lsst.ip.diffim gbdffd78dd5+aceb78a6b6
dipoleFitTask.py
Go to the documentation of this file.
2# LSST Data Management System
3# Copyright 2008-2016 AURA/LSST.
4#
5# This product includes software developed by the
6# LSST Project (http://www.lsst.org/).
7#
8# This program is free software: you can redistribute it and/or modify
9# it under the terms of the GNU General Public License as published by
10# the Free Software Foundation, either version 3 of the License, or
11# (at your option) any later version.
12#
13# This program is distributed in the hope that it will be useful,
14# but WITHOUT ANY WARRANTY; without even the implied warranty of
15# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16# GNU General Public License for more details.
17#
18# You should have received a copy of the LSST License Statement and
19# the GNU General Public License along with this program. If not,
20# see <https://www.lsstcorp.org/LegalNotices/>.
21#
22
23import numpy as np
24import warnings
25
26import lsst.afw.image as afwImage
27import lsst.meas.base as measBase
28import lsst.afw.table as afwTable
29import lsst.afw.detection as afwDet
30import lsst.geom as geom
31from lsst.log import Log
32import lsst.pex.exceptions as pexExcept
33import lsst.pex.config as pexConfig
34from lsst.pipe.base import Struct, timeMethod
35
36__all__ = ("DipoleFitTask", "DipoleFitPlugin", "DipoleFitTaskConfig", "DipoleFitPluginConfig",
37 "DipoleFitAlgorithm")
38
39
40# Create a new measurement task (`DipoleFitTask`) that can handle all other SFM tasks but can
41# pass a separate pos- and neg- exposure/image to the `DipoleFitPlugin`s `run()` method.
42
43
44class DipoleFitPluginConfig(measBase.SingleFramePluginConfig):
45 """Configuration for DipoleFitPlugin
46 """
47
48 fitAllDiaSources = pexConfig.Field(
49 dtype=float, default=False,
50 doc="""Attempte dipole fit of all diaSources (otherwise just the ones consisting of overlapping
51 positive and negative footprints)""")
52
53 maxSeparation = pexConfig.Field(
54 dtype=float, default=5.,
55 doc="Assume dipole is not separated by more than maxSeparation * psfSigma")
56
57 relWeight = pexConfig.Field(
58 dtype=float, default=0.5,
59 doc="""Relative weighting of pre-subtraction images (higher -> greater influence of pre-sub.
60 images on fit)""")
61
62 tolerance = pexConfig.Field(
63 dtype=float, default=1e-7,
64 doc="Fit tolerance")
65
66 fitBackground = pexConfig.Field(
67 dtype=int, default=1,
68 doc="Set whether and how to fit for linear gradient in pre-sub. images. Possible values:"
69 "0: do not fit background at all"
70 "1 (default): pre-fit the background using linear least squares and then do not fit it as part"
71 "of the dipole fitting optimization"
72 "2: pre-fit the background using linear least squares (as in 1), and use the parameter"
73 "estimates from that fit as starting parameters for an integrated re-fit of the background")
74
75 fitSeparateNegParams = pexConfig.Field(
76 dtype=bool, default=False,
77 doc="Include parameters to fit for negative values (flux, gradient) separately from pos.")
78
79 # Config params for classification of detected diaSources as dipole or not
80 minSn = pexConfig.Field(
81 dtype=float, default=np.sqrt(2) * 5.0,
82 doc="Minimum quadrature sum of positive+negative lobe S/N to be considered a dipole")
83
84 maxFluxRatio = pexConfig.Field(
85 dtype=float, default=0.65,
86 doc="Maximum flux ratio in either lobe to be considered a dipole")
87
88 maxChi2DoF = pexConfig.Field(
89 dtype=float, default=0.05,
90 doc="""Maximum Chi2/DoF significance of fit to be considered a dipole.
91 Default value means \"Choose a chi2DoF corresponding to a significance level of at most 0.05\"
92 (note this is actually a significance, not a chi2 value).""")
93
94
95class DipoleFitTaskConfig(measBase.SingleFrameMeasurementConfig):
96 """Measurement of detected diaSources as dipoles
97
98 Currently we keep the "old" DipoleMeasurement algorithms turned on.
99 """
100
101 def setDefaults(self):
102 measBase.SingleFrameMeasurementConfig.setDefaults(self)
103
104 self.plugins.names = ["base_CircularApertureFlux",
105 "base_PixelFlags",
106 "base_SkyCoord",
107 "base_PsfFlux",
108 "base_SdssCentroid",
109 "base_SdssShape",
110 "base_GaussianFlux",
111 "base_PeakLikelihoodFlux",
112 "base_PeakCentroid",
113 "base_NaiveCentroid",
114 "ip_diffim_NaiveDipoleCentroid",
115 "ip_diffim_NaiveDipoleFlux",
116 "ip_diffim_PsfDipoleFlux",
117 "ip_diffim_ClassificationDipole",
118 ]
119
120 self.slots.calibFlux = None
121 self.slots.modelFlux = None
122 self.slots.gaussianFlux = None
123 self.slots.shape = "base_SdssShape"
124 self.slots.centroid = "ip_diffim_NaiveDipoleCentroid"
125 self.doReplaceWithNoisedoReplaceWithNoise = False
126
127
128class DipoleFitTask(measBase.SingleFrameMeasurementTask):
129 """A task that fits a dipole to a difference image, with an optional separate detection image.
130
131 Because it subclasses SingleFrameMeasurementTask, and calls
132 SingleFrameMeasurementTask.run() from its run() method, it still
133 can be used identically to a standard SingleFrameMeasurementTask.
134 """
135
136 ConfigClass = DipoleFitTaskConfig
137 _DefaultName = "ip_diffim_DipoleFit"
138
139 def __init__(self, schema, algMetadata=None, **kwargs):
140
141 measBase.SingleFrameMeasurementTask.__init__(self, schema, algMetadata, **kwargs)
142
143 dpFitPluginConfig = self.config.plugins['ip_diffim_DipoleFit']
144
145 self.dipoleFitterdipoleFitter = DipoleFitPlugin(dpFitPluginConfig, name=self._DefaultName_DefaultName,
146 schema=schema, metadata=algMetadata)
147
148 @timeMethod
149 def run(self, sources, exposure, posExp=None, negExp=None, **kwargs):
150 """Run dipole measurement and classification
151
152 Parameters
153 ----------
155 ``diaSources`` that will be measured using dipole measurement
156 exposure : `lsst.afw.image.Exposure`
157 The difference exposure on which the ``diaSources`` of the ``sources`` parameter
158 were detected. If neither ``posExp`` nor ``negExp`` are set, then the dipole is also
159 fitted directly to this difference image.
160 posExp : `lsst.afw.image.Exposure`, optional
161 "Positive" exposure, typically a science exposure, or None if unavailable
162 When `posExp` is `None`, will compute `posImage = exposure + negExp`.
163 negExp : `lsst.afw.image.Exposure`, optional
164 "Negative" exposure, typically a template exposure, or None if unavailable
165 When `negExp` is `None`, will compute `negImage = posExp - exposure`.
166 **kwargs
167 Additional keyword arguments for `lsst.meas.base.sfm.SingleFrameMeasurementTask`.
168 """
169
170 measBase.SingleFrameMeasurementTask.run(self, sources, exposure, **kwargs)
171
172 if not sources:
173 return
174
175 for source in sources:
176 self.dipoleFitterdipoleFitter.measure(source, exposure, posExp, negExp)
177
178
179class DipoleModel(object):
180 """Lightweight class containing methods for generating a dipole model for fitting
181 to sources in diffims, used by DipoleFitAlgorithm.
182
183 See also:
184 `DMTN-007: Dipole characterization for image differencing <https://dmtn-007.lsst.io>`_.
185 """
186
187 def __init__(self):
188 import lsstDebug
189 self.debugdebug = lsstDebug.Info(__name__).debug
190 self.loglog = Log.getLogger(__name__)
191
192 def makeBackgroundModel(self, in_x, pars=None):
193 """Generate gradient model (2-d array) with up to 2nd-order polynomial
194
195 Parameters
196 ----------
197 in_x : `numpy.array`
198 (2, w, h)-dimensional `numpy.array`, containing the
199 input x,y meshgrid providing the coordinates upon which to
200 compute the gradient. This will typically be generated via
201 `_generateXYGrid()`. `w` and `h` correspond to the width and
202 height of the desired grid.
203 pars : `list` of `float`, optional
204 Up to 6 floats for up
205 to 6 2nd-order 2-d polynomial gradient parameters, in the
206 following order: (intercept, x, y, xy, x**2, y**2). If `pars`
207 is emtpy or `None`, do nothing and return `None` (for speed).
208
209 Returns
210 -------
211 result : `None` or `numpy.array`
212 return None, or 2-d numpy.array of width/height matching
213 input bbox, containing computed gradient values.
214 """
215
216 # Don't fit for other gradient parameters if the intercept is not included.
217 if (pars is None) or (len(pars) <= 0) or (pars[0] is None):
218 return
219
220 y, x = in_x[0, :], in_x[1, :]
221 gradient = np.full_like(x, pars[0], dtype='float64')
222 if len(pars) > 1 and pars[1] is not None:
223 gradient += pars[1] * x
224 if len(pars) > 2 and pars[2] is not None:
225 gradient += pars[2] * y
226 if len(pars) > 3 and pars[3] is not None:
227 gradient += pars[3] * (x * y)
228 if len(pars) > 4 and pars[4] is not None:
229 gradient += pars[4] * (x * x)
230 if len(pars) > 5 and pars[5] is not None:
231 gradient += pars[5] * (y * y)
232
233 return gradient
234
235 def _generateXYGrid(self, bbox):
236 """Generate a meshgrid covering the x,y coordinates bounded by bbox
237
238 Parameters
239 ----------
240 bbox : `lsst.geom.Box2I`
241 input Bounding Box defining the coordinate limits
242
243 Returns
244 -------
245 in_x : `numpy.array`
246 (2, w, h)-dimensional numpy array containing the grid indexing over x- and
247 y- coordinates
248 """
249
250 x, y = np.mgrid[bbox.getBeginY():bbox.getEndY(), bbox.getBeginX():bbox.getEndX()]
251 in_x = np.array([y, x]).astype(np.float64)
252 in_x[0, :] -= np.mean(in_x[0, :])
253 in_x[1, :] -= np.mean(in_x[1, :])
254 return in_x
255
256 def _getHeavyFootprintSubimage(self, fp, badfill=np.nan, grow=0):
257 """Extract the image from a ``~lsst.afw.detection.HeavyFootprint``
258 as an `lsst.afw.image.ImageF`.
259
260 Parameters
261 ----------
263 HeavyFootprint to use to generate the subimage
264 badfill : `float`, optional
265 Value to fill in pixels in extracted image that are outside the footprint
266 grow : `int`
267 Optionally grow the footprint by this amount before extraction
268
269 Returns
270 -------
271 subim2 : `lsst.afw.image.ImageF`
272 An `~lsst.afw.image.ImageF` containing the subimage.
273 """
274 bbox = fp.getBBox()
275 if grow > 0:
276 bbox.grow(grow)
277
278 subim2 = afwImage.ImageF(bbox, badfill)
279 fp.getSpans().unflatten(subim2.getArray(), fp.getImageArray(), bbox.getCorners()[0])
280 return subim2
281
282 def fitFootprintBackground(self, source, posImage, order=1):
283 """Fit a linear (polynomial) model of given order (max 2) to the background of a footprint.
284
285 Only fit the pixels OUTSIDE of the footprint, but within its bounding box.
286
287 Parameters
288 ----------
290 SourceRecord, the footprint of which is to be fit
291 posImage : `lsst.afw.image.Exposure`
292 The exposure from which to extract the footprint subimage
293 order : `int`
294 Polynomial order of background gradient to fit.
295
296 Returns
297 -------
298 pars : `tuple` of `float`
299 `tuple` of length (1 if order==0; 3 if order==1; 6 if order == 2),
300 containing the resulting fit parameters
301 """
302
303 # TODO look into whether to use afwMath background methods -- see
304 # http://lsst-web.ncsa.illinois.edu/doxygen/x_masterDoxyDoc/_background_example.html
305 fp = source.getFootprint()
306 bbox = fp.getBBox()
307 bbox.grow(3)
308 posImg = afwImage.ImageF(posImage.getMaskedImage().getImage(), bbox, afwImage.PARENT)
309
310 # This code constructs the footprint image so that we can identify the pixels that are
311 # outside the footprint (but within the bounding box). These are the pixels used for
312 # fitting the background.
313 posHfp = afwDet.HeavyFootprintF(fp, posImage.getMaskedImage())
314 posFpImg = self._getHeavyFootprintSubimage_getHeavyFootprintSubimage(posHfp, grow=3)
315
316 isBg = np.isnan(posFpImg.getArray()).ravel()
317
318 data = posImg.getArray().ravel()
319 data = data[isBg]
320 B = data
321
322 x, y = np.mgrid[bbox.getBeginY():bbox.getEndY(), bbox.getBeginX():bbox.getEndX()]
323 x = x.astype(np.float64).ravel()
324 x -= np.mean(x)
325 x = x[isBg]
326 y = y.astype(np.float64).ravel()
327 y -= np.mean(y)
328 y = y[isBg]
329 b = np.ones_like(x, dtype=np.float64)
330
331 M = np.vstack([b]).T # order = 0
332 if order == 1:
333 M = np.vstack([b, x, y]).T
334 elif order == 2:
335 M = np.vstack([b, x, y, x**2., y**2., x*y]).T
336
337 pars = np.linalg.lstsq(M, B, rcond=-1)[0]
338 return pars
339
340 def makeStarModel(self, bbox, psf, xcen, ycen, flux):
341 """Generate a 2D image model of a single PDF centered at the given coordinates.
342
343 Parameters
344 ----------
345 bbox : `lsst.geom.Box`
346 Bounding box marking pixel coordinates for generated model
347 psf : TODO: DM-17458
348 Psf model used to generate the 'star'
349 xcen : `float`
350 Desired x-centroid of the 'star'
351 ycen : `float`
352 Desired y-centroid of the 'star'
353 flux : `float`
354 Desired flux of the 'star'
355
356 Returns
357 -------
358 p_Im : `lsst.afw.image.Image`
359 2-d stellar image of width/height matching input ``bbox``,
360 containing PSF with given centroid and flux
361 """
362
363 # Generate the psf image, normalize to flux
364 psf_img = psf.computeImage(geom.Point2D(xcen, ycen)).convertF()
365 psf_img_sum = np.nansum(psf_img.getArray())
366 psf_img *= (flux/psf_img_sum)
367
368 # Clip the PSF image bounding box to fall within the footprint bounding box
369 psf_box = psf_img.getBBox()
370 psf_box.clip(bbox)
371 psf_img = afwImage.ImageF(psf_img, psf_box, afwImage.PARENT)
372
373 # Then actually crop the psf image.
374 # Usually not necessary, but if the dipole is near the edge of the image...
375 # Would be nice if we could compare original pos_box with clipped pos_box and
376 # see if it actually was clipped.
377 p_Im = afwImage.ImageF(bbox)
378 tmpSubim = afwImage.ImageF(p_Im, psf_box, afwImage.PARENT)
379 tmpSubim += psf_img
380
381 return p_Im
382
383 def makeModel(self, x, flux, xcenPos, ycenPos, xcenNeg, ycenNeg, fluxNeg=None,
384 b=None, x1=None, y1=None, xy=None, x2=None, y2=None,
385 bNeg=None, x1Neg=None, y1Neg=None, xyNeg=None, x2Neg=None, y2Neg=None,
386 **kwargs):
387 """Generate dipole model with given parameters.
388
389 This is the function whose sum-of-squared difference from data
390 is minimized by `lmfit`.
391
392 x : TODO: DM-17458
393 Input independent variable. Used here as the grid on
394 which to compute the background gradient model.
395 flux : `float`
396 Desired flux of the positive lobe of the dipole
397 xcenPos : `float`
398 Desired x-centroid of the positive lobe of the dipole
399 ycenPos : `float`
400 Desired y-centroid of the positive lobe of the dipole
401 xcenNeg : `float`
402 Desired x-centroid of the negative lobe of the dipole
403 ycenNeg : `float`
404 Desired y-centroid of the negative lobe of the dipole
405 fluxNeg : `float`, optional
406 Desired flux of the negative lobe of the dipole, set to 'flux' if None
407 b, x1, y1, xy, x2, y2 : `float`
408 Gradient parameters for positive lobe.
409 bNeg, x1Neg, y1Neg, xyNeg, x2Neg, y2Neg : `float`, optional
410 Gradient parameters for negative lobe.
411 They are set to the corresponding positive values if None.
412
413 **kwargs
414 Keyword arguments passed through ``lmfit`` and
415 used by this function. These must include:
416
417 - ``psf`` Psf model used to generate the 'star'
418 - ``rel_weight`` Used to signify least-squares weighting of posImage/negImage
419 relative to diffim. If ``rel_weight == 0`` then posImage/negImage are ignored.
420 - ``bbox`` Bounding box containing region to be modelled
421
422 Returns
423 -------
424 zout : `numpy.array`
425 Has width and height matching the input bbox, and
426 contains the dipole model with given centroids and flux(es). If
427 ``rel_weight`` = 0, this is a 2-d array with dimensions matching
428 those of bbox; otherwise a stack of three such arrays,
429 representing the dipole (diffim), positive and negative images
430 respectively.
431 """
432
433 psf = kwargs.get('psf')
434 rel_weight = kwargs.get('rel_weight') # if > 0, we're including pre-sub. images
435 fp = kwargs.get('footprint')
436 bbox = fp.getBBox()
437
438 if fluxNeg is None:
439 fluxNeg = flux
440
441 if self.debugdebug:
442 self.loglog.debug('%.2f %.2f %.2f %.2f %.2f %.2f',
443 flux, fluxNeg, xcenPos, ycenPos, xcenNeg, ycenNeg)
444 if x1 is not None:
445 self.loglog.debug(' %.2f %.2f %.2f', b, x1, y1)
446 if xy is not None:
447 self.loglog.debug(' %.2f %.2f %.2f', xy, x2, y2)
448
449 posIm = self.makeStarModelmakeStarModel(bbox, psf, xcenPos, ycenPos, flux)
450 negIm = self.makeStarModelmakeStarModel(bbox, psf, xcenNeg, ycenNeg, fluxNeg)
451
452 in_x = x
453 if in_x is None: # use the footprint to generate the input grid
454 y, x = np.mgrid[bbox.getBeginY():bbox.getEndY(), bbox.getBeginX():bbox.getEndX()]
455 in_x = np.array([x, y]) * 1.
456 in_x[0, :] -= in_x[0, :].mean() # center it!
457 in_x[1, :] -= in_x[1, :].mean()
458
459 if b is not None:
460 gradient = self.makeBackgroundModelmakeBackgroundModel(in_x, (b, x1, y1, xy, x2, y2))
461
462 # If bNeg is None, then don't fit the negative background separately
463 if bNeg is not None:
464 gradientNeg = self.makeBackgroundModelmakeBackgroundModel(in_x, (bNeg, x1Neg, y1Neg, xyNeg, x2Neg, y2Neg))
465 else:
466 gradientNeg = gradient
467
468 posIm.getArray()[:, :] += gradient
469 negIm.getArray()[:, :] += gradientNeg
470
471 # Generate the diffIm model
472 diffIm = afwImage.ImageF(bbox)
473 diffIm += posIm
474 diffIm -= negIm
475
476 zout = diffIm.getArray()
477 if rel_weight > 0.:
478 zout = np.append([zout], [posIm.getArray(), negIm.getArray()], axis=0)
479
480 return zout
481
482
483class DipoleFitAlgorithm(object):
484 """Fit a dipole model using an image difference.
485
486 See also:
487 `DMTN-007: Dipole characterization for image differencing <https://dmtn-007.lsst.io>`_.
488 """
489
490 # This is just a private version number to sync with the ipython notebooks that I have been
491 # using for algorithm development.
492 _private_version_ = '0.0.5'
493
494 # Below is a (somewhat incomplete) list of improvements
495 # that would be worth investigating, given the time:
496
497 # todo 1. evaluate necessity for separate parameters for pos- and neg- images
498 # todo 2. only fit background OUTSIDE footprint (DONE) and dipole params INSIDE footprint (NOT DONE)?
499 # todo 3. correct normalization of least-squares weights based on variance planes
500 # todo 4. account for PSFs that vary across the exposures (should be happening by default?)
501 # todo 5. correctly account for NA/masks (i.e., ignore!)
502 # todo 6. better exception handling in the plugin
503 # todo 7. better classification of dipoles (e.g. by comparing chi2 fit vs. monopole?)
504 # todo 8. (DONE) Initial fast estimate of background gradient(s) params -- perhaps using numpy.lstsq
505 # todo 9. (NOT NEEDED - see (2)) Initial fast test whether a background gradient needs to be fit
506 # todo 10. (DONE) better initial estimate for flux when there's a strong gradient
507 # todo 11. (DONE) requires a new package `lmfit` -- investiate others? (astropy/scipy/iminuit?)
508
509 def __init__(self, diffim, posImage=None, negImage=None):
510 """Algorithm to run dipole measurement on a diaSource
511
512 Parameters
513 ----------
514 diffim : `lsst.afw.image.Exposure`
515 Exposure on which the diaSources were detected
516 posImage : `lsst.afw.image.Exposure`
517 "Positive" exposure from which the template was subtracted
518 negImage : `lsst.afw.image.Exposure`
519 "Negative" exposure which was subtracted from the posImage
520 """
521
522 self.diffimdiffim = diffim
523 self.posImageposImage = posImage
524 self.negImagenegImage = negImage
525 self.psfSigmapsfSigma = None
526 if diffim is not None:
527 self.psfSigmapsfSigma = diffim.getPsf().computeShape().getDeterminantRadius()
528
529 self.loglog = Log.getLogger(__name__)
530
531 import lsstDebug
532 self.debugdebug = lsstDebug.Info(__name__).debug
533
534 def fitDipoleImpl(self, source, tol=1e-7, rel_weight=0.5,
535 fitBackground=1, bgGradientOrder=1, maxSepInSigma=5.,
536 separateNegParams=True, verbose=False):
537 """Fit a dipole model to an input difference image.
538
539 Actually, fits the subimage bounded by the input source's
540 footprint) and optionally constrain the fit using the
541 pre-subtraction images posImage and negImage.
542
543 Parameters
544 ----------
545 source : TODO: DM-17458
546 TODO: DM-17458
547 tol : float, optional
548 TODO: DM-17458
549 rel_weight : `float`, optional
550 TODO: DM-17458
551 fitBackground : `int`, optional
552 TODO: DM-17458
553 bgGradientOrder : `int`, optional
554 TODO: DM-17458
555 maxSepInSigma : `float`, optional
556 TODO: DM-17458
557 separateNegParams : `bool`, optional
558 TODO: DM-17458
559 verbose : `bool`, optional
560 TODO: DM-17458
561
562 Returns
563 -------
564 result : `lmfit.MinimizerResult`
565 return `lmfit.MinimizerResult` object containing the fit
566 parameters and other information.
567 """
568
569 # Only import lmfit if someone wants to use the new DipoleFitAlgorithm.
570 import lmfit
571
572 fp = source.getFootprint()
573 bbox = fp.getBBox()
574 subim = afwImage.MaskedImageF(self.diffimdiffim.getMaskedImage(), bbox=bbox, origin=afwImage.PARENT)
575
576 z = diArr = subim.getArrays()[0]
577 weights = 1. / subim.getArrays()[2] # get the weights (=1/variance)
578
579 if rel_weight > 0. and ((self.posImageposImage is not None) or (self.negImagenegImage is not None)):
580 if self.negImagenegImage is not None:
581 negSubim = afwImage.MaskedImageF(self.negImagenegImage.getMaskedImage(), bbox, origin=afwImage.PARENT)
582 if self.posImageposImage is not None:
583 posSubim = afwImage.MaskedImageF(self.posImageposImage.getMaskedImage(), bbox, origin=afwImage.PARENT)
584 if self.posImageposImage is None: # no science image provided; generate it from diffim + negImage
585 posSubim = subim.clone()
586 posSubim += negSubim
587 if self.negImagenegImage is None: # no template provided; generate it from the posImage - diffim
588 negSubim = posSubim.clone()
589 negSubim -= subim
590
591 z = np.append([z], [posSubim.getArrays()[0],
592 negSubim.getArrays()[0]], axis=0)
593 # Weight the pos/neg images by rel_weight relative to the diffim
594 weights = np.append([weights], [1. / posSubim.getArrays()[2] * rel_weight,
595 1. / negSubim.getArrays()[2] * rel_weight], axis=0)
596 else:
597 rel_weight = 0. # a short-cut for "don't include the pre-subtraction data"
598
599 # It seems that `lmfit` requires a static functor as its optimized method, which eliminates
600 # the ability to pass a bound method or other class method. Here we write a wrapper which
601 # makes this possible.
602 def dipoleModelFunctor(x, flux, xcenPos, ycenPos, xcenNeg, ycenNeg, fluxNeg=None,
603 b=None, x1=None, y1=None, xy=None, x2=None, y2=None,
604 bNeg=None, x1Neg=None, y1Neg=None, xyNeg=None, x2Neg=None, y2Neg=None,
605 **kwargs):
606 """Generate dipole model with given parameters.
607
608 It simply defers to `modelObj.makeModel()`, where `modelObj` comes
609 out of `kwargs['modelObj']`.
610 """
611 modelObj = kwargs.pop('modelObj')
612 return modelObj.makeModel(x, flux, xcenPos, ycenPos, xcenNeg, ycenNeg, fluxNeg=fluxNeg,
613 b=b, x1=x1, y1=y1, xy=xy, x2=x2, y2=y2,
614 bNeg=bNeg, x1Neg=x1Neg, y1Neg=y1Neg, xyNeg=xyNeg,
615 x2Neg=x2Neg, y2Neg=y2Neg, **kwargs)
616
617 dipoleModel = DipoleModel()
618
619 modelFunctor = dipoleModelFunctor # dipoleModel.makeModel does not work for now.
620 # Create the lmfit model (lmfit uses scipy 'leastsq' option by default - Levenberg-Marquardt)
621 # Note we can also tell it to drop missing values from the data.
622 gmod = lmfit.Model(modelFunctor, verbose=verbose, missing='drop')
623 # independent_vars=independent_vars) #, param_names=param_names)
624
625 # Add the constraints for centroids, fluxes.
626 # starting constraint - near centroid of footprint
627 fpCentroid = np.array([fp.getCentroid().getX(), fp.getCentroid().getY()])
628 cenNeg = cenPos = fpCentroid
629
630 pks = fp.getPeaks()
631
632 if len(pks) >= 1:
633 cenPos = pks[0].getF() # if individual (merged) peaks were detected, use those
634 if len(pks) >= 2: # peaks are already sorted by centroid flux so take the most negative one
635 cenNeg = pks[-1].getF()
636
637 # For close/faint dipoles the starting locs (min/max) might be way off, let's help them a bit.
638 # First assume dipole is not separated by more than 5*psfSigma.
639 maxSep = self.psfSigmapsfSigma * maxSepInSigma
640
641 # As an initial guess -- assume the dipole is close to the center of the footprint.
642 if np.sum(np.sqrt((np.array(cenPos) - fpCentroid)**2.)) > maxSep:
643 cenPos = fpCentroid
644 if np.sum(np.sqrt((np.array(cenNeg) - fpCentroid)**2.)) > maxSep:
645 cenPos = fpCentroid
646
647 # parameter hints/constraints: https://lmfit.github.io/lmfit-py/model.html#model-param-hints-section
648 # might make sense to not use bounds -- see http://lmfit.github.io/lmfit-py/bounds.html
649 # also see this discussion -- https://github.com/scipy/scipy/issues/3129
650 gmod.set_param_hint('xcenPos', value=cenPos[0],
651 min=cenPos[0]-maxSep, max=cenPos[0]+maxSep)
652 gmod.set_param_hint('ycenPos', value=cenPos[1],
653 min=cenPos[1]-maxSep, max=cenPos[1]+maxSep)
654 gmod.set_param_hint('xcenNeg', value=cenNeg[0],
655 min=cenNeg[0]-maxSep, max=cenNeg[0]+maxSep)
656 gmod.set_param_hint('ycenNeg', value=cenNeg[1],
657 min=cenNeg[1]-maxSep, max=cenNeg[1]+maxSep)
658
659 # Use the (flux under the dipole)*5 for an estimate.
660 # Lots of testing showed that having startingFlux be too high was better than too low.
661 startingFlux = np.nansum(np.abs(diArr) - np.nanmedian(np.abs(diArr))) * 5.
662 posFlux = negFlux = startingFlux
663
664 # TBD: set max. flux limit?
665 gmod.set_param_hint('flux', value=posFlux, min=0.1)
666
667 if separateNegParams:
668 # TBD: set max negative lobe flux limit?
669 gmod.set_param_hint('fluxNeg', value=np.abs(negFlux), min=0.1)
670
671 # Fixed parameters (don't fit for them if there are no pre-sub images or no gradient fit requested):
672 # Right now (fitBackground == 1), we fit a linear model to the background and then subtract
673 # it from the data and then don't fit the background again (this is faster).
674 # A slower alternative (fitBackground == 2) is to use the estimated background parameters as
675 # starting points in the integrated model fit. That is currently not performed by default,
676 # but might be desirable in some cases.
677 bgParsPos = bgParsNeg = (0., 0., 0.)
678 if ((rel_weight > 0.) and (fitBackground != 0) and (bgGradientOrder >= 0)):
679 pbg = 0.
680 bgFitImage = self.posImageposImage if self.posImageposImage is not None else self.negImagenegImage
681 # Fit the gradient to the background (linear model)
682 bgParsPos = bgParsNeg = dipoleModel.fitFootprintBackground(source, bgFitImage,
683 order=bgGradientOrder)
684
685 # Generate the gradient and subtract it from the pre-subtraction image data
686 if fitBackground == 1:
687 in_x = dipoleModel._generateXYGrid(bbox)
688 pbg = dipoleModel.makeBackgroundModel(in_x, tuple(bgParsPos))
689 z[1, :] -= pbg
690 z[1, :] -= np.nanmedian(z[1, :])
691 posFlux = np.nansum(z[1, :])
692 gmod.set_param_hint('flux', value=posFlux*1.5, min=0.1)
693
694 if separateNegParams and self.negImagenegImage is not None:
695 bgParsNeg = dipoleModel.fitFootprintBackground(source, self.negImagenegImage,
696 order=bgGradientOrder)
697 pbg = dipoleModel.makeBackgroundModel(in_x, tuple(bgParsNeg))
698 z[2, :] -= pbg
699 z[2, :] -= np.nanmedian(z[2, :])
700 if separateNegParams:
701 negFlux = np.nansum(z[2, :])
702 gmod.set_param_hint('fluxNeg', value=negFlux*1.5, min=0.1)
703
704 # Do not subtract the background from the images but include the background parameters in the fit
705 if fitBackground == 2:
706 if bgGradientOrder >= 0:
707 gmod.set_param_hint('b', value=bgParsPos[0])
708 if separateNegParams:
709 gmod.set_param_hint('bNeg', value=bgParsNeg[0])
710 if bgGradientOrder >= 1:
711 gmod.set_param_hint('x1', value=bgParsPos[1])
712 gmod.set_param_hint('y1', value=bgParsPos[2])
713 if separateNegParams:
714 gmod.set_param_hint('x1Neg', value=bgParsNeg[1])
715 gmod.set_param_hint('y1Neg', value=bgParsNeg[2])
716 if bgGradientOrder >= 2:
717 gmod.set_param_hint('xy', value=bgParsPos[3])
718 gmod.set_param_hint('x2', value=bgParsPos[4])
719 gmod.set_param_hint('y2', value=bgParsPos[5])
720 if separateNegParams:
721 gmod.set_param_hint('xyNeg', value=bgParsNeg[3])
722 gmod.set_param_hint('x2Neg', value=bgParsNeg[4])
723 gmod.set_param_hint('y2Neg', value=bgParsNeg[5])
724
725 y, x = np.mgrid[bbox.getBeginY():bbox.getEndY(), bbox.getBeginX():bbox.getEndX()]
726 in_x = np.array([x, y]).astype(np.float)
727 in_x[0, :] -= in_x[0, :].mean() # center it!
728 in_x[1, :] -= in_x[1, :].mean()
729
730 # Instead of explicitly using a mask to ignore flagged pixels, just set the ignored pixels'
731 # weights to 0 in the fit. TBD: need to inspect mask planes to set this mask.
732 mask = np.ones_like(z, dtype=bool) # TBD: set mask values to False if the pixels are to be ignored
733
734 # I'm not sure about the variance planes in the diffim (or convolved pre-sub. images
735 # for that matter) so for now, let's just do an un-weighted least-squares fit
736 # (override weights computed above).
737 weights = mask.astype(np.float64)
738 if self.posImageposImage is not None and rel_weight > 0.:
739 weights = np.array([np.ones_like(diArr), np.ones_like(diArr)*rel_weight,
740 np.ones_like(diArr)*rel_weight])
741
742 # Set the weights to zero if mask is False
743 if np.any(~mask):
744 weights[~mask] = 0.
745
746 # Note that although we can, we're not required to set initial values for params here,
747 # since we set their param_hint's above.
748 # Can add "method" param to not use 'leastsq' (==levenberg-marquardt), e.g. "method='nelder'"
749 with warnings.catch_warnings():
750 warnings.simplefilter("ignore") # temporarily turn off silly lmfit warnings
751 result = gmod.fit(z, weights=weights, x=in_x,
752 verbose=verbose,
753 fit_kws={'ftol': tol, 'xtol': tol, 'gtol': tol,
754 'maxfev': 250}, # see scipy docs
755 psf=self.diffimdiffim.getPsf(), # hereon: kwargs that get passed to genDipoleModel()
756 rel_weight=rel_weight,
757 footprint=fp,
758 modelObj=dipoleModel)
759
760 if verbose: # the ci_report() seems to fail if neg params are constrained -- TBD why.
761 # Never wanted in production - this takes a long time (longer than the fit!)
762 # This is how to get confidence intervals out:
763 # https://lmfit.github.io/lmfit-py/confidence.html and
764 # http://cars9.uchicago.edu/software/python/lmfit/model.html
765 print(result.fit_report(show_correl=False))
766 if separateNegParams:
767 print(result.ci_report())
768
769 return result
770
771 def fitDipole(self, source, tol=1e-7, rel_weight=0.1,
772 fitBackground=1, maxSepInSigma=5., separateNegParams=True,
773 bgGradientOrder=1, verbose=False, display=False):
774 """Fit a dipole model to an input ``diaSource`` (wraps `fitDipoleImpl`).
775
776 Actually, fits the subimage bounded by the input source's
777 footprint) and optionally constrain the fit using the
778 pre-subtraction images self.posImageposImage (science) and
779 self.negImagenegImage (template). Wraps the output into a
780 `pipeBase.Struct` named tuple after computing additional
781 statistics such as orientation and SNR.
782
783 Parameters
784 ----------
786 Record containing the (merged) dipole source footprint detected on the diffim
787 tol : `float`, optional
788 Tolerance parameter for scipy.leastsq() optimization
789 rel_weight : `float`, optional
790 Weighting of posImage/negImage relative to the diffim in the fit
791 fitBackground : `int`, {0, 1, 2}, optional
792 How to fit linear background gradient in posImage/negImage
793
794 - 0: do not fit background at all
795 - 1 (default): pre-fit the background using linear least squares and then do not fit it
796 as part of the dipole fitting optimization
797 - 2: pre-fit the background using linear least squares (as in 1), and use the parameter
798 estimates from that fit as starting parameters for an integrated "re-fit" of the
799 background as part of the overall dipole fitting optimization.
800 maxSepInSigma : `float`, optional
801 Allowed window of centroid parameters relative to peak in input source footprint
802 separateNegParams : `bool`, optional
803 Fit separate parameters to the flux and background gradient in
804 bgGradientOrder : `int`, {0, 1, 2}, optional
805 Desired polynomial order of background gradient
806 verbose: `bool`, optional
807 Be verbose
808 display
809 Display input data, best fit model(s) and residuals in a matplotlib window.
810
811 Returns
812 -------
813 result : `struct`
814 `pipeBase.Struct` object containing the fit parameters and other information.
815
816 result : `callable`
817 `lmfit.MinimizerResult` object for debugging and error estimation, etc.
818
819 Notes
820 -----
821 Parameter `fitBackground` has three options, thus it is an integer:
822
823 """
824
825 fitResult = self.fitDipoleImplfitDipoleImpl(
826 source, tol=tol, rel_weight=rel_weight, fitBackground=fitBackground,
827 maxSepInSigma=maxSepInSigma, separateNegParams=separateNegParams,
828 bgGradientOrder=bgGradientOrder, verbose=verbose)
829
830 # Display images, model fits and residuals (currently uses matplotlib display functions)
831 if display:
832 fp = source.getFootprint()
833 self.displayFitResultsdisplayFitResults(fp, fitResult)
834
835 fitParams = fitResult.best_values
836 if fitParams['flux'] <= 1.: # usually around 0.1 -- the minimum flux allowed -- i.e. bad fit.
837 out = Struct(posCentroidX=np.nan, posCentroidY=np.nan,
838 negCentroidX=np.nan, negCentroidY=np.nan,
839 posFlux=np.nan, negFlux=np.nan, posFluxErr=np.nan, negFluxErr=np.nan,
840 centroidX=np.nan, centroidY=np.nan, orientation=np.nan,
841 signalToNoise=np.nan, chi2=np.nan, redChi2=np.nan)
842 return out, fitResult
843
844 centroid = ((fitParams['xcenPos'] + fitParams['xcenNeg']) / 2.,
845 (fitParams['ycenPos'] + fitParams['ycenNeg']) / 2.)
846 dx, dy = fitParams['xcenPos'] - fitParams['xcenNeg'], fitParams['ycenPos'] - fitParams['ycenNeg']
847 angle = np.arctan2(dy, dx) / np.pi * 180. # convert to degrees (should keep as rad?)
848
849 # Exctract flux value, compute signalToNoise from flux/variance_within_footprint
850 # Also extract the stderr of flux estimate.
851 def computeSumVariance(exposure, footprint):
852 box = footprint.getBBox()
853 subim = afwImage.MaskedImageF(exposure.getMaskedImage(), box, origin=afwImage.PARENT)
854 return np.sqrt(np.nansum(subim.getArrays()[1][:, :]))
855
856 fluxVal = fluxVar = fitParams['flux']
857 fluxErr = fluxErrNeg = fitResult.params['flux'].stderr
858 if self.posImageposImage is not None:
859 fluxVar = computeSumVariance(self.posImageposImage, source.getFootprint())
860 else:
861 fluxVar = computeSumVariance(self.diffimdiffim, source.getFootprint())
862
863 fluxValNeg, fluxVarNeg = fluxVal, fluxVar
864 if separateNegParams:
865 fluxValNeg = fitParams['fluxNeg']
866 fluxErrNeg = fitResult.params['fluxNeg'].stderr
867 if self.negImagenegImage is not None:
868 fluxVarNeg = computeSumVariance(self.negImagenegImage, source.getFootprint())
869
870 try:
871 signalToNoise = np.sqrt((fluxVal/fluxVar)**2 + (fluxValNeg/fluxVarNeg)**2)
872 except ZeroDivisionError: # catch divide by zero - should never happen.
873 signalToNoise = np.nan
874
875 out = Struct(posCentroidX=fitParams['xcenPos'], posCentroidY=fitParams['ycenPos'],
876 negCentroidX=fitParams['xcenNeg'], negCentroidY=fitParams['ycenNeg'],
877 posFlux=fluxVal, negFlux=-fluxValNeg, posFluxErr=fluxErr, negFluxErr=fluxErrNeg,
878 centroidX=centroid[0], centroidY=centroid[1], orientation=angle,
879 signalToNoise=signalToNoise, chi2=fitResult.chisqr, redChi2=fitResult.redchi)
880
881 # fitResult may be returned for debugging
882 return out, fitResult
883
884 def displayFitResults(self, footprint, result):
885 """Display data, model fits and residuals (currently uses matplotlib display functions).
886
887 Parameters
888 ----------
889 footprint : TODO: DM-17458
890 Footprint containing the dipole that was fit
891 result : `lmfit.MinimizerResult`
892 `lmfit.MinimizerResult` object returned by `lmfit` optimizer
893
894 Returns
895 -------
896 fig : `matplotlib.pyplot.plot`
897 """
898 try:
899 import matplotlib.pyplot as plt
900 except ImportError as err:
901 self.loglog.warning('Unable to import matplotlib: %s', err)
902 raise err
903
904 def display2dArray(arr, title='Data', extent=None):
905 """Use `matplotlib.pyplot.imshow` to display a 2-D array with a given coordinate range.
906 """
907 fig = plt.imshow(arr, origin='lower', interpolation='none', cmap='gray', extent=extent)
908 plt.title(title)
909 plt.colorbar(fig, cmap='gray')
910 return fig
911
912 z = result.data
913 fit = result.best_fit
914 bbox = footprint.getBBox()
915 extent = (bbox.getBeginX(), bbox.getEndX(), bbox.getBeginY(), bbox.getEndY())
916 if z.shape[0] == 3:
917 fig = plt.figure(figsize=(8, 8))
918 for i in range(3):
919 plt.subplot(3, 3, i*3+1)
920 display2dArray(z[i, :], 'Data', extent=extent)
921 plt.subplot(3, 3, i*3+2)
922 display2dArray(fit[i, :], 'Model', extent=extent)
923 plt.subplot(3, 3, i*3+3)
924 display2dArray(z[i, :] - fit[i, :], 'Residual', extent=extent)
925 return fig
926 else:
927 fig = plt.figure(figsize=(8, 2.5))
928 plt.subplot(1, 3, 1)
929 display2dArray(z, 'Data', extent=extent)
930 plt.subplot(1, 3, 2)
931 display2dArray(fit, 'Model', extent=extent)
932 plt.subplot(1, 3, 3)
933 display2dArray(z - fit, 'Residual', extent=extent)
934 return fig
935
936 plt.show()
937
938
939@measBase.register("ip_diffim_DipoleFit")
940class DipoleFitPlugin(measBase.SingleFramePlugin):
941 """A single frame measurement plugin that fits dipoles to all merged (two-peak) ``diaSources``.
942
943 This measurement plugin accepts up to three input images in
944 its `measure` method. If these are provided, it includes data
945 from the pre-subtraction posImage (science image) and optionally
946 negImage (template image) to constrain the fit. The meat of the
947 fitting routines are in the class `~lsst.module.name.DipoleFitAlgorithm`.
948
949 Notes
950 -----
951 The motivation behind this plugin and the necessity for including more than
952 one exposure are documented in DMTN-007 (http://dmtn-007.lsst.io).
953
954 This class is named `ip_diffim_DipoleFit` so that it may be used alongside
955 the existing `ip_diffim_DipoleMeasurement` classes until such a time as those
956 are deemed to be replaceable by this.
957 """
958
959 ConfigClass = DipoleFitPluginConfig
960 DipoleFitAlgorithmClass = DipoleFitAlgorithm # Pointer to the class that performs the fit
961
962 FAILURE_EDGE = 1 # too close to the edge
963 FAILURE_FIT = 2 # failure in the fitting
964 FAILURE_NOT_DIPOLE = 4 # input source is not a putative dipole to begin with
965
966 @classmethod
968 """Set execution order to `FLUX_ORDER`.
969
970 This includes algorithms that require both `getShape()` and `getCentroid()`,
971 in addition to a Footprint and its Peaks.
972 """
973 return cls.FLUX_ORDER
974
975 def __init__(self, config, name, schema, metadata):
976 measBase.SingleFramePlugin.__init__(self, config, name, schema, metadata)
977
978 self.loglog = Log.getLogger(name)
979
980 self._setupSchema_setupSchema(config, name, schema, metadata)
981
982 def _setupSchema(self, config, name, schema, metadata):
983 # Get a FunctorKey that can quickly look up the "blessed" centroid value.
984 self.centroidKeycentroidKey = afwTable.Point2DKey(schema["slot_Centroid"])
985
986 # Add some fields for our outputs, and save their Keys.
987 # Use setattr() to programmatically set the pos/neg named attributes to values, e.g.
988 # self.posCentroidKeyX = 'ip_diffim_DipoleFit_pos_centroid_x'
989
990 for pos_neg in ['pos', 'neg']:
991
992 key = schema.addField(
993 schema.join(name, pos_neg, "instFlux"), type=float, units="count",
994 doc="Dipole {0} lobe flux".format(pos_neg))
995 setattr(self, ''.join((pos_neg, 'FluxKey')), key)
996
997 key = schema.addField(
998 schema.join(name, pos_neg, "instFluxErr"), type=float, units="count",
999 doc="1-sigma uncertainty for {0} dipole flux".format(pos_neg))
1000 setattr(self, ''.join((pos_neg, 'FluxErrKey')), key)
1001
1002 for x_y in ['x', 'y']:
1003 key = schema.addField(
1004 schema.join(name, pos_neg, "centroid", x_y), type=float, units="pixel",
1005 doc="Dipole {0} lobe centroid".format(pos_neg))
1006 setattr(self, ''.join((pos_neg, 'CentroidKey', x_y.upper())), key)
1007
1008 for x_y in ['x', 'y']:
1009 key = schema.addField(
1010 schema.join(name, "centroid", x_y), type=float, units="pixel",
1011 doc="Dipole centroid")
1012 setattr(self, ''.join(('centroidKey', x_y.upper())), key)
1013
1014 self.fluxKeyfluxKey = schema.addField(
1015 schema.join(name, "instFlux"), type=float, units="count",
1016 doc="Dipole overall flux")
1017
1018 self.orientationKeyorientationKey = schema.addField(
1019 schema.join(name, "orientation"), type=float, units="deg",
1020 doc="Dipole orientation")
1021
1022 self.separationKeyseparationKey = schema.addField(
1023 schema.join(name, "separation"), type=float, units="pixel",
1024 doc="Pixel separation between positive and negative lobes of dipole")
1025
1026 self.chi2dofKeychi2dofKey = schema.addField(
1027 schema.join(name, "chi2dof"), type=float,
1028 doc="Chi2 per degree of freedom of dipole fit")
1029
1030 self.signalToNoiseKeysignalToNoiseKey = schema.addField(
1031 schema.join(name, "signalToNoise"), type=float,
1032 doc="Estimated signal-to-noise of dipole fit")
1033
1034 self.classificationFlagKeyclassificationFlagKey = schema.addField(
1035 schema.join(name, "flag", "classification"), type="Flag",
1036 doc="Flag indicating diaSource is classified as a dipole")
1037
1038 self.classificationAttemptedFlagKeyclassificationAttemptedFlagKey = schema.addField(
1039 schema.join(name, "flag", "classificationAttempted"), type="Flag",
1040 doc="Flag indicating diaSource was attempted to be classified as a dipole")
1041
1042 self.flagKeyflagKey = schema.addField(
1043 schema.join(name, "flag"), type="Flag",
1044 doc="General failure flag for dipole fit")
1045
1046 self.edgeFlagKeyedgeFlagKey = schema.addField(
1047 schema.join(name, "flag", "edge"), type="Flag",
1048 doc="Flag set when dipole is too close to edge of image")
1049
1050 def measure(self, measRecord, exposure, posExp=None, negExp=None):
1051 """Perform the non-linear least squares minimization on the putative dipole source.
1052
1053 Parameters
1054 ----------
1055 measRecord : `lsst.afw.table.SourceRecord`
1056 diaSources that will be measured using dipole measurement
1057 exposure : `lsst.afw.image.Exposure`
1058 Difference exposure on which the diaSources were detected; `exposure = posExp-negExp`
1059 If both `posExp` and `negExp` are `None`, will attempt to fit the
1060 dipole to just the `exposure` with no constraint.
1061 posExp : `lsst.afw.image.Exposure`, optional
1062 "Positive" exposure, typically a science exposure, or None if unavailable
1063 When `posExp` is `None`, will compute `posImage = exposure + negExp`.
1064 negExp : `lsst.afw.image.Exposure`, optional
1065 "Negative" exposure, typically a template exposure, or None if unavailable
1066 When `negExp` is `None`, will compute `negImage = posExp - exposure`.
1067
1068 Notes
1069 -----
1070 The main functionality of this routine was placed outside of
1071 this plugin (into `DipoleFitAlgorithm.fitDipole()`) so that
1072 `DipoleFitAlgorithm.fitDipole()` can be called separately for
1073 testing (@see `tests/testDipoleFitter.py`)
1074
1075 Returns
1076 -------
1077 result : TODO: DM-17458
1078 TODO: DM-17458
1079 """
1080
1081 result = None
1082 pks = measRecord.getFootprint().getPeaks()
1083
1084 # Check if the footprint consists of a putative dipole - else don't fit it.
1085 if (
1086 (len(pks) <= 1) # one peak in the footprint - not a dipole
1087 or (len(pks) > 1 and (np.sign(pks[0].getPeakValue())
1088 == np.sign(pks[-1].getPeakValue()))) # peaks are same sign - not a dipole
1089 ):
1090 measRecord.set(self.classificationFlagKeyclassificationFlagKey, False)
1091 measRecord.set(self.classificationAttemptedFlagKeyclassificationAttemptedFlagKey, False)
1092 self.failfail(measRecord, measBase.MeasurementError('not a dipole', self.FAILURE_NOT_DIPOLEFAILURE_NOT_DIPOLE))
1093 if not self.config.fitAllDiaSources:
1094 return result
1095
1096 try:
1097 alg = self.DipoleFitAlgorithmClassDipoleFitAlgorithmClass(exposure, posImage=posExp, negImage=negExp)
1098 result, _ = alg.fitDipole(
1099 measRecord, rel_weight=self.config.relWeight,
1100 tol=self.config.tolerance,
1101 maxSepInSigma=self.config.maxSeparation,
1102 fitBackground=self.config.fitBackground,
1103 separateNegParams=self.config.fitSeparateNegParams,
1104 verbose=False, display=False)
1105 except pexExcept.LengthError:
1106 self.failfail(measRecord, measBase.MeasurementError('edge failure', self.FAILURE_EDGEFAILURE_EDGE))
1107 except Exception:
1108 self.failfail(measRecord, measBase.MeasurementError('dipole fit failure', self.FAILURE_FITFAILURE_FIT))
1109
1110 if result is None:
1111 measRecord.set(self.classificationFlagKeyclassificationFlagKey, False)
1112 measRecord.set(self.classificationAttemptedFlagKeyclassificationAttemptedFlagKey, False)
1113 return result
1114
1115 self.loglog.debug("Dipole fit result: %d %s", measRecord.getId(), str(result))
1116
1117 if result.posFlux <= 1.: # usually around 0.1 -- the minimum flux allowed -- i.e. bad fit.
1118 self.failfail(measRecord, measBase.MeasurementError('dipole fit failure', self.FAILURE_FITFAILURE_FIT))
1119
1120 # add chi2, coord/flux uncertainties (TBD), dipole classification
1121 # Add the relevant values to the measRecord
1122 measRecord[self.posFluxKey] = result.posFlux
1123 measRecord[self.posFluxErrKey] = result.signalToNoise # to be changed to actual sigma!
1124 measRecord[self.posCentroidKeyX] = result.posCentroidX
1125 measRecord[self.posCentroidKeyY] = result.posCentroidY
1126
1127 measRecord[self.negFluxKey] = result.negFlux
1128 measRecord[self.negFluxErrKey] = result.signalToNoise # to be changed to actual sigma!
1129 measRecord[self.negCentroidKeyX] = result.negCentroidX
1130 measRecord[self.negCentroidKeyY] = result.negCentroidY
1131
1132 # Dia source flux: average of pos+neg
1133 measRecord[self.fluxKeyfluxKey] = (abs(result.posFlux) + abs(result.negFlux))/2.
1134 measRecord[self.orientationKeyorientationKey] = result.orientation
1135 measRecord[self.separationKeyseparationKey] = np.sqrt((result.posCentroidX - result.negCentroidX)**2.
1136 + (result.posCentroidY - result.negCentroidY)**2.)
1137 measRecord[self.centroidKeyX] = result.centroidX
1138 measRecord[self.centroidKeyY] = result.centroidY
1139
1140 measRecord[self.signalToNoiseKeysignalToNoiseKey] = result.signalToNoise
1141 measRecord[self.chi2dofKeychi2dofKey] = result.redChi2
1142
1143 self.doClassifydoClassify(measRecord, result.chi2)
1144
1145 def doClassify(self, measRecord, chi2val):
1146 """Classify a source as a dipole.
1147
1148 Parameters
1149 ----------
1150 measRecord : TODO: DM-17458
1151 TODO: DM-17458
1152 chi2val : TODO: DM-17458
1153 TODO: DM-17458
1154
1155 Notes
1156 -----
1157 Sources are classified as dipoles, or not, according to three criteria:
1158
1159 1. Does the total signal-to-noise surpass the ``minSn``?
1160 2. Are the pos/neg fluxes greater than 1.0 and no more than 0.65 (``maxFluxRatio``)
1161 of the total flux? By default this will never happen since ``posFlux == negFlux``.
1162 3. Is it a good fit (``chi2dof`` < 1)? (Currently not used.)
1163 """
1164
1165 # First, does the total signal-to-noise surpass the minSn?
1166 passesSn = measRecord[self.signalToNoiseKeysignalToNoiseKey] > self.config.minSn
1167
1168 # Second, are the pos/neg fluxes greater than 1.0 and no more than 0.65 (param maxFluxRatio)
1169 # of the total flux? By default this will never happen since posFlux = negFlux.
1170 passesFluxPos = (abs(measRecord[self.posFluxKey])
1171 / (measRecord[self.fluxKeyfluxKey]*2.)) < self.config.maxFluxRatio
1172 passesFluxPos &= (abs(measRecord[self.posFluxKey]) >= 1.0)
1173 passesFluxNeg = (abs(measRecord[self.negFluxKey])
1174 / (measRecord[self.fluxKeyfluxKey]*2.)) < self.config.maxFluxRatio
1175 passesFluxNeg &= (abs(measRecord[self.negFluxKey]) >= 1.0)
1176 allPass = (passesSn and passesFluxPos and passesFluxNeg) # and passesChi2)
1177
1178 # Third, is it a good fit (chi2dof < 1)?
1179 # Use scipy's chi2 cumulative distrib to estimate significance
1180 # This doesn't really work since I don't trust the values in the variance plane (which
1181 # affects the least-sq weights, which affects the resulting chi2).
1182 # But I'm going to keep this here for future use.
1183 if False:
1184 from scipy.stats import chi2
1185 ndof = chi2val / measRecord[self.chi2dofKeychi2dofKey]
1186 significance = chi2.cdf(chi2val, ndof)
1187 passesChi2 = significance < self.config.maxChi2DoF
1188 allPass = allPass and passesChi2
1189
1190 measRecord.set(self.classificationAttemptedFlagKeyclassificationAttemptedFlagKey, True)
1191
1192 if allPass: # Note cannot pass `allPass` into the `measRecord.set()` call below...?
1193 measRecord.set(self.classificationFlagKeyclassificationFlagKey, True)
1194 else:
1195 measRecord.set(self.classificationFlagKeyclassificationFlagKey, False)
1196
1197 def fail(self, measRecord, error=None):
1198 """Catch failures and set the correct flags.
1199 """
1200
1201 measRecord.set(self.flagKeyflagKey, True)
1202 if error is not None:
1203 if error.getFlagBit() == self.FAILURE_EDGEFAILURE_EDGE:
1204 self.loglog.warning('DipoleFitPlugin not run on record %d: %s', measRecord.getId(), str(error))
1205 measRecord.set(self.edgeFlagKeyedgeFlagKey, True)
1206 if error.getFlagBit() == self.FAILURE_FITFAILURE_FIT:
1207 self.loglog.warning('DipoleFitPlugin failed on record %d: %s', measRecord.getId(), str(error))
1208 measRecord.set(self.flagKeyflagKey, True)
1209 if error.getFlagBit() == self.FAILURE_NOT_DIPOLEFAILURE_NOT_DIPOLE:
1210 self.loglog.debug('DipoleFitPlugin not run on record %d: %s',
1211 measRecord.getId(), str(error))
1212 measRecord.set(self.classificationAttemptedFlagKeyclassificationAttemptedFlagKey, False)
1213 measRecord.set(self.flagKeyflagKey, True)
1214 else:
1215 self.loglog.warning('DipoleFitPlugin failed on record %d', measRecord.getId())
def fitDipoleImpl(self, source, tol=1e-7, rel_weight=0.5, fitBackground=1, bgGradientOrder=1, maxSepInSigma=5., separateNegParams=True, verbose=False)
def fitDipole(self, source, tol=1e-7, rel_weight=0.1, fitBackground=1, maxSepInSigma=5., separateNegParams=True, bgGradientOrder=1, verbose=False, display=False)
def displayFitResults(self, footprint, result)
def __init__(self, diffim, posImage=None, negImage=None)
def fail(self, measRecord, error=None)
def measure(self, measRecord, exposure, posExp=None, negExp=None)
def doClassify(self, measRecord, chi2val)
def _setupSchema(self, config, name, schema, metadata)
def __init__(self, config, name, schema, metadata)
def __init__(self, schema, algMetadata=None, **kwargs)
def run(self, sources, exposure, posExp=None, negExp=None, **kwargs)
def _getHeavyFootprintSubimage(self, fp, badfill=np.nan, grow=0)
def makeModel(self, x, flux, xcenPos, ycenPos, xcenNeg, ycenNeg, fluxNeg=None, b=None, x1=None, y1=None, xy=None, x2=None, y2=None, bNeg=None, x1Neg=None, y1Neg=None, xyNeg=None, x2Neg=None, y2Neg=None, **kwargs)
def makeBackgroundModel(self, in_x, pars=None)
def fitFootprintBackground(self, source, posImage, order=1)
def makeStarModel(self, bbox, psf, xcen, ycen, flux)