convert_3d_counts_to_cf`
for computing :math:`\\xi(s, \mu)` from the pair counts.
.. versionadded:: 2.1.0
Parameters
----------
autocorr: boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
nthreads: integer
The number of OpenMP threads to use. Has no effect if OpenMP was not
enabled during library compilation.
binfile: string or an list/array of floats
For string input: filename specifying the ``s`` bins for
``DDsmu_mocks``. The file should contain white-space separated values
of (smin, smax) specifying each ``s`` bin wanted. The bins
need to be contiguous and sorted in increasing order (smallest bins
come first).
For array-like input: A sequence of ``s`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
mu_max: double. Must be in range (0.0, 1.0]
A double-precision value for the maximum cosine of the angular
separation from the line of sight (LOS). Here, LOS is taken to be
along the Z direction.
Note: Only pairs with :math:`0 <= \cos(\\theta_{LOS}) < \mu_{max}`
are counted (no equality).
nmu_bins: int
The number of linear ``mu`` bins, with the bins ranging from
from (0, :math:`\mu_{max}`)
X1/Y1/Z1 : array-like, real (float/double)
The array of X/Y/Z positions for the first set of points.
Calculations are done in the precision of the supplied arrays.
weights1 : array-like, real (float/double), shape (n_particles,) or \
(n_weights_per_particle,n_particles), optional
Weights for computing a weighted pair count.
weight_type : str, optional
The type of pair weighting to apply.
Options: "pair_product", None; Default: None.
periodic : boolean
Boolean flag to indicate periodic boundary conditions.
X2/Y2/Z2 : array-like, real (float/double)
Array of XYZ positions for the second set of points. *Must* be the same
precision as the X1/Y1/Z1 arrays. Only required when ``autocorr==0``.
weights2 : array-like, real (float/double), shape (n_particles,) or \
(n_weights_per_particle,n_particles), optional
Weights for computing a weighted pair count.
verbose : boolean (default false)
Boolean flag to control output of informational messages
boxsize : double
The side-length of the cube in the cosmological simulation.
Present to facilitate exact calculations for periodic wrapping.
If boxsize is not supplied, then the wrapping is done based on
the maximum difference within each dimension of the X/Y/Z arrays.
output_savg : boolean (default false)
Boolean flag to output the average ``s`` for each bin. Code will
run slower if you set this flag. Also, note, if you are calculating
in single-precision, ``s`` will suffer from numerical loss of
precision and can not be trusted. If you need accurate ``s``
values, then pass in double precision arrays for the particle positions.
(xyz)bin_refine_factor: integer (default (2,2,1) typical values in [1-3])
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer (default 100, typical values in [50-300])
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer : boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa : integer (default -1)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [-1, AVX, SSE42, FALLBACK]
Setting isa to -1 will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``AVX`` and ``AVX`` is not available on the computer, then the code will
revert to using ``FALLBACK`` (even though ``SSE42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are* benchmarking,
then the integer values correspond to the ``enum`` for the instruction set
defined in ``utils/defs.h``.
Returns
--------
results : A python list
A python list containing ``nmu_bins`` of [smin, smax, savg, mu_max, npairs, weightavg]
for each spatial bin specified in the ``binfile``. There will be a total of ``nmu_bins``
ranging from [0, ``mu_max``) *per* spatial bin. If ``output_savg`` is not set, then ``savg``
will be set to 0.0 for all bins; similarly for ``weight_avg``. ``npairs``
contains the number of pairs in that bin.
time : if ``c_api_timer`` is set, then the return value contains the time spent
in the API; otherwise time is set to 0.0
Example
-------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.theory.DDsmu import DDsmu
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> N = 10000
>>> boxsize = 420.0
>>> nthreads = 4
>>> autocorr = 1
>>> mu_max = 1.0
>>> seed = 42
>>> nmu_bins = 10
>>> np.random.seed(seed)
>>> X = np.random.uniform(0, boxsize, N)
>>> Y = np.random.uniform(0, boxsize, N)
>>> Z = np.random.uniform(0, boxsize, N)
>>> weights = np.ones_like(X)
>>> results = DDsmu(autocorr, nthreads, binfile, mu_max, nmu_bins,
... X, Y, Z, weights1=weights, weight_type='pair_product', output_savg=True)
>>> for r in results[100:]: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10.1f}"
... " {4:10d} {5:10.6f}".format(r['smin'], r['smax'],
... r['savg'], r['mu_max'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
5.788530 8.249250 7.148213 0.1 230 1.000000
5.788530 8.249250 7.157218 0.2 236 1.000000
5.788530 8.249250 7.165338 0.3 208 1.000000
5.788530 8.249250 7.079905 0.4 252 1.000000
5.788530 8.249250 7.251661 0.5 184 1.000000
5.788530 8.249250 7.118536 0.6 222 1.000000
5.788530 8.249250 7.083466 0.7 238 1.000000
5.788530 8.249250 7.198184 0.8 170 1.000000
5.788530 8.249250 7.127409 0.9 208 1.000000
5.788530 8.249250 6.973090 1.0 206 1.000000
8.249250 11.756000 10.149183 0.1 592 1.000000
8.249250 11.756000 10.213009 0.2 634 1.000000
8.249250 11.756000 10.192220 0.3 532 1.000000
8.249250 11.756000 10.246931 0.4 544 1.000000
8.249250 11.756000 10.102675 0.5 530 1.000000
8.249250 11.756000 10.276180 0.6 644 1.000000
8.249250 11.756000 10.251264 0.7 666 1.000000
8.249250 11.756000 10.138399 0.8 680 1.000000
8.249250 11.756000 10.191916 0.9 566 1.000000
8.249250 11.756000 10.243229 1.0 608 1.000000
11.756000 16.753600 14.552776 0.1 1734 1.000000
11.756000 16.753600 14.579991 0.2 1806 1.000000
11.756000 16.753600 14.599611 0.3 1802 1.000000
11.756000 16.753600 14.471100 0.4 1820 1.000000
11.756000 16.753600 14.480192 0.5 1740 1.000000
11.756000 16.753600 14.493679 0.6 1746 1.000000
11.756000 16.753600 14.547713 0.7 1722 1.000000
11.756000 16.753600 14.465390 0.8 1750 1.000000
11.756000 16.753600 14.547465 0.9 1798 1.000000
11.756000 16.753600 14.440975 1.0 1828 1.000000
16.753600 23.875500 20.720406 0.1 5094 1.000000
16.753600 23.875500 20.735403 0.2 5004 1.000000
16.753600 23.875500 20.721069 0.3 5172 1.000000
16.753600 23.875500 20.723648 0.4 5014 1.000000
16.753600 23.875500 20.650621 0.5 5094 1.000000
16.753600 23.875500 20.688135 0.6 5076 1.000000
16.753600 23.875500 20.735691 0.7 4910 1.000000
16.753600 23.875500 20.714097 0.8 4864 1.000000
16.753600 23.875500 20.751836 0.9 4954 1.000000
16.753600 23.875500 20.721183 1.0 5070 1.000000
"""
try:
from Corrfunc._countpairs import countpairs_s_mu as DDsmu_extn
except ImportError:
msg = "Could not import the C extension for the 3-D "\
"redshift-space pair counter."
raise ImportError(msg)
import numpy as np
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins
from future.utils import bytes_to_native_str
# Broadcast scalar weights to arrays
if weights1 is not None:
weights1 = np.atleast_1d(weights1)
if weights2 is not None:
weights2 = np.atleast_1d(weights2)
# Check if mu_max is scalar
if not np.isscalar(mu_max):
msg = "The parameter `mu_max` = {0}, has size = {1}. "\
"The code is expecting a scalar quantity (and not "\
"not a list, array)".format(mu_max, np.size(mu_max))
raise TypeError(msg)
# Check that mu_max is within (0.0, 1.0]
if mu_max <= 0.0 or mu_max > 1.0:
msg = "The parameter `mu_max` = {0}, is the max. of cosine of an "
"angle and should be within (0.0, 1.0]".format(mu_max)
raise ValueError(msg)
if not autocorr:
if X2 is None or Y2 is None or Z2 is None:
msg = "Must pass valid arrays for X2/Y2/Z2 for "\
"computing cross-correlation"
raise ValueError(msg)
# If only one set of points has weights, set the other to uniform weights
if weights1 is None and weights2 is not None:
weights1 = np.ones_like(weights2)
if weights2 is None and weights1 is not None:
weights2 = np.ones_like(weights1)
else:
X2 = np.empty(1)
Y2 = np.empty(1)
Z2 = np.empty(1)
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'X2', 'Y2', 'Z2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
sbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time = DDsmu_extn(autocorr, nthreads,
sbinfile,
mu_max, nmu_bins,
X1, Y1, Z1,
periodic=periodic,
verbose=verbose,
boxsize=boxsize,
output_savg=output_savg,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa, **kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(sbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'smin'), np.float),
(bytes_to_native_str(b'smax'), np.float),
(bytes_to_native_str(b'savg'), np.float),
(bytes_to_native_str(b'mu_max'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float),])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def DDrppi_mocks(autocorr,
cosmology,
nthreads,
pimax,
binfile,
RA1,
DEC1,
CZ1,
weights1=None,
RA2=None,
DEC2=None,
CZ2=None,
weights2=None,
is_comoving_dist=False,
verbose=False,
output_rpavg=False,
fast_divide=False,
xbin_refine_factor=2,
ybin_refine_factor=2,
zbin_refine_factor=1,
max_cells_per_dim=100,
c_api_timer=False,
isa=r'fastest',
weight_type=None):
"""
Calculate the 2-D pair-counts corresponding to the projected correlation
function, :math:`\\xi(r_p, \pi)`. Pairs which are separated by less
than the ``rp`` bins (specified in ``binfile``) in the
X-Y plane, and less than ``pimax`` in the Z-dimension are
counted. The input positions are expected to be on-sky co-ordinates.
This module is suitable for calculating correlation functions for mock
catalogs.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
Returns a numpy structured array containing the pair counts for the
specified bins.
.. note:: that this module only returns pair counts and not the actual
correlation function :math:`\\xi(r_p, \pi)` or :math:`wp(r_p)`. See the
utilities :py:mod:`Corrfunc.utils.convert_3d_counts_to_cf` and
:py:mod:`Corrfunc.utils.convert_rp_pi_counts_to_wp` for computing
:math:`\\xi(r_p, \pi)` and :math:`wp(r_p)` respectively from the
pair counts.
Parameters
-----------
autocorr : boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
cosmology : integer, required
Integer choice for setting cosmology. Valid values are 1->LasDamas
cosmology and 2->Planck cosmology. If you need arbitrary cosmology,
easiest way is to convert the ``CZ`` values into co-moving distance,
based on your preferred cosmology. Set ``is_comoving_dist=True``, to
indicate that the co-moving distance conversion has already been done.
Choices:
1. LasDamas cosmology. :math:`\\Omega_m=0.25`, :math:`\\Omega_\Lambda=0.75`
2. Planck cosmology. :math:`\\Omega_m=0.302`, :math:`\\Omega_\Lambda=0.698`
To setup a new cosmology, add an entry to the function,
``init_cosmology`` in ``ROOT/utils/cosmology_params.c`` and re-install
the entire package.
nthreads : integer
The number of OpenMP threads to use. Has no effect if OpenMP was not
enabled during library compilation.
pimax : double
A double-precision value for the maximum separation along
the Z-dimension.
Distances along the :math:`\\pi` direction are binned with unit
depth. For instance, if ``pimax=40``, then 40 bins will be created
along the ``pi`` direction. Only pairs with ``0 <= dz < pimax``
are counted (no equality).
binfile: string or an list/array of floats
For string input: filename specifying the ``rp`` bins for
``DDrppi_mocks``. The file should contain white-space separated values
of (rpmin, rpmax) for each ``rp`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``rp`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
RA1 : array-like, real (float/double)
The array of Right Ascensions for the first set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Calculations are done in the precision of the supplied arrays.
DEC1 : array-like, real (float/double)
Array of Declinations for the first set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1.
CZ1 : array-like, real (float/double)
Array of (Speed Of Light * Redshift) values for the first set of
points. Code will try to detect cases where ``redshifts`` have been
passed and multiply the entire array with the ``speed of light``.
If is_comoving_dist is set, then ``CZ1`` is interpreted as the
co-moving distance, rather than `cz`.
weights1 : array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or (n_positions,).
`weight_type` specifies how these weights are used; results are returned
in the `weightavg` field. If only one of weights1 and weights2 is
specified, the other will be set to uniform weights.
RA2 : array-like, real (float/double)
The array of Right Ascensions for the second set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Must be of same precision type as RA1/DEC1/CZ1.
DEC2 : array-like, real (float/double)
Array of Declinations for the second set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1/DEC1/CZ1.
CZ2 : array-like, real (float/double)
Array of (Speed Of Light * Redshift) values for the second set of
points. Code will try to detect cases where ``redshifts`` have been
passed and multiply the entire array with the ``speed of light``.
If is_comoving_dist is set, then ``CZ2`` is interpreted as the
co-moving distance, rather than `cz`.
Must be of same precision type as RA1/DEC1/CZ1.
weights2 : array-like, real (float/double), optional
Same as weights1, but for the second set of positions
is_comoving_dist : boolean (default false)
Boolean flag to indicate that ``cz`` values have already been
converted into co-moving distances. This flag allows arbitrary
cosmologies to be used in ``Corrfunc``.
verbose : boolean (default false)
Boolean flag to control output of informational messages
output_rpavg : boolean (default false)
Boolean flag to output the average ``rp`` for each bin. Code will
run slower if you set this flag.
If you are calculating in single-precision, ``rpavg`` will suffer
suffer from numerical loss of precision and can not be trusted. If
you need accurate ``rpavg`` values, then pass in double precision
arrays for the particle positions.
fast_divide : boolean (default false)
Boolean flag to replace the division in ``AVX`` implementation with an
approximate reciprocal, followed by two Newton-Raphson steps. Improves
runtime by ~15-20%. Loss of precision is at the 5-6th decimal place.
(xyz)bin_refine_factor : integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer : boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa : string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code
will revert to using ``fallback`` (even though ``sse42`` might be
available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type : string, optional
The type of weighting to apply. One of ["pair_product", None]. Default: None.
Returns
--------
results : Numpy structured array
A numpy structured array containing [rpmin, rpmax, rpavg, pimax, npairs, weightavg]
for each radial bin specified in the ``binfile``. If ``output_ravg`` is
not set, then ``rpavg`` will be set to 0.0 for all bins; similarly for
``weightavg``. ``npairs``
contains the number of pairs in that bin and can be used to compute the
actual :math:`\\xi(r_p, \pi)` or :math:`wp(rp)` by combining with
(DR, RR) counts.
api_time : float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only the time
spent within the C library and ignores all python overhead.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.mocks.DDrppi_mocks import DDrppi_mocks
>>> import math
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../mocks/tests/", "bins")
>>> N = 100000
>>> boxsize = 420.0
>>> seed = 42
>>> np.random.seed(seed)
>>> X = np.random.uniform(-0.5*boxsize, 0.5*boxsize, N)
>>> Y = np.random.uniform(-0.5*boxsize, 0.5*boxsize, N)
>>> Z = np.random.uniform(-0.5*boxsize, 0.5*boxsize, N)
>>> weights = np.ones_like(X)
>>> CZ = np.sqrt(X*X + Y*Y + Z*Z)
>>> inv_cz = 1.0/CZ
>>> X *= inv_cz
>>> Y *= inv_cz
>>> Z *= inv_cz
>>> DEC = 90.0 - np.arccos(Z)*180.0/math.pi
>>> RA = (np.arctan2(Y, X)*180.0/math.pi) + 180.0
>>> autocorr = 1
>>> cosmology = 1
>>> nthreads = 2
>>> pimax = 40.0
>>> results = DDrppi_mocks(autocorr, cosmology, nthreads,
... pimax, binfile, RA, DEC, CZ,
... weights1=weights, weight_type='pair_product',
... output_rpavg=True, is_comoving_dist=True)
>>> for r in results[519:]: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10.1f}"
... " {4:10d} {5:10.6f}".format(r['rmin'], r['rmax'],
... r['rpavg'], r['pimax'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
11.359969 16.852277 14.285169 40.0 104850 1.000000
16.852277 25.000000 21.181246 1.0 274144 1.000000
16.852277 25.000000 21.190844 2.0 272876 1.000000
16.852277 25.000000 21.183321 3.0 272294 1.000000
16.852277 25.000000 21.188486 4.0 272506 1.000000
16.852277 25.000000 21.170832 5.0 272100 1.000000
16.852277 25.000000 21.165379 6.0 271788 1.000000
16.852277 25.000000 21.175246 7.0 270040 1.000000
16.852277 25.000000 21.187417 8.0 269492 1.000000
16.852277 25.000000 21.172066 9.0 269682 1.000000
16.852277 25.000000 21.182460 10.0 268266 1.000000
16.852277 25.000000 21.170594 11.0 268744 1.000000
16.852277 25.000000 21.178608 12.0 266820 1.000000
16.852277 25.000000 21.187184 13.0 266510 1.000000
16.852277 25.000000 21.184937 14.0 265484 1.000000
16.852277 25.000000 21.180184 15.0 265258 1.000000
16.852277 25.000000 21.191504 16.0 262952 1.000000
16.852277 25.000000 21.187746 17.0 262602 1.000000
16.852277 25.000000 21.189778 18.0 260206 1.000000
16.852277 25.000000 21.188882 19.0 259410 1.000000
16.852277 25.000000 21.185684 20.0 256806 1.000000
16.852277 25.000000 21.194036 21.0 255574 1.000000
16.852277 25.000000 21.184115 22.0 255406 1.000000
16.852277 25.000000 21.178255 23.0 252394 1.000000
16.852277 25.000000 21.184644 24.0 252220 1.000000
16.852277 25.000000 21.187020 25.0 251668 1.000000
16.852277 25.000000 21.183827 26.0 249648 1.000000
16.852277 25.000000 21.183121 27.0 247160 1.000000
16.852277 25.000000 21.180872 28.0 246238 1.000000
16.852277 25.000000 21.185251 29.0 246030 1.000000
16.852277 25.000000 21.183488 30.0 242124 1.000000
16.852277 25.000000 21.194538 31.0 242426 1.000000
16.852277 25.000000 21.190702 32.0 239778 1.000000
16.852277 25.000000 21.188985 33.0 239046 1.000000
16.852277 25.000000 21.187092 34.0 237640 1.000000
16.852277 25.000000 21.185515 35.0 236256 1.000000
16.852277 25.000000 21.190278 36.0 233536 1.000000
16.852277 25.000000 21.183240 37.0 233274 1.000000
16.852277 25.000000 21.183796 38.0 231628 1.000000
16.852277 25.000000 21.200668 39.0 230378 1.000000
16.852277 25.000000 21.181153 40.0 229006 1.000000
"""
try:
from Corrfunc._countpairs_mocks import countpairs_rp_pi_mocks as\
DDrppi_extn
except ImportError:
msg = "Could not import the C extension for the on-sky"\
"pair counter."
raise ImportError(msg)
import numpy as np
from warnings import warn
from Corrfunc.utils import translate_isa_string_to_enum, fix_ra_dec,\
return_file_with_rbins, convert_to_native_endian,\
is_native_endian
from future.utils import bytes_to_native_str
# Broadcast scalar weights to arrays
if weights1 is not None:
weights1 = np.atleast_1d(weights1)
if weights2 is not None:
weights2 = np.atleast_1d(weights2)
if not autocorr:
if RA2 is None or DEC2 is None or CZ2 is None:
msg = "Must pass valid arrays for RA2/DEC2/CZ2 for "\
"computing cross-correlation"
raise ValueError(msg)
# If only one set of points has weights, set the other to uniform weights
if weights1 is None and weights2 is not None:
weights1 = np.ones_like(weights2)
if weights2 is None and weights1 is not None:
weights2 = np.ones_like(weights1)
else:
RA2 = np.empty(1)
DEC2 = np.empty(1)
CZ2 = np.empty(1)
# Warn about non-native endian arrays
if not all(
is_native_endian(arr)
for arr in [RA1, DEC1, CZ1, weights1, RA2, DEC2, CZ2, weights2]):
warn(
'One or more input array has non-native endianness! A copy will be made with the correct endianness.'
)
RA1, DEC1, CZ1, weights1, RA2, DEC2, CZ2, weights2 = [
convert_to_native_endian(arr)
for arr in [RA1, DEC1, CZ1, weights1, RA2, DEC2, CZ2, weights2]
]
fix_ra_dec(RA1, DEC1)
if autocorr == 0:
fix_ra_dec(RA2, DEC2)
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'RA2', 'DEC2', 'CZ2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time = DDrppi_extn(autocorr,
cosmology,
nthreads,
pimax,
rbinfile,
RA1,
DEC1,
CZ1,
is_comoving_dist=is_comoving_dist,
verbose=verbose,
output_rpavg=output_rpavg,
fast_divide=fast_divide,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa,
**kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'rmin'), np.float),
(bytes_to_native_str(b'rmax'), np.float),
(bytes_to_native_str(b'rpavg'), np.float),
(bytes_to_native_str(b'pimax'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def find_fastest_wp_bin_refs(boxsize,
pimax,
nthreads,
binfile,
X,
Y,
Z,
verbose=False,
output_rpavg=False,
max_cells_per_dim=100,
isa=r'fastest',
maxbinref=3,
nrepeats=3,
return_runtimes=False):
"""
Finds the combination of ``bin refine factors`` that produces the
fastest computation for the given dataset and ``rp`` limits.
Parameters
-----------
boxsize: double
A double-precision value for the boxsize of the simulation
in same units as the particle positions and the ``rp`` bins.
pimax: double
A double-precision value for the maximum separation along
the Z-dimension.
.. note:: Only pairs with ``0 <= dz < pimax`` are counted (no equality).
nthreads: integer
Number of threads to use.
binfile: string or an list/array of floats
For string input: filename specifying the ``rp`` bins for
``DDrppi_mocks``. The file should contain white-space separated values
of (rpmin, rpmax) for each ``rp`` wanted. The bins do not need to be
contiguous but must be in increasing order (smallest bins come first).
For array-like input: A sequence of ``rp`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input, specifying 15 (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
X/Y/Z: arraytype, real (float/double)
Particle positions in the 3 axes. Must be within [0, boxsize]
and specified in the same units as ``rp_bins`` and boxsize. All
3 arrays must be of the same floating-point type.
Calculations will be done in the same precision as these arrays,
i.e., calculations will be in floating point if XYZ are single
precision arrays (C float type); or in double-precision if XYZ
are double precision arrays (C double type).
verbose: boolean (default false)
Boolean flag to control output of informational messages
output_rpavg: boolean (default false)
Boolean flag to output the average ``rp`` for each bin. Code will
run slower if you set this flag.
.. note:: If you are calculating in single-precision, ``rpavg`` will
suffer from numerical loss of precision and can not be trusted. If
you need accurate ``rpavg`` values, then pass in double precision
arrays for the particle positions.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code will
revert to using ``fallback`` (even though ``sse42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
maxbinref: integer (default 3)
The maximum ``bin refine factor`` to use along each dimension. From
experience, values larger than 3 do not improve ``wp`` runtime.
Runtime of module scales as ``maxbinref^3``, so change the value of
``maxbinref`` with caution.
nrepeats: integer (default 3)
Number of times to repeat the timing for an individual run. Accounts
for the dispersion in runtimes on computers with multiple user
processes.
return_runtimes: boolean (default ``false``)
If set, also returns the array of runtimes.
Returns
--------
(nx, ny, nz) : tuple of integers
The combination of ``bin refine factors`` along each dimension that
produces the fastest code.
runtimes: numpy structured array
if ``return_runtimes`` is set, then the return value is a tuple
containing ((nx, ny, nz), runtimes). ``runtimes`` is a ``numpy``
structured array containing the fields, [``nx``, ``ny``, ``nz``,
``avg_runtime``, ``sigma_time``]. Here, ``avg_runtime`` is the
average time, measured over ``nrepeats`` invocations, spent in
the python extension. ``sigma_time`` is the dispersion of the
run times across those ``nrepeats`` invocations.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.io import read_catalog
>>> from Corrfunc.theory.wp import find_fastest_wp_bin_refs
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> X, Y, Z = read_catalog(return_dtype=np.float32)
>>> boxsize = 420.0
>>> pimax = 40.0
>>> nthreads = 4
>>> verbose = 1
>>> best, _ = find_fastest_wp_bin_refs(boxsize, pimax, nthreads, binfile,
... X, Y, Z, maxbinref=2, nrepeats=3,
... verbose=verbose,
... return_runtimes=True)
>>> print(best) # doctest:+SKIP
(2, 2, 1)
.. note:: Since the result might change depending on the computer, doctest
is skipped for this function.
"""
try:
from Corrfunc._countpairs import countpairs_wp as wp_extn
except ImportError:
msg = "Could not import the C extension for the projected "\
"correlation function."
raise ImportError(msg)
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins
import itertools
import numpy as np
from future.utils import bytes_to_native_str
import time
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
bin_refs = np.arange(1, maxbinref + 1)
bin_ref_perms = itertools.product(bin_refs, bin_refs, bin_refs)
dtype = np.dtype([(bytes_to_native_str(b'nx'), np.int),
(bytes_to_native_str(b'ny'), np.int),
(bytes_to_native_str(b'nz'), np.int),
(bytes_to_native_str(b'avg_time'), np.float),
(bytes_to_native_str(b'sigma_time'), np.float)])
all_runtimes = np.zeros(maxbinref**3, dtype=dtype)
all_runtimes[:] = np.inf
for ii, (nx, ny, nz) in enumerate(bin_ref_perms):
total_runtime = 0.0
total_sqr_runtime = 0.0
for _ in range(nrepeats):
t0 = time.time()
extn_results, _, _ = wp_extn(boxsize,
pimax,
nthreads,
rbinfile,
X,
Y,
Z,
verbose=verbose,
output_rpavg=output_rpavg,
xbin_refine_factor=nx,
ybin_refine_factor=ny,
zbin_refine_factor=nz,
max_cells_per_dim=max_cells_per_dim,
isa=integer_isa)
t1 = time.time()
if extn_results is None:
msg = "RuntimeError occurred with perms = ({0}, {1}, {2})".\
format(nx, ny, nz)
print(msg)
print("Continuing...")
continue
dt = (t1 - t0)
total_runtime += dt
total_sqr_runtime += dt * dt
avg_runtime = total_runtime / nrepeats
# variance = E(X^2) - E^2(X)
# disp = sqrt(variance)
runtime_disp = np.sqrt(total_sqr_runtime / nrepeats -
avg_runtime * avg_runtime)
all_runtimes[ii]['nx'] = nx
all_runtimes[ii]['ny'] = ny
all_runtimes[ii]['nz'] = nz
all_runtimes[ii]['avg_time'] = avg_runtime
all_runtimes[ii]['sigma_time'] = runtime_disp
if delete_after_use:
import os
os.remove(rbinfile)
all_runtimes.sort(order=('avg_time', 'sigma_time'))
results = (all_runtimes[0]['nx'], all_runtimes[0]['ny'],
all_runtimes[0]['nz'])
optional_returns = return_runtimes
if not optional_returns:
ret = results
else:
ret = (results, )
if return_runtimes:
ret += (all_runtimes, )
return ret
def DDsmu_mocks(autocorr,
cosmology,
nthreads,
mu_max,
nmu_bins,
binfile,
RA1,
DEC1,
CZ1,
weights1=None,
RA2=None,
DEC2=None,
CZ2=None,
weights2=None,
is_comoving_dist=False,
verbose=False,
output_savg=False,
fast_divide_and_NR_steps=0,
xbin_refine_factor=2,
ybin_refine_factor=2,
zbin_refine_factor=1,
max_cells_per_dim=100,
c_api_timer=False,
isa='fastest',
weight_type=None):
"""
Calculate the 2-D pair-counts corresponding to the projected correlation
function, :math:`\\xi(s, \mu)`. The pairs are counted in bins of
radial separation and cosine of angle to the line-of-sight (LOS). The
input positions are expected to be on-sky co-ordinates. This module is
suitable for calculating correlation functions for mock catalogs.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
Returns a numpy structured array containing the pair counts for the
specified bins.
.. note:: This module only returns pair counts and not the actual
correlation function :math:`\\xi(s, \mu)`. See the
utilities :py:mod:`Corrfunc.utils.convert_3d_counts_to_cf`
for computing :math:`\\xi(s, \mu)` from the pair counts.
.. versionadded:: 2.1.0
Parameters
----------
autocorr: boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
cosmology: integer, required
Integer choice for setting cosmology. Valid values are 1->LasDamas
cosmology and 2->Planck cosmology. If you need arbitrary cosmology,
easiest way is to convert the ``CZ`` values into co-moving distance,
based on your preferred cosmology. Set ``is_comoving_dist=True``, to
indicate that the co-moving distance conversion has already been done.
Choices:
1. LasDamas cosmology. :math:`\\Omega_m=0.25`, :math:`\\Omega_\Lambda=0.75`
2. Planck cosmology. :math:`\\Omega_m=0.302`, :math:`\\Omega_\Lambda=0.698`
To setup a new cosmology, add an entry to the function,
``init_cosmology`` in ``ROOT/utils/cosmology_params.c`` and re-install
the entire package.
nthreads: integer
The number of OpenMP threads to use. Has no effect if OpenMP was not
enabled during library compilation.
mu_max: double. Must be in range [0.0, 1.0]
A double-precision value for the maximum cosine of the angular
separation from the line of sight (LOS). Here, ``mu`` is defined as
the angle between ``s`` and ``l``. If :math:`v_1` and :math:`v_2`
represent the vectors to each point constituting the pair, then
:math:`s := v_1 - v_2` and :math:`l := 1/2 (v_1 + v_2)`.
Note: Only pairs with :math:`0 <= \cos(\\theta_{LOS}) < \mu_{max}`
are counted (no equality).
nmu_bins: int
The number of linear ``mu`` bins, with the bins ranging from
from (0, :math:`\mu_{max}`)
binfile: string or an list/array of floats
For string input: filename specifying the ``s`` bins for
``DDsmu_mocks``. The file should contain white-space separated values
of (smin, smax) specifying each ``s`` bin wanted. The bins
need to be contiguous and sorted in increasing order (smallest bins
come first).
For array-like input: A sequence of ``s`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
RA1: array-like, real (float/double)
The array of Right Ascensions for the first set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Calculations are done in the precision of the supplied arrays.
DEC1: array-like, real (float/double)
Array of Declinations for the first set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1.
CZ1: array-like, real (float/double)
Array of (Speed Of Light * Redshift) values for the first set of
points. Code will try to detect cases where ``redshifts`` have been
passed and multiply the entire array with the ``speed of light``.
If is_comoving_dist is set, then ``CZ1`` is interpreted as the
co-moving distance, rather than `cz`.
weights1: array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions)
or (n_positions,). `weight_type` specifies how these weights are used;
results are returned in the `weightavg` field. If only one of
``weights1`` or ``weights2`` is specified, the other will be set
to uniform weights.
RA2: array-like, real (float/double)
The array of Right Ascensions for the second set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Must be of same precision type as RA1/DEC1/CZ1.
DEC2: array-like, real (float/double)
Array of Declinations for the second set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1/DEC1/CZ1.
CZ2: array-like, real (float/double)
Array of (Speed Of Light * Redshift) values for the second set of
points. Code will try to detect cases where ``redshifts`` have been
passed and multiply the entire array with the ``speed of light``.
If is_comoving_dist is set, then ``CZ2`` is interpreted as the
co-moving distance, rather than `cz`.
Must be of same precision type as RA1/DEC1/CZ1.
weights2: array-like, real (float/double), optional
Same as weights1, but for the second set of positions
is_comoving_dist: boolean (default false)
Boolean flag to indicate that ``cz`` values have already been
converted into co-moving distances. This flag allows arbitrary
cosmologies to be used in ``Corrfunc``.
verbose: boolean (default false)
Boolean flag to control output of informational messages
output_savg: boolean (default false)
Boolean flag to output the average ``s`` for each bin. Code will
run slower if you set this flag. Also, note, if you are calculating
in single-precision, ``savg`` will suffer from numerical loss of
precision and can not be trusted. If you need accurate ``savg``
values, then pass in double precision arrays for the particle
positions.
fast_divide_and_NR_steps: integer (default 0)
Replaces the division in ``AVX`` implementation with an approximate
reciprocal, followed by ``fast_divide_and_NR_steps`` of Newton-Raphson.
Can improve runtime by ~15-20% on older computers. Value of 0 uses
the standard division operation.
(xyz)bin_refine_factor: integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer: boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code
will revert to using ``fallback`` (even though ``sse42`` might be
available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type: string, optional
The type of weighting to apply. One of ["pair_product", None]. Default: None.
Returns
--------
results: Numpy structured array
A numpy structured array containing [smin, smax, savg, mumax, npairs, weightavg]
for each separation bin specified in the ``binfile``. If ``output_savg`` is
not set, then ``savg`` will be set to 0.0 for all bins; similarly for
``weightavg``. ``npairs`` contains the number of pairs in that bin and
can be used to compute the actual :math:`\\xi(s, \mu)` by combining
with (DR, RR) counts.
api_time: float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only
the time spent within the C library and ignores all python overhead.
"""
try:
from Corrfunc._countpairs_mocks import countpairs_s_mu_mocks as\
DDsmu_extn
except ImportError:
msg = "Could not import the C extension for the on-sky"\
"pair counter."
raise ImportError(msg)
import numpy as np
from Corrfunc.utils import translate_isa_string_to_enum, fix_ra_dec,\
return_file_with_rbins, sys_pipes
from future.utils import bytes_to_native_str
# Broadcast scalar weights to arrays
if weights1 is not None:
weights1 = np.atleast_1d(weights1)
if weights2 is not None:
weights2 = np.atleast_1d(weights2)
# Check if mu_max is scalar
if not np.isscalar(mu_max):
msg = "The parameter `mu_max` = {0}, has size = {1}. "\
"The code is expecting a scalar quantity (and not "\
"not a list, array)".format(mu_max, np.size(mu_max))
raise TypeError(msg)
# Check that mu_max is within (0.0, 1.0]
if mu_max <= 0.0 or mu_max > 1.0:
msg = "The parameter `mu_max` = {0}, is the max. of cosine of an "
"angle and should be within (0.0, 1.0]".format(mu_max)
raise ValueError(msg)
if not autocorr:
if RA2 is None or DEC2 is None or CZ2 is None:
msg = "Must pass valid arrays for RA2/DEC2/CZ2 for "\
"computing cross-correlation"
raise ValueError(msg)
# If only one set of points has weights, set the other to uniform weights
if weights1 is None and weights2 is not None:
weights1 = np.ones_like(weights2)
if weights2 is None and weights1 is not None:
weights2 = np.ones_like(weights1)
else:
RA2 = np.empty(1)
DEC2 = np.empty(1)
CZ2 = np.empty(1)
fix_ra_dec(RA1, DEC1)
if autocorr == 0:
fix_ra_dec(RA2, DEC2)
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'RA2', 'DEC2', 'CZ2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
sbinfile, delete_after_use = return_file_with_rbins(binfile)
with sys_pipes():
extn_results = DDsmu_extn(
autocorr,
cosmology,
nthreads,
mu_max,
nmu_bins,
sbinfile,
RA1,
DEC1,
CZ1,
is_comoving_dist=is_comoving_dist,
verbose=verbose,
output_savg=output_savg,
fast_divide_and_NR_steps=fast_divide_and_NR_steps,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa,
**kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
else:
extn_results, proj, proj_tensor, api_time = extn_results
if delete_after_use:
import os
os.remove(sbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'smin'), np.float),
(bytes_to_native_str(b'smax'), np.float),
(bytes_to_native_str(b'savg'), np.float),
(bytes_to_native_str(b'mumax'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
nbin = len(extn_results)
results = np.zeros(nbin, dtype=results_dtype)
for ii, r in enumerate(extn_results):
results['smin'][ii] = r[0]
results['smax'][ii] = r[1]
results['savg'][ii] = r[2]
results['mumax'][ii] = r[3]
results['npairs'][ii] = r[4]
results['weightavg'][ii] = r[5]
proj = np.array(proj)
projt = np.array(proj_tensor)
# Keep projt as 1d array
# nprojbins = len(proj)
# projt = np.zeros((nprojbins, nprojbins))
# for i in range(nprojbins):
# for j in range(nprojbins):
# projt[i][j] = proj_tensor[i*nprojbins+j]
if not c_api_timer:
return results, proj, projt
else:
return results, proj, projt, api_time
def xi(boxsize,
nthreads,
binfile,
X,
Y,
Z,
weights=None,
weight_type=None,
verbose=False,
output_ravg=False,
xbin_refine_factor=2,
ybin_refine_factor=2,
zbin_refine_factor=1,
max_cells_per_dim=100,
c_api_timer=False,
isa=r'fastest'):
"""
Function to compute the projected correlation function in a
periodic cosmological box. Pairs which are separated by less
than the ``r`` bins (specified in ``binfile``) in 3-D real space.
If ``weights`` are provided, the resulting correlation function
is weighted. The weighting scheme depends on ``weight_type``.
.. note:: Pairs are double-counted. And if ``rmin`` is set to
0.0, then all the self-pairs (i'th particle with itself) are
added to the first bin => minimum number of pairs in the first bin
is the total number of particles.
Parameters
-----------
boxsize: double
A double-precision value for the boxsize of the simulation
in same units as the particle positions and the ``r`` bins.
nthreads: integer
Number of threads to use.
binfile: string or an list/array of floats
For string input: filename specifying the ``r`` bins for
``xi``. The file should contain white-space separated values
of (rmin, rmax) for each ``r`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``r`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
X/Y/Z: arraytype, real (float/double)
Particle positions in the 3 axes. Must be within [0, boxsize]
and specified in the same units as ``rp_bins`` and boxsize. All
3 arrays must be of the same floating-point type.
Calculations will be done in the same precision as these arrays,
i.e., calculations will be in floating point if XYZ are single
precision arrays (C float type); or in double-precision if XYZ
are double precision arrays (C double type).
weights: array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or
(n_positions,). `weight_type` specifies how these weights are used;
results are returned in the `weightavg` field.
verbose: boolean (default false)
Boolean flag to control output of informational messages
output_ravg: boolean (default false)
Boolean flag to output the average ``r`` for each bin. Code will
run slower if you set this flag.
Note: If you are calculating in single-precision, ``rpavg`` will
suffer from numerical loss of precision and can not be trusted. If
you need accurate ``rpavg`` values, then pass in double precision
arrays for the particle positions.
(xyz)bin_refine_factor: integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer: boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code will
revert to using ``fallback`` (even though ``sse42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type: string, optional, Default: None.
The type of weighting to apply. One of ["pair_product", None].
Returns
--------
results: Numpy structured array
A numpy structured array containing [rmin, rmax, ravg, xi, npairs, weightavg] for
each radial specified in the ``binfile``. If ``output_ravg`` is not
set then ``ravg`` will be set to 0.0 for all bins; similarly for ``weightavg``.
``xi`` contains the correlation function while ``npairs`` contains the number of
pairs in that bin. If using weights, ``xi`` will be weighted while ``npairs``
will not be.
api_time: float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only the time spent
within the C library and ignores all python overhead.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.theory.xi import xi
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> N = 100000
>>> boxsize = 420.0
>>> nthreads = 4
>>> seed = 42
>>> np.random.seed(seed)
>>> X = np.random.uniform(0, boxsize, N)
>>> Y = np.random.uniform(0, boxsize, N)
>>> Z = np.random.uniform(0, boxsize, N)
>>> weights = np.ones_like(X)
>>> results = xi(boxsize, nthreads, binfile, X, Y, Z, weights=weights, weight_type='pair_product', output_ravg=True)
>>> for r in results: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10.6f} {4:10d} {5:10.6f}"
... .format(r['rmin'], r['rmax'],
... r['ravg'], r['xi'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
0.167536 0.238755 0.226592 -0.205733 4 1.000000
0.238755 0.340251 0.289277 -0.176729 12 1.000000
0.340251 0.484892 0.426819 -0.051829 40 1.000000
0.484892 0.691021 0.596187 -0.131853 106 1.000000
0.691021 0.984777 0.850100 -0.049207 336 1.000000
0.984777 1.403410 1.225112 0.028543 1052 1.000000
1.403410 2.000000 1.737153 0.011403 2994 1.000000
2.000000 2.850200 2.474588 0.005405 8614 1.000000
2.850200 4.061840 3.532018 -0.014098 24448 1.000000
4.061840 5.788530 5.022241 -0.010784 70996 1.000000
5.788530 8.249250 7.160648 -0.001588 207392 1.000000
8.249250 11.756000 10.207213 -0.000323 601002 1.000000
11.756000 16.753600 14.541171 0.000007 1740084 1.000000
16.753600 23.875500 20.728773 -0.001595 5028058 1.000000
"""
try:
from Corrfunc._countpairs import countpairs_xi as xi_extn
except ImportError:
msg = "Could not import the C extension for the projected "\
"correlation function."
raise ImportError(msg)
import numpy as np
from warnings import warn
from future.utils import bytes_to_native_str
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins, convert_to_native_endian,\
is_native_endian
# Broadcast scalar weights to arrays
if weights is not None:
weights = np.atleast_1d(weights)
# Warn about non-native endian arrays
if not all(is_native_endian(arr) for arr in [X, Y, Z, weights]):
warn(
'One or more input array has non-native endianness! A copy will be made with the correct endianness.'
)
X, Y, Z, weights = [
convert_to_native_endian(arr) for arr in [X, Y, Z, weights]
]
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights', 'weight_type']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time = xi_extn(boxsize,
nthreads,
rbinfile,
X,
Y,
Z,
verbose=verbose,
output_ravg=output_ravg,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa,
**kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'rmin'), np.float),
(bytes_to_native_str(b'rmax'), np.float),
(bytes_to_native_str(b'ravg'), np.float),
(bytes_to_native_str(b'xi'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def DDrppi(autocorr, nthreads, pimax, binfile, X1, Y1, Z1, weights1=None,
periodic=True, X2=None, Y2=None, Z2=None, weights2=None,
verbose=False, boxsize=0.0, output_rpavg=False,
xbin_refine_factor=2, ybin_refine_factor=2,
zbin_refine_factor=1, max_cells_per_dim=100,
c_api_timer=False, isa=r'fastest', weight_type=None):
"""
Calculate the 3-D pair-counts corresponding to the real-space correlation
function, :math:`\\xi(r_p, \pi)` or :math:`\\wp(r_p)`. Pairs which are
separated by less than the ``rp`` bins (specified in ``binfile``) in the
X-Y plane, and less than ``pimax`` in the Z-dimension are
counted.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
.. note:: that this module only returns pair counts and not the actual
correlation function :math:`\\xi(r_p, \pi)` or :math:`wp(r_p)`. See the
utilities :py:mod:`Corrfunc.utils.convert_3d_counts_to_cf` and
:py:mod:`Corrfunc.utils.convert_rp_pi_counts_to_wp` for computing
:math:`\\xi(r_p, \pi)` and :math:`wp(r_p)` respectively from the
pair counts.
Parameters
-----------
autocorr: boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
nthreads: integer
The number of OpenMP threads to use. Has no effect if OpenMP was not
enabled during library compilation.
pimax: double
A double-precision value for the maximum separation along
the Z-dimension.
Distances along the :math:``\\pi`` direction are binned with unit
depth. For instance, if ``pimax=40``, then 40 bins will be created
along the ``pi`` direction.
Note: Only pairs with ``0 <= dz < pimax`` are counted (no equality).
binfile: string or an list/array of floats
For string input: filename specifying the ``rp`` bins for
``DDrppi``. The file should contain white-space separated values
of (rpmin, rpmax) for each ``rp`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``rp`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0. This
array does not need to be sorted.
X1/Y1/Z1: array-like, real (float/double)
The array of X/Y/Z positions for the first set of points.
Calculations are done in the precision of the supplied arrays.
weights1: array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or (n_positions,).
`weight_type` specifies how these weights are used; results are returned
in the `weightavg` field. If only one of weights1 and weights2 is
specified, the other will be set to uniform weights.
X2/Y2/Z2: array-like, real (float/double)
Array of XYZ positions for the second set of points. *Must* be the same
precision as the X1/Y1/Z1 arrays. Only required when ``autocorr==0``.
weights2: array-like, real (float/double), optional
Same as weights1, but for the second set of positions
periodic: boolean
Boolean flag to indicate periodic boundary conditions.
verbose: boolean (default false)
Boolean flag to control output of informational messages
boxsize: double
The side-length of the cube in the cosmological simulation.
Present to facilitate exact calculations for periodic wrapping.
If boxsize is not supplied, then the wrapping is done based on
the maximum difference within each dimension of the X/Y/Z arrays.
output_rpavg: boolean (default false)
Boolean flag to output the average ``rp`` for each bin. Code will
run slower if you set this flag.
Note: If you are calculating in single-precision, ``rpavg`` will
suffer from numerical loss of precision and can not be trusted. If
you need accurate ``rpavg`` values, then pass in double precision
arrays for the particle positions.
(xyz)bin_refine_factor: integer, default is (2,2,1); typically within [1-3]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
c_api_timer: boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa: string (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Possible
options are: [``fastest``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code will
revert to using ``fallback`` (even though ``sse42`` might be available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type: string, optional
The type of weighting to apply. One of ["pair_product", None]. Default: None.
Returns
--------
results: Numpy structured array
A numpy structured array containing [rpmin, rpmax, rpavg, pimax, npairs, weightavg]
for each radial bin specified in the ``binfile``. If ``output_rpavg``
is not set, then ``rpavg`` will be set to 0.0 for all bins; similarly for
``weightavg``. ``npairs`` contains the number of pairs in that bin and can
be used to compute :math:`\\xi(r_p, \pi)` by combining with (DR, RR) counts.
api_time: float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only the time
spent within the C library and ignores all python overhead.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.theory.DDrppi import DDrppi
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../theory/tests/", "bins")
>>> N = 10000
>>> boxsize = 420.0
>>> nthreads = 4
>>> autocorr = 1
>>> pimax = 40.0
>>> seed = 42
>>> np.random.seed(seed)
>>> X = np.random.uniform(0, boxsize, N)
>>> Y = np.random.uniform(0, boxsize, N)
>>> Z = np.random.uniform(0, boxsize, N)
>>> weights = np.ones_like(X)
>>> results = DDrppi(autocorr, nthreads, pimax, binfile,
... X, Y, Z, weights1=weights, weight_type='pair_product', output_rpavg=True)
>>> for r in results[519:]: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10.1f}"
... " {4:10d} {5:10.6f}".format(r['rmin'], r['rmax'],
... r['rpavg'], r['pimax'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
11.756000 16.753600 14.379250 40.0 1150 1.000000
16.753600 23.875500 20.449131 1.0 2604 1.000000
16.753600 23.875500 20.604834 2.0 2370 1.000000
16.753600 23.875500 20.523989 3.0 2428 1.000000
16.753600 23.875500 20.475181 4.0 2462 1.000000
16.753600 23.875500 20.458005 5.0 2532 1.000000
16.753600 23.875500 20.537162 6.0 2522 1.000000
16.753600 23.875500 20.443087 7.0 2422 1.000000
16.753600 23.875500 20.474580 8.0 2360 1.000000
16.753600 23.875500 20.420360 9.0 2512 1.000000
16.753600 23.875500 20.478355 10.0 2472 1.000000
16.753600 23.875500 20.485268 11.0 2406 1.000000
16.753600 23.875500 20.372985 12.0 2420 1.000000
16.753600 23.875500 20.647998 13.0 2378 1.000000
16.753600 23.875500 20.556208 14.0 2420 1.000000
16.753600 23.875500 20.527992 15.0 2462 1.000000
16.753600 23.875500 20.581017 16.0 2380 1.000000
16.753600 23.875500 20.491819 17.0 2346 1.000000
16.753600 23.875500 20.534440 18.0 2496 1.000000
16.753600 23.875500 20.529129 19.0 2512 1.000000
16.753600 23.875500 20.501946 20.0 2500 1.000000
16.753600 23.875500 20.513349 21.0 2544 1.000000
16.753600 23.875500 20.471915 22.0 2430 1.000000
16.753600 23.875500 20.450651 23.0 2354 1.000000
16.753600 23.875500 20.550753 24.0 2460 1.000000
16.753600 23.875500 20.540262 25.0 2490 1.000000
16.753600 23.875500 20.559572 26.0 2350 1.000000
16.753600 23.875500 20.534245 27.0 2382 1.000000
16.753600 23.875500 20.511302 28.0 2508 1.000000
16.753600 23.875500 20.491632 29.0 2456 1.000000
16.753600 23.875500 20.592493 30.0 2386 1.000000
16.753600 23.875500 20.506234 31.0 2484 1.000000
16.753600 23.875500 20.482109 32.0 2538 1.000000
16.753600 23.875500 20.518463 33.0 2544 1.000000
16.753600 23.875500 20.482515 34.0 2534 1.000000
16.753600 23.875500 20.503124 35.0 2382 1.000000
16.753600 23.875500 20.471307 36.0 2356 1.000000
16.753600 23.875500 20.384231 37.0 2554 1.000000
16.753600 23.875500 20.454012 38.0 2458 1.000000
16.753600 23.875500 20.585543 39.0 2394 1.000000
16.753600 23.875500 20.504965 40.0 2500 1.000000
"""
try:
from Corrfunc._countpairs import countpairs_rp_pi as DDrppi_extn
except ImportError:
msg = "Could not import the C extension for the 3-D "\
"real-space pair counter."
raise ImportError(msg)
import numpy as np
from warnings import warn
from Corrfunc.utils import translate_isa_string_to_enum,\
return_file_with_rbins, convert_to_native_endian,\
is_native_endian
from future.utils import bytes_to_native_str
# Broadcast scalar weights to arrays
if weights1 is not None:
weights1 = np.atleast_1d(weights1)
if weights2 is not None:
weights2 = np.atleast_1d(weights2)
if not autocorr:
if X2 is None or Y2 is None or Z2 is None:
msg = "Must pass valid arrays for X2/Y2/Z2 for "\
"computing cross-correlation"
raise ValueError(msg)
# If only one set of points has weights, set the other to uniform weights
if weights1 is None and weights2 is not None:
weights1 = np.ones_like(weights2)
if weights2 is None and weights1 is not None:
weights2 = np.ones_like(weights1)
else:
X2 = np.empty(1)
Y2 = np.empty(1)
Z2 = np.empty(1)
# Warn about non-native endian arrays
if not all(is_native_endian(arr) for arr in [X1, Y1, Z1, weights1, X2, Y2, Z2, weights2]):
warn('One or more input array has non-native endianness! A copy will be made with the correct endianness.')
X1, Y1, Z1, weights1, X2, Y2, Z2, weights2 = [convert_to_native_endian(arr) for arr in [X1, Y1, Z1, weights1, X2, Y2, Z2, weights2]]
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'X2', 'Y2', 'Z2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
extn_results, api_time = DDrppi_extn(autocorr, nthreads,
pimax, rbinfile,
X1, Y1, Z1,
periodic=periodic,
verbose=verbose,
boxsize=boxsize,
output_rpavg=output_rpavg,
xbin_refine_factor=xbin_refine_factor,
ybin_refine_factor=ybin_refine_factor,
zbin_refine_factor=zbin_refine_factor,
max_cells_per_dim=max_cells_per_dim,
c_api_timer=c_api_timer,
isa=integer_isa, **kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'rmin'), np.float),
(bytes_to_native_str(b'rmax'), np.float),
(bytes_to_native_str(b'rpavg'), np.float),
(bytes_to_native_str(b'pimax'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float),])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def DDtheta_mocks(autocorr, nthreads, binfile,
RA1, DEC1, weights1=None,
RA2=None, DEC2=None, weights2=None,
link_in_dec=True, link_in_ra=True,
verbose=False, output_thetaavg=False,
fast_acos=False, ra_refine_factor=2,
dec_refine_factor=2, max_cells_per_dim=100,
copy_particles=True, enable_min_sep_opt=True,
c_api_timer=False, isa=r'fastest', weight_type=None):
"""
Function to compute the angular correlation function for points on
the sky (i.e., mock catalogs or observed galaxies).
Returns a numpy structured array containing the pair counts for the
specified angular bins.
If ``weights`` are provided, the resulting pair counts are weighted. The
weighting scheme depends on ``weight_type``.
.. note:: This module only returns pair counts and not the actual
correlation function :math:`\\omega(\theta)`. See
:py:mod:`Corrfunc.utils.convert_3d_counts_to_cf` for computing
:math:`\\omega(\theta)` from the pair counts returned.
Parameters
-----------
autocorr : boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
nthreads : integer
Number of threads to use.
binfile: string or an list/array of floats. Units: degrees.
For string input: filename specifying the ``theta`` bins for
``DDtheta_mocks``. The file should contain white-space separated values
of (thetamin, thetamax) for each ``theta`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``theta`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0
degrees. This array does not need to be sorted.
RA1 : array-like, real (float/double)
The array of Right Ascensions for the first set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Calculations are done in the precision of the supplied arrays.
DEC1 : array-like, real (float/double)
Array of Declinations for the first set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1.
weights1 : array_like, real (float/double), optional
A scalar, or an array of weights of shape (n_weights, n_positions) or
(n_positions,). `weight_type` specifies how these weights are used;
results are returned in the `weightavg` field. If only one of weights1
and weights2 is specified, the other will be set to uniform weights.
RA2 : array-like, real (float/double)
The array of Right Ascensions for the second set of points. RA's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA's are in [-180, 180.0]. For peace of mind, always supply
RA's in [0.0, 360.0].
Must be of same precision type as RA1/DEC1.
DEC2 : array-like, real (float/double)
Array of Declinations for the second set of points. DEC's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC's in [-90.0, 90.0].
Must be of same precision type as RA1/DEC1.
weights2 : array-like, real (float/double), optional
Same as weights1, but for the second set of positions
link_in_dec : boolean (default True)
Boolean flag to create lattice in Declination. Code runs faster with
this option. However, if the angular separations are too small, then
linking in declination might produce incorrect results. When running
for the first time, check your results by comparing with the output
of the code for ``link_in_dec=False`` and ``link_in_ra=False``.
link_in_ra : boolean (default True)
Boolean flag to create lattice in Right Ascension. Setting this option
implies ``link_in_dec=True``. Similar considerations as ``link_in_dec``
described above.
If you disable both ``link_in_dec`` and ``link_in_ra``, then
the code reduces to a brute-force pair counter. No lattices are created
at all. For very small angular separations, the brute-force method
might be the most numerically stable method.
verbose : boolean (default false)
Boolean flag to control output of informational messages
output_thetaavg : boolean (default false)
Boolean flag to output the average ``\theta`` for each bin. Code will
run slower if you set this flag.
If you are calculating in single-precision, ``thetaavg`` will
suffer from numerical loss of precision and can not be trusted. If you
need accurate ``thetaavg`` values, then pass in double precision arrays
for ``RA/DEC``.
Code will run significantly slower if you enable this option.
Use the keyword ``fast_acos`` if you can tolerate some loss of
precision.
fast_acos : boolean (default false)
Flag to use numerical approximation for the ``arccos`` - gives better
performance at the expense of some precision. Relevant only if
``output_thetaavg==True``.
Developers: Two versions already coded up in ``utils/fast_acos.h``,
so you can choose the version you want. There are also notes on how
to implement faster (and less accurate) functions, particularly
relevant if you know your ``theta`` range is limited. If you implement
a new version, then you will have to reinstall the entire Corrfunc
package.
Note: Tests will fail if you run the tests with``fast_acos=True``.
(radec)_refine_factor : integer, default is (2,2); typically within [1-5]
Controls the refinement on the cell sizes. Can have up to a 20% impact
on runtime.
Only two refine factors are to be specified and these
correspond to ``ra`` and ``dec`` (rather, than the usual three of
``(xyz)bin_refine_factor`` for all other correlation functions).
max_cells_per_dim : integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^3. Only increase if ``thetamax``
is too small relative to the boxsize (and increasing helps the
runtime).
copy_particles: boolean (default True)
Boolean flag to make a copy of the particle positions
If set to False, the particles will be re-ordered in-place
.. versionadded:: 2.3.0
enable_min_sep_opt: boolean (default true)
Boolean flag to allow optimizations based on min. separation between
pairs of cells. Here to allow for comparison studies.
.. versionadded:: 2.3.0
c_api_timer : boolean (default false)
Boolean flag to measure actual time spent in the C libraries. Here
to allow for benchmarking and scaling studies.
isa: string, case-insensitive (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Options
are: [``fastest``, ``avx512f``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code
will revert to using ``fallback`` (even though ``sse42`` might be
available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
weight_type : string, optional (default None)
The type of weighting to apply. One of ["pair_product", None].
Returns
--------
results : Numpy structured array
A numpy structured array containing [thetamin, thetamax, thetaavg,
npairs, weightavg] for each angular bin specified in the ``binfile``.
If ``output_thetaavg`` is not set then ``thetavg`` will be set to 0.0
for all bins; similarly for ``weightavg``. ``npairs`` contains the
number of pairs in that bin.
api_time : float, optional
Only returned if ``c_api_timer`` is set. ``api_time`` measures only
the time spent within the C library and ignores all python overhead.
Example
--------
>>> from __future__ import print_function
>>> import numpy as np
>>> import time
>>> from math import pi
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.mocks.DDtheta_mocks import DDtheta_mocks
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../mocks/tests/", "angular_bins")
>>> N = 100000
>>> nthreads = 4
>>> seed = 42
>>> np.random.seed(seed)
>>> RA = np.random.uniform(0.0, 2.0*pi, N)*180.0/pi
>>> cos_theta = np.random.uniform(-1.0, 1.0, N)
>>> DEC = 90.0 - np.arccos(cos_theta)*180.0/pi
>>> weights = np.ones_like(RA)
>>> autocorr = 1
>>> for isa in ['AVX', 'SSE42', 'FALLBACK']:
... for link_in_dec in [False, True]:
... for link_in_ra in [False, True]:
... results = DDtheta_mocks(autocorr, nthreads, binfile,
... RA, DEC, output_thetaavg=True,
... weights1=weights, weight_type='pair_product',
... link_in_dec=link_in_dec, link_in_ra=link_in_ra,
... isa=isa, verbose=True)
>>> for r in results: print("{0:10.6f} {1:10.6f} {2:10.6f} {3:10d} {4:10.6f}".
... format(r['thetamin'], r['thetamax'],
... r['thetaavg'], r['npairs'], r['weightavg']))
... # doctest: +NORMALIZE_WHITESPACE
0.010000 0.014125 0.012272 62 1.000000
0.014125 0.019953 0.016978 172 1.000000
0.019953 0.028184 0.024380 298 1.000000
0.028184 0.039811 0.034321 598 1.000000
0.039811 0.056234 0.048535 1164 1.000000
0.056234 0.079433 0.068385 2438 1.000000
0.079433 0.112202 0.096631 4658 1.000000
0.112202 0.158489 0.136834 9414 1.000000
0.158489 0.223872 0.192967 19098 1.000000
0.223872 0.316228 0.272673 37848 1.000000
0.316228 0.446684 0.385344 75520 1.000000
0.446684 0.630957 0.543973 150938 1.000000
0.630957 0.891251 0.768406 301854 1.000000
0.891251 1.258925 1.085273 599896 1.000000
1.258925 1.778279 1.533461 1200238 1.000000
1.778279 2.511886 2.166009 2396338 1.000000
2.511886 3.548134 3.059159 4775162 1.000000
3.548134 5.011872 4.321445 9532582 1.000000
5.011872 7.079458 6.104214 19001930 1.000000
7.079458 10.000000 8.622400 37842502 1.000000
"""
try:
from Corrfunc._countpairs_mocks import countpairs_theta_mocks as\
DDtheta_mocks_extn
except ImportError:
msg = "Could not import the C extension for the angular "\
"correlation function for mocks."
raise ImportError(msg)
import numpy as np
from Corrfunc.utils import translate_isa_string_to_enum, fix_ra_dec,\
return_file_with_rbins, convert_to_native_endian,\
sys_pipes, process_weights
from future.utils import bytes_to_native_str
if autocorr == 0:
if RA2 is None or DEC2 is None:
msg = "Must pass valid arrays for RA2/DEC2 for "\
"computing cross-correlation"
raise ValueError(msg)
else:
RA2 = np.empty(1)
DEC2 = np.empty(1)
weights1, weights2 = process_weights(weights1, weights2, RA1, RA2, weight_type, autocorr)
# Ensure all input arrays are native endian
RA1, DEC1, weights1, RA2, DEC2, weights2 = [
convert_to_native_endian(arr, warn=True) for arr in
[RA1, DEC1, weights1, RA2, DEC2, weights2]]
fix_ra_dec(RA1, DEC1)
if autocorr == 0:
fix_ra_dec(RA2, DEC2)
if link_in_ra is True:
link_in_dec = True
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'RA2', 'DEC2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
with sys_pipes():
extn_results = DDtheta_mocks_extn(autocorr, nthreads, rbinfile,
RA1, DEC1,
verbose=verbose,
link_in_dec=link_in_dec,
link_in_ra=link_in_ra,
output_thetaavg=output_thetaavg,
fast_acos=fast_acos,
ra_refine_factor=ra_refine_factor,
dec_refine_factor=dec_refine_factor,
max_cells_per_dim=max_cells_per_dim,
copy_particles=copy_particles,
enable_min_sep_opt=enable_min_sep_opt,
c_api_timer=c_api_timer,
isa=integer_isa, **kwargs)
if extn_results is None:
msg = "RuntimeError occurred"
raise RuntimeError(msg)
else:
extn_results, api_time = extn_results
if delete_after_use:
import os
os.remove(rbinfile)
results_dtype = np.dtype([(bytes_to_native_str(b'thetamin'), np.float),
(bytes_to_native_str(b'thetamax'), np.float),
(bytes_to_native_str(b'thetaavg'), np.float),
(bytes_to_native_str(b'npairs'), np.uint64),
(bytes_to_native_str(b'weightavg'), np.float)])
results = np.array(extn_results, dtype=results_dtype)
if not c_api_timer:
return results
else:
return results, api_time
def find_fastest_DDtheta_mocks_bin_refs(autocorr, nthreads, binfile,
RA1, DEC1, RA2=None, DEC2=None,
link_in_dec=True, link_in_ra=True,
verbose=False, output_thetaavg=False,
max_cells_per_dim=100,
isa=r'fastest',
maxbinref=3, nrepeats=3,
return_runtimes=False):
"""
Parameters
----------
autocorr : boolean, required
Boolean flag for auto/cross-correlation. If autocorr is set to 1,
then the second set of particle positions are not required.
nthreads : integer
Number of threads to use.
binfile: string or an list/array of floats. Units: degrees.
For string input: filename specifying the ``theta`` bins for
``DDtheta_mocks``. The file should contain white-space separated values
of (thetamin, thetamax) for each ``theta`` wanted. The bins need to be
contiguous and sorted in increasing order (smallest bins come first).
For array-like input: A sequence of ``theta`` values that provides the
bin-edges. For example,
``np.logspace(np.log10(0.1), np.log10(10.0), 15)`` is a valid
input specifying **14** (logarithmic) bins between 0.1 and 10.0
degrees. This array does not need to be sorted.
RA1 : array-like, real (float/double)
The array of Right Ascensions for the first set of points. RA1's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA1's are in [-180, 180.0]. For peace of mind, always supply
RA1's in [0.0, 360.0].
Calculations are done in the precision of the supplied arrays.
DEC1 : array-like, real (float/double)
Array of Declinations for the first set of points. DEC1's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC1's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC1's in [-90.0, 90.0].
Must be of same precision type as RA1.
RA2 : array-like, real (float/double)
The array of Right Ascensions for the second set of points. RA2's
are expected to be in [0.0, 360.0], but the code will try to fix cases
where the RA2's are in [-180, 180.0]. For peace of mind, always supply
RA2's in [0.0, 360.0].
Must be of same precision type as RA1/DEC1.
DEC2 : array-like, real (float/double)
Array of Declinations for the second set of points. DEC2's are expected
to be in the [-90.0, 90.0], but the code will try to fix cases where
the DEC2's are in [0.0, 180.0]. Again, for peace of mind, always supply
DEC2's in [-90.0, 90.0].
Must be of same precision type as RA1/DEC1.
verbose : boolean (default false)
Boolean flag to control output of informational messages
output_thetaavg : boolean (default false)
Boolean flag to output the average ``\theta`` for each bin. Code will
run slower if you set this flag.
If you are calculating in single-precision, ``thetaavg`` will
suffer from numerical loss of precision and can not be trusted. If you
need accurate ``thetaavg`` values, then pass in double precision arrays
for ``RA/DEC``.
isa: string, case-insensitive (default ``fastest``)
Controls the runtime dispatch for the instruction set to use. Options
are: [``fastest``, ``avx512f``, ``avx``, ``sse42``, ``fallback``]
Setting isa to ``fastest`` will pick the fastest available instruction
set on the current computer. However, if you set ``isa`` to, say,
``avx`` and ``avx`` is not available on the computer, then the code
will revert to using ``fallback`` (even though ``sse42`` might be
available).
Unless you are benchmarking the different instruction sets, you should
always leave ``isa`` to the default value. And if you *are*
benchmarking, then the string supplied here gets translated into an
``enum`` for the instruction set defined in ``utils/defs.h``.
max_cells_per_dim: integer, default is 100, typical values in [50-300]
Controls the maximum number of cells per dimension. Total number of
cells can be up to (max_cells_per_dim)^2. Only increase if ``rpmax`` is
too small relative to the boxsize (and increasing helps the runtime).
maxbinref: integer (default 3)
The maximum ``bin refine factor`` to use along each dimension.
Runtime of module scales as ``maxbinref^2``, so change the value of
``maxbinref`` with caution.
Note that ``max_cells_per_dim`` might need to be increased
to accommodate really large ``maxbinref``.
nrepeats: integer (default 3)
Number of times to repeat the timing for an individual run. Accounts
for the dispersion in runtimes on computers with multiple user
processes.
return_runtimes: boolean (default ``false``)
If set, also returns the array of runtimes.
Returns
-------
(nRA, nDEC) : tuple of integers
The combination of ``bin refine factors`` along each dimension that
produces the fastest code.
runtimes : numpy structured array
Only returned if ``return_runtimes`` is set, then the return value
is a tuple containing ((nRA, nDEC), runtimes). ``runtimes`` is a
``numpy`` structured array containing the fields, [``nRA``, ``nDEC``,
``avg_runtime``, ``sigma_time``]. Here, ``avg_runtime`` is the
average time, measured over ``nrepeats`` invocations, spent in
the python extension. ``sigma_time`` is the dispersion of the
run times across those ``nrepeats`` invocations.
Example
-------
>>> from __future__ import print_function
>>> import numpy as np
>>> import time
>>> from math import pi
>>> from os.path import dirname, abspath, join as pjoin
>>> import Corrfunc
>>> from Corrfunc.mocks.DDtheta_mocks \
import find_fastest_DDtheta_mocks_bin_refs
>>> binfile = pjoin(dirname(abspath(Corrfunc.__file__)),
... "../mocks/tests/", "angular_bins")
>>> N = 100000
>>> nthreads = 4
>>> seed = 42
>>> np.random.seed(seed)
>>> RA1 = np.random.uniform(0.0, 2.0*pi, N)*180.0/pi
>>> cos_theta = np.random.uniform(-1.0, 1.0, N)
>>> DEC1 = 90.0 - np.arccos(cos_theta)*180.0/pi
>>> autocorr = 1
>>> best, _ = find_fastest_DDtheta_mocks_bin_refs(autocorr, nthreads,\
... binfile, RA1, DEC1,\
... return_runtimes=True)
>>> print(best) # doctest:+SKIP
(2, 1)
.. note:: Since the result might change depending on the computer,
doctest is skipped for this function.
"""
weights1 = None
weights2 = None
weight_type = None
try:
from Corrfunc._countpairs_mocks import countpairs_theta_mocks as\
DDtheta_mocks_extn
except ImportError:
msg = "Could not import the C extension for the angular "\
"correlation function for mocks."
raise ImportError(msg)
import numpy as np
from Corrfunc.utils import translate_isa_string_to_enum, fix_ra_dec,\
return_file_with_rbins, convert_to_native_endian, process_weights
from future.utils import bytes_to_native_str
import itertools
import time
weights1, weights2 = process_weights(weights1, weights2, RA1, RA2,
weight_type, autocorr)
# Ensure all input arrays are native endian
RA1, DEC1, weights1, RA2, DEC2, weights2 = [
convert_to_native_endian(arr, warn=True) for arr in
[RA1, DEC1, weights1, RA2, DEC2, weights2]]
fix_ra_dec(RA1, DEC1)
if autocorr == 0:
fix_ra_dec(RA2, DEC2)
# Passing None parameters breaks the parsing code, so avoid this
kwargs = {}
for k in ['weights1', 'weights2', 'weight_type', 'RA2', 'DEC2']:
v = locals()[k]
if v is not None:
kwargs[k] = v
integer_isa = translate_isa_string_to_enum(isa)
rbinfile, delete_after_use = return_file_with_rbins(binfile)
bin_refs = np.arange(1, maxbinref + 1)
if link_in_ra:
bin_ref_perms = itertools.product(bin_refs, bin_refs)
nperms = maxbinref ** 2
else:
bin_ref_perms = [(1, binref) for binref in bin_refs]
nperms = maxbinref
dtype = np.dtype([(bytes_to_native_str(b'nRA'), np.int),
(bytes_to_native_str(b'nDEC'), np.int),
(bytes_to_native_str(b'avg_time'), np.float),
(bytes_to_native_str(b'sigma_time'), np.float)])
all_runtimes = np.zeros(nperms, dtype=dtype)
all_runtimes[:] = np.inf
if autocorr == 0:
if RA2 is None or DEC2 is None:
msg = "Must pass valid arrays for RA2/DEC2 for "\
"computing cross-correlation."
raise ValueError(msg)
else:
RA2 = np.empty(1)
DEC2 = np.empty(1)
if link_in_ra:
link_in_dec = True
if not link_in_dec:
msg = "Error: Brute-force calculation without any gridding " \
"is forced, as link_in_dec and link_in_ra are both set " \
"to False. Please set at least one of link_in_dec, link_in_ra=True " \
"to enable gridding along DEC or along both RA and DEC."
raise ValueError(msg)
if verbose and not link_in_ra:
msg = "INFO: Since ``link_in_ra`` is not set, only gridding in declination " \
"Checking with refinements in declination ranging from [1, {}] and a " \
"maximum of {} bins".format(maxbinref, max_cells_per_dim)
print(msg)
for ii, (nRA, nDEC) in enumerate(bin_ref_perms):
total_runtime = 0.0
total_sqr_runtime = 0.0
for _ in range(nrepeats):
t0 = time.perf_counter()
extn_results = DDtheta_mocks_extn(autocorr, nthreads, rbinfile,
RA1, DEC1, RA2, DEC2,
link_in_dec=link_in_dec,
link_in_ra=link_in_ra,
verbose=verbose,
output_thetaavg=output_thetaavg,
ra_refine_factor=nRA,
dec_refine_factor=nDEC,
max_cells_per_dim=max_cells_per_dim,
isa=integer_isa)
t1 = time.perf_counter()
if extn_results is None:
msg = "RuntimeError occurred with perms = ({0}, {1})".\
format(nRA, nDEC)
raise ValueError(msg)
dt = (t1 - t0)
total_runtime += dt
total_sqr_runtime += dt * dt
avg_runtime = total_runtime / nrepeats
# variance = E(X^2) - E^2(X)
# disp = sqrt(variance)
runtime_disp = np.sqrt(total_sqr_runtime / nrepeats -
avg_runtime * avg_runtime)
all_runtimes[ii]['nRA'] = nRA
all_runtimes[ii]['nDEC'] = nDEC
all_runtimes[ii]['avg_time'] = avg_runtime
all_runtimes[ii]['sigma_time'] = runtime_disp
if delete_after_use:
import os
os.remove(rbinfile)
all_runtimes.sort(order=('avg_time', 'sigma_time'))
results = (all_runtimes[0]['nRA'],
all_runtimes[0]['nDEC'])
optional_returns = return_runtimes
if not optional_returns:
ret = results
else:
ret = (results,)
if return_runtimes:
ret += (all_runtimes,)
return ret
