Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Ch/kltransform fsorder #102

Draft
wants to merge 6 commits into
base: master
Choose a base branch
from
Draft

Ch/kltransform fsorder #102

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
ace5ff9
feat(kltransform): allow F/S diagonalisation order
Jul 2, 2020
de682ea
Merge branch 'master' into ch/kltransform
Feb 1, 2021
b7733b1
fix(doublekl): black
Feb 2, 2021
fc33e42
Merge branch 'master' into ch/kltransform-fsorder
Mar 26, 2021
853e72f
docs(KLTransform): explain diagonalisation order and _dkl
Mar 26, 2021
14b571c
fix(doublekl): remove unnecessary if/else statement
Mar 29, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 14 additions & 4 deletions drift/core/doublekl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@ class DoubleKL(kltransform.KLTransform):
"""

foreground_threshold = config.Property(proptype=float, default=100.0)
_dkl = True

def _transform_m(self, mi):

Expand All @@ -42,13 +43,22 @@ def _transform_m(self, mi):
cs, cn = [cv.reshape(nside, nside) for cv in self.sn_covariance(mi)]

# Find joint eigenbasis and transformation matrix
evals, evecs2, ac = kltransform.eigh_gen(
cs, cn, message="m = %d; KL step 1" % mi
)
if self.diagonalisation_order == "sf":
evals, evecs2, ac = kltransform.eigh_gen(
cs, cn, message="m = %d; KL step 1" % mi
)
else:
evals, evecs2, ac = kltransform.eigh_gen(
cn, cs, message="m = %d; KL step 1" % mi
)

evecs = evecs2.T.conj()

# Get the indices that extract the high S/F ratio modes
ind = np.where(evals > self.foreground_threshold)
if self.diagonalisation_order == "sf":
ind = np.where(evals > self.foreground_threshold)
else:
ind = np.where(evals < self.foreground_threshold)

# Construct evextra dictionary (holding foreground ratio)
evextra = {"ac": ac, "f_evals": evals.copy()}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we consider using sf_evals for S/F transform and fs_evals for F/S transform, to make it more explicit which eigenvalues are being stored?

Expand Down
66 changes: 53 additions & 13 deletions drift/core/kltransform.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -154,6 +154,9 @@ class KLTransform(config.Reader):
If True, throw away modes below a S/N `threshold`.
threshold : scalar
S/N threshold to cut modes at.
diagonalisation_order : string, optional
Options are 'sf' for eigen value decomposition with S/F eigen values and
'fs' for F/S eigen values.
inverse : boolean
If True construct and cache inverse transformation.
use_thermal, use_foregrounds : boolean
Expand All @@ -168,6 +171,8 @@ class KLTransform(config.Reader):

threshold = config.Property(proptype=float, default=0.1, key="threshold")

diagonalisation_order = config.enum(["sf", "fs"], default="sf")

_foreground_regulariser = config.Property(
proptype=float, default=1e-14, key="regulariser"
)
Expand All @@ -182,6 +187,7 @@ class KLTransform(config.Reader):

_cvfg = None
_cvsg = None
_dkl = None

@property
def _evfile(self):
Expand Down Expand Up @@ -341,7 +347,12 @@ def _transform_m(self, mi):

# Perform the generalised eigenvalue problem to get the KL-modes.
st = time.time()
evals, evecs, ac = eigh_gen(cvb_sr, cvb_nr, message="m = %d" % mi)

if self.diagonalisation_order == "sf":
evals, evecs, ac = eigh_gen(cvb_sr, cvb_nr, message="m = %d" % mi)
else:
evals, evecs, ac = eigh_gen(cvb_nr, cvb_sr, message="m = %d" % mi)

et = time.time()
print("Time =", (et - st))

Expand Down Expand Up @@ -391,18 +402,27 @@ def transform_save(self, mi):
evalsf = np.zeros(nside, dtype=np.float64)
if evals.size != 0:
evalsf[(-evals.size) :] = evals

f.create_dataset("evals_full", data=evalsf)

# Discard eigenmodes with S/N below threshold if requested.
if self.subset:
i_ev = np.searchsorted(evals, self.threshold)

evals = evals[i_ev:]
evecs = evecs[i_ev:]
print(
"Modes with S/N > %f: %i of %i"
% (self.threshold, evals.size, evalsf.size)
)
# The second step in the double KL will be eigen values of S/N regardless of first step KL step. We save the double KL values in S/N order.
if self.diagonalisation_order == "sf" or self._dkl:
sjforeman marked this conversation as resolved.
Show resolved Hide resolved
evals = evals[i_ev:]
evecs = evecs[i_ev:]
print(
"Modes with S/N > %f: %i of %i"
% (self.threshold, evals.size, evalsf.size)
)
else:
evals = evals[:i_ev]
evecs = evecs[:i_ev]
print(
"Modes with N/S < %f: %i of %i"
% (self.threshold, evals.size, evalsf.size)
)

# Write out potentially reduced eigen spectrum.
f.create_dataset("evals", data=evals)
Expand All @@ -411,7 +431,10 @@ def transform_save(self, mi):

if self.inverse:
if self.subset:
inv = inv[i_ev:]
if self.diagonalisation_order == "sf" or self._dkl:
inv = inv[i_ev:]
else:
inv = inv[:i_ev]

f.create_dataset("evinv", data=inv)

Expand Down Expand Up @@ -461,6 +484,7 @@ def evfunc(mi):
if f["evals_full"].shape[0] > 0:
ev = f["evals_full"][:]
evf[-ev.size :] = ev

f.close()

return evf
Expand Down Expand Up @@ -495,6 +519,9 @@ def generate(self, regen=False):
if mpiutil.rank0:
st = time.time()
print("======== Starting KL calculation ========")
print(
"======= Diagonalisation order: %s =======" % self.diagonalisation_order
)

# Iterate list over MPI processes.
for mi in mpiutil.mpirange(self.telescope.mmax + 1):
Expand Down Expand Up @@ -566,7 +593,10 @@ def modes_m(self, mi, threshold=None):
if startind == evals.size:
modes = None, None
else:
modes = (evals[startind:], f["evecs"][startind:])
if self.diagonalisation_order == "sf" or self._dkl:
modes = (evals[startind:], f["evecs"][startind:])
else:
modes = (evals[:startind], f["evecs"][:startind])

# If old data file perform complex conjugate
modes = (
Expand Down Expand Up @@ -623,7 +653,10 @@ def evals_m(self, mi, threshold=None):
if startind == evals.size:
modes = None
else:
modes = evals[startind:]
if self.diagonalisation_order == "sf" or self._dkl:
modes = evals[startind:]
else:
modes = evals[:startind]

f.close()

Expand Down Expand Up @@ -656,7 +689,10 @@ def invmodes_m(self, mi, threshold=None):

if threshold != None:
nevals = evals.size
inv = inv[(-nevals):]
if self.diagonalisation_order == "sf" or self._dkl:
inv = inv[(-nevals):]
else:
inv = inv[:nevals]

return inv.T

Expand Down Expand Up @@ -892,7 +928,11 @@ def project_sky(self, sky, mlist=None, threshold=None, harmonic=False):
def _proj(mi):
p1 = self.project_sky_vector_forward(mi, alm[:, :, mi], threshold)
p2 = np.zeros(nmodes, dtype=np.complex128)
p2[-p1.size :] = p1
if self.diagonalisation_order == "sf" or self._dkl:
p2[-p1.size :] = p1
else:
p2[: p1.size] = p1

return p2

# Map over list of m's and project sky onto eigenbasis
Expand Down