Skip to content

Commit

Permalink
Refactor by LSK, round 2
Browse files Browse the repository at this point in the history
  • Loading branch information
@leeskelvin
leeskelvin committed Jul 30, 2024
1 parent 6d8e9e6 commit fea12d4
Showing 1 changed file with 98 additions and 109 deletions.
207 changes: 98 additions & 109 deletions python/lsst/pipe/tasks/matchBackgrounds.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@

import lsstDebug
import numpy as np
from lsst.afw.image import LOCAL, PARENT, ImageF, Mask, MaskedImageF
from lsst.afw.image import LOCAL, PARENT, ExposureF, ImageF, Mask, MaskedImageF
from lsst.afw.math import (
MEAN,
MEANCLIP,
Expand All @@ -35,6 +35,7 @@
ApproximateControl,
BackgroundControl,
BackgroundList,
BackgroundMI,
StatisticsControl,
makeBackground,
makeStatistics,
Expand Down Expand Up @@ -84,11 +85,11 @@ class MatchBackgroundsConfig(PipelineTaskConfig, pipelineConnections=MatchBackgr
doc="Visit ID of the reference warp. If None, the best warp is chosen from the list of warps.",
optional=True,
)
bestRefWeightCoverage = RangeField(
bestRefWeightChi2 = RangeField(
dtype=float,
doc="Coverage weight (Number of pixels overlapping the patch) when calculating the best reference "
"exposure. Higher weights prefer exposures with high coverage. Ignored when a ref visit supplied.",
default=0.4,
doc="Mean background goodness of fit statistic weight when calculating the best reference exposure. "
"Higher weights prefer exposures with flatter backgrounds. Ignored when ref visit supplied.",
default=0.2,
min=0.0,
max=1.0,
)
Expand All @@ -100,10 +101,18 @@ class MatchBackgroundsConfig(PipelineTaskConfig, pipelineConnections=MatchBackgr
min=0.0,
max=1.0,
)
bestRefWeightLevel = RangeField(
bestRefWeightGlobalCoverage = RangeField(
dtype=float,
doc="Mean background level weight when calculating the best reference exposure. "
"Higher weights prefer exposures with low mean background levels. Ignored when ref visit supplied.",
doc="Global coverage weight (total number of valid pixels) when calculating the best reference "
"exposure. Higher weights prefer exposures with high coverage. Ignored when a ref visit supplied.",
default=0.2,
min=0.0,
max=1.0,
)
bestRefWeightEdgeCoverage = RangeField(
dtype=float,
doc="Edge coverage weight (number of valid edge pixels) when calculating the best reference "
"exposure. Higher weights prefer exposures with high coverage. Ignored when a ref visit supplied.",
default=0.2,
min=0.0,
max=1.0,
Expand Down Expand Up @@ -143,7 +152,7 @@ class MatchBackgroundsConfig(PipelineTaskConfig, pipelineConnections=MatchBackgr
)
binSize = Field[int](
doc="Bin size for gridding the difference image and fitting a spatial model.",
default=256,
default=1024,
)
interpStyle = ChoiceField(
dtype=str,
Expand Down Expand Up @@ -213,11 +222,16 @@ class MatchBackgroundsTask(PipelineTask):

def __init__(self, *args, **kwargs):
super().__init__(**kwargs)
self.statsFlag = stringToStatisticsProperty(self.config.gridStatistic)
self.statsCtrl = StatisticsControl()
# TODO: Check that setting the mask planes here work - these planes
# can vary from exposure to exposure, I think?
self.statsCtrl.setAndMask(Mask.getPlaneBitMask(self.config.badMaskPlanes))
self.statsCtrl.setNanSafe(True)
self.statsCtrl.setNumSigmaClip(self.config.numSigmaClip)
self.statsCtrl.setNumIter(self.config.numIter)
self.stringToInterpStyle = stringToInterpStyle(self.config.interpStyle)
self.undersampleStyle = stringToUndersampleStyle(self.config.undersampleStyle)

def runQuantum(self, butlerQC, inputRefs, outputRefs):
inputs = butlerQC.get(inputRefs)
Expand Down Expand Up @@ -333,154 +347,129 @@ def _defineWarps(self, warps, refWarpVisit=None):
minimizing a cost function that penalizes high variance, high
background level, and low coverage.
To find a reference warp, the following config parameters are used:
- ``bestRefWeightCoverage``
- ``bestRefWeightVariance``
- ``bestRefWeightLevel``
Parameters
----------
warps : `list`[`~lsst.afw.image.Exposure`]
List of warped science exposures.
warps : `list`[`~lsst.daf.butler.DeferredDatasetHandle`]
List of warped exposures (of type `~lsst.afw.image.ExposureF`).
refWarpVisit : `int`, optional
Visit ID of the reference warp.
If None, the best warp is chosen from the list of warps.
Returns
-------
refWarp : `~lsst.afw.image.Exposure`
refWarp : `~lsst.afw.image.ExposureF`
Reference warped exposure.
compWarps : `list`[`~lsst.afw.image.Exposure`]
List of warped science exposures to compare to the reference.
refWarpIndex : `int`
Index of the reference removed from the list of warps.
Notes
-----
This method modifies the input list of warps in place by removing the
reference warp from it.
"""
# User a reference visit, if one has been supplied
# User-defined reference visit, if one has been supplied
if refWarpVisit:
warpVisits = [warp.dataId["visit"] for warp in warps]
warpVisits = [warpDDH.dataId["visit"] for warpDDH in warps]
try:
refWarp = warps.pop(warpVisits.index(refWarpVisit))
refWarpIndex = warpVisits.index(refWarpVisit)
refWarpDDH = warps.pop(refWarpIndex)
self.log.info("Using user-supplied reference visit %d", refWarpVisit)
return refWarp
return refWarpDDH.get(), refWarpIndex
except ValueError:
raise TaskError(f"Reference visit {refWarpVisit} is not found in the list of warps.")

# Extract mean/var/npoints for each warp
warpMeans = []
warpVars = []
warpNPoints = []
warpChi2s = [] # Background goodness of fit
warpVars = [] # Variance
warpNPointsGlobal = [] # Global coverage
warpNPointsEdge = [] # Edge coverage
for warpDDH in warps:
warp = warpDDH.get()

# First check if any image edge is all NaN
# If so, don't use
leftBool = np.all(np.isnan(warp.image.array[:, 0]))
rightBool = np.all(np.isnan(warp.image.array[:, warp.image.getHeight() - 1]))
bottomBool = np.all(np.isnan(warp.image.array[0, :]))
topBool = np.all(np.isnan(warp.image.array[warp.image.getWidth() - 1, :]))
if np.any([leftBool, rightBool, bottomBool, topBool]):
continue

warp.image.array *= warp.getPhotoCalib().instFluxToNanojansky(1) # Convert image plane to nJy

# Select reference based on BG of plane sky-subtracted images
bkgd, __, __, __ = self._setupBackground(warp)

weightByInverseVariance = self.config.approxWeighting
actrl = ApproximateControl(ApproximateControl.CHEBYSHEV, 1, 1, weightByInverseVariance)
undersampleStyle = stringToUndersampleStyle(self.config.undersampleStyle)
approx = bkgd.getApproximate(actrl, undersampleStyle)
bgSubImage = ImageF(warp.image.array - approx.getImage().array)

warpStats = makeStatistics(bgSubImage, warp.mask, MEAN | VARIANCE | NPOINT, self.statsCtrl)
warpMean, _ = warpStats.getResult(MEAN)
instFluxToNanojansky = warp.getPhotoCalib().instFluxToNanojansky(1)
warp.image *= instFluxToNanojansky # Images in nJy to facilitate difference imaging
warp.variance *= instFluxToNanojansky
warpBg, _ = self._makeBackground(warp)

# Return an approximation to the background
approxCtrl = ApproximateControl(ApproximateControl.CHEBYSHEV, 1, 1, self.config.approxWeighting)
warpApprox = warpBg.getApproximate(approxCtrl, self.undersampleStyle)
warpBgSub = ImageF(warp.image.array - warpApprox.getImage().array)

warpStats = makeStatistics(warpBgSub, warp.mask, VARIANCE | NPOINT, self.statsCtrl)
# TODO: need to modify this to account for the background mask
warpChi2 = np.nansum(warpBgSub.array**2 / warp.variance.array)
warpVar, _ = warpStats.getResult(VARIANCE)
warpNPoint, _ = warpStats.getResult(NPOINT)
warpMeans.append(warpMean)
warpNPointGlobal, _ = warpStats.getResult(NPOINT)
warpNPointEdge = (
np.sum(~np.isnan(warp.image.array[:, 0])) # Left edge
+ np.sum(~np.isnan(warp.image.array[:, -1])) # Right edge
+ np.sum(~np.isnan(warp.image.array[0, :])) # Bottom edge
+ np.sum(~np.isnan(warp.image.array[-1, :])) # Top edge
)
warpChi2s.append(warpChi2)
warpVars.append(warpVar)
warpNPoints.append(int(warpNPoint))

if len(warpNPoints) == 0:
raise TaskError("No suitable reference visit found in list of warps.")
warpNPointsGlobal.append(int(warpNPointGlobal))
warpNPointsEdge.append(warpNPointEdge)

# Normalize mean/var/npoints to range from 0 to 1
warpMeansFrac = np.array(warpMeans) / np.nanmax(warpMeans)
warpChi2sFrac = np.array(warpChi2s) / np.nanmax(warpChi2s)
warpVarsFrac = np.array(warpVars) / np.nanmax(warpVars)
warpNPointsFrac = np.nanmin(warpNPoints) / np.array(warpNPoints)
warpNPointsGlobalFrac = np.nanmin(warpNPointsGlobal) / np.array(warpNPointsGlobal)
warpNPointsEdgeFrac = np.nanmin(warpNPointsEdge) / np.array(warpNPointsEdge)

# Calculate cost function values
costFunctionVals = self.config.bestRefWeightLevel * warpMeansFrac
costFunctionVals = self.config.bestRefWeightChi2 * warpChi2sFrac
costFunctionVals += self.config.bestRefWeightVariance * warpVarsFrac
costFunctionVals += self.config.bestRefWeightCoverage * warpNPointsFrac
costFunctionVals += self.config.bestRefWeightGlobalCoverage * warpNPointsGlobalFrac
costFunctionVals += self.config.bestRefWeightEdgeCoverage * warpNPointsEdgeFrac

ind = np.nanargmin(costFunctionVals)
refWarp = warps.pop(ind)
self.log.info("Using best reference visit %d", refWarp.dataId["visit"])
return refWarp, ind

def _fluxScale(self, exposure):
"""Scales image to nJy flux using photometric calibration.
def _makeBackground(self, warp: ExposureF) -> tuple[BackgroundMI, BackgroundControl]:
"""Generate a simple binned background masked image for warped data.
Parameters
----------
exposure: `lsst.afw.image._exposure.ExposureF`
Exposure to scale.
warp: `~lsst.afw.image.ExposureF`
Warped exposure for which to estimate background.
Returns
-------
maskedImage: `lsst.afw.image._maskedImage.MaskedImageF`
Flux-scaled masked exposure.
warpBgMI: `~lsst.afw.math.BackgroundMI`
Background-subtracted masked image.
bgCtrl: `~lsst.afw.math.BackgroundControl`
Background control object.
"""
maskedImage = exposure.getMaskedImage()
fluxZp = exposure.getPhotoCalib().instFluxToNanojansky(1)
exposure.image.array *= fluxZp
nx = warp.getWidth() // self.config.binSize
ny = warp.getHeight() // self.config.binSize

return maskedImage
bgCtrl = BackgroundControl(nx, ny, self.statsCtrl, self.statsFlag)
bgCtrl.setUndersampleStyle(self.config.undersampleStyle)
warpBgMI = makeBackground(warp.getMaskedImage(), bgCtrl)

def _setupBackground(self, warp):
"""Set up and return a background model container and associated images
and controllers.
return warpBgMI, bgCtrl

Uses the following config parameters:
- ``gridStatistic``
- ``numSigmaClip``
- ``numIter``
- ``binSize``
- ``undersampleStyle``
def _fluxScale(self, exposure):
"""Scales image to nJy flux using photometric calibration.
Parameters
----------
warp: `lsst.afw.image._exposure.ExposureF`
Warped exposure or difference image for which to estimate
background.
exposure: `lsst.afw.image._exposure.ExposureF`
Exposure to scale.
Returns
-------
bkgd: `lsst.afw.math.BackgroundMI`
Background model container.
bctrl: `lsst.afw.math.BackgroundControl`
Background model control
warpMI: `lsst.afw.image._maskedImage.MaskedImageF`
Masked image associated with warp
statsFlag: `lsst.afw.math.Property`
Flag for grid statistic property (self.config.gridStatistic)
maskedImage: `lsst.afw.image._maskedImage.MaskedImageF`
Flux-scaled masked exposure.
"""
statsFlag = stringToStatisticsProperty(self.config.gridStatistic)
self.statsCtrl.setNumSigmaClip(self.config.numSigmaClip)
self.statsCtrl.setNumIter(self.config.numIter)

warpMI = warp.getMaskedImage()

width = warpMI.getWidth()
height = warpMI.getHeight()
nx = width // self.config.binSize
if width % self.config.binSize != 0:
nx += 1
ny = height // self.config.binSize
if height % self.config.binSize != 0:
ny += 1

bctrl = BackgroundControl(nx, ny, self.statsCtrl, statsFlag)
bctrl.setUndersampleStyle(self.config.undersampleStyle)

bkgd = makeBackground(warpMI, bctrl)
maskedImage = exposure.getMaskedImage()
fluxZp = exposure.getPhotoCalib().instFluxToNanojansky(1)
exposure.image.array *= fluxZp

return bkgd, bctrl, warpMI, statsFlag
return maskedImage

@timeMethod
def matchBackgrounds(self, refExposure, sciExposure):
Expand Down

0 comments on commit fea12d4

Please sign in to comment.