Coverage for python/lsst/meas/astrom/matcher_probabilistic.py: 26%
210 statements
« prev ^ index » next coverage.py v7.3.2, created at 2023-10-13 10:25 +0000
1# This file is part of meas_astrom.
2#
3# Developed for the LSST Data Management System.
4# This product includes software developed by the LSST Project
5# (https://www.lsst.org).
6# See the COPYRIGHT file at the top-level directory of this distribution
7# for details of code ownership.
8#
9# This program is free software: you can redistribute it and/or modify
10# it under the terms of the GNU General Public License as published by
11# the Free Software Foundation, either version 3 of the License, or
12# (at your option) any later version.
13#
14# This program is distributed in the hope that it will be useful,
15# but WITHOUT ANY WARRANTY; without even the implied warranty of
16# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17# GNU General Public License for more details.
18#
19# You should have received a copy of the GNU General Public License
20# along with this program. If not, see <https://www.gnu.org/licenses/>.
22__all__ = ['ConvertCatalogCoordinatesConfig', 'MatchProbabilisticConfig', 'MatcherProbabilistic']
24import lsst.pex.config as pexConfig
26from dataclasses import dataclass
27import logging
28import numpy as np
29import pandas as pd
30from scipy.spatial import cKDTree
31import time
32from typing import Callable, Set
34logger_default = logging.getLogger(__name__)
37def _mul_column(column: np.array, value: float):
38 if value is not None and value != 1:
39 column *= value
40 return column
43def _radec_to_xyz(ra, dec):
44 """Convert input ra/dec coordinates to spherical unit vectors.
46 Parameters
47 ----------
48 ra, dec: `numpy.ndarray`
49 Arrays of right ascension/declination in degrees.
51 Returns
52 -------
53 vectors : `numpy.ndarray`, (N, 3)
54 Output unit vectors.
55 """
56 if ra.size != dec.size:
57 raise ValueError('ra and dec must be same size')
58 ras = np.radians(ra)
59 decs = np.radians(dec)
60 vectors = np.empty((ras.size, 3))
62 sin_dec = np.sin(np.pi / 2 - decs)
63 vectors[:, 0] = sin_dec * np.cos(ras)
64 vectors[:, 1] = sin_dec * np.sin(ras)
65 vectors[:, 2] = np.cos(np.pi / 2 - decs)
67 return vectors
70@dataclass
71class CatalogExtras:
72 """Store frequently-reference (meta)data relevant for matching a catalog.
74 Parameters
75 ----------
76 catalog : `pandas.DataFrame`
77 A pandas catalog to store extra information for.
78 select : `numpy.array`
79 A numpy boolean array of the same length as catalog to be used for
80 target selection.
81 """
82 n: int
83 indices: np.array
84 select: np.array
86 coordinate_factor: float = None
88 def __init__(self, catalog: pd.DataFrame, select: np.array = None, coordinate_factor: float = None):
89 self.n = len(catalog)
90 self.select = np.ones(self.n, dtype=bool) if select is None else select
91 self.indices = np.flatnonzero(select) if select is not None else np.arange(self.n)
92 self.coordinate_factor = coordinate_factor
95@dataclass(frozen=True)
96class ComparableCatalog:
97 """A catalog with sources with coordinate columns in some standard format/units.
99 catalog : `pandas.DataFrame`
100 A catalog with comparable coordinate columns.
101 column_coord1 : `str`
102 The first spatial coordinate column name.
103 column_coord2 : `str`
104 The second spatial coordinate column name.
105 coord1 : `numpy.array`
106 The first spatial coordinate values.
107 coord2 : `numpy.array`
108 The second spatial coordinate values.
109 extras : `CatalogExtras`
110 Extra cached (meta)data for the `catalog`.
111 """
112 catalog: pd.DataFrame
113 column_coord1: str
114 column_coord2: str
115 coord1: np.array
116 coord2: np.array
117 extras: CatalogExtras
120class ConvertCatalogCoordinatesConfig(pexConfig.Config):
121 """Configuration for the MatchProbabilistic matcher.
122 """
123 column_ref_coord1 = pexConfig.Field(
124 dtype=str,
125 default='ra',
126 doc='The reference table column for the first spatial coordinate (usually x or ra).',
127 )
128 column_ref_coord2 = pexConfig.Field(
129 dtype=str,
130 default='dec',
131 doc='The reference table column for the second spatial coordinate (usually y or dec).'
132 'Units must match column_ref_coord1.',
133 )
134 column_target_coord1 = pexConfig.Field(
135 dtype=str,
136 default='coord_ra',
137 doc='The target table column for the first spatial coordinate (usually x or ra).'
138 'Units must match column_ref_coord1.',
139 )
140 column_target_coord2 = pexConfig.Field(
141 dtype=str,
142 default='coord_dec',
143 doc='The target table column for the second spatial coordinate (usually y or dec).'
144 'Units must match column_ref_coord2.',
145 )
146 coords_spherical = pexConfig.Field(
147 dtype=bool,
148 default=True,
149 doc='Whether column_*_coord[12] are spherical coordinates (ra/dec) or not (pixel x/y)',
150 )
151 coords_ref_factor = pexConfig.Field(
152 dtype=float,
153 default=1.0,
154 doc='Multiplicative factor for reference catalog coordinates.'
155 'If coords_spherical is true, this must be the number of degrees per unit increment of '
156 'column_ref_coord[12]. Otherwise, it must convert the coordinate to the same units'
157 ' as the target coordinates.',
158 )
159 coords_target_factor = pexConfig.Field(
160 dtype=float,
161 default=1.0,
162 doc='Multiplicative factor for target catalog coordinates.'
163 'If coords_spherical is true, this must be the number of degrees per unit increment of '
164 'column_target_coord[12]. Otherwise, it must convert the coordinate to the same units'
165 ' as the reference coordinates.',
166 )
167 coords_ref_to_convert = pexConfig.DictField( 167 ↛ exitline 167 didn't jump to the function exit
168 default=None,
169 optional=True,
170 keytype=str,
171 itemtype=str,
172 dictCheck=lambda x: len(x) == 2,
173 doc='Dict mapping sky coordinate columns to be converted to pixel columns',
174 )
175 mag_zeropoint_ref = pexConfig.Field(
176 dtype=float,
177 default=31.4,
178 doc='Magnitude zeropoint for reference catalog.',
179 )
181 def format_catalogs(
182 self,
183 catalog_ref: pd.DataFrame,
184 catalog_target: pd.DataFrame,
185 select_ref: np.array = None,
186 select_target: np.array = None,
187 radec_to_xy_func: Callable = None,
188 return_converted_columns: bool = False,
189 **kwargs,
190 ):
191 """Format matched catalogs that may require coordinate conversions.
193 Parameters
194 ----------
195 catalog_ref : `pandas.DataFrame`
196 A reference catalog for comparison to `catalog_target`.
197 catalog_target : `pandas.DataFrame`
198 A target catalog with measurements for comparison to `catalog_ref`.
199 select_ref : `numpy.ndarray`, (Nref,)
200 A boolean array of len `catalog_ref`, True for valid match candidates.
201 select_target : `numpy.ndarray`, (Ntarget,)
202 A boolean array of len `catalog_target`, True for valid match candidates.
203 radec_to_xy_func : `typing.Callable`
204 Function taking equal-length ra, dec arrays and returning an ndarray of
205 - ``x``: current parameter (`float`).
206 - ``extra_args``: additional arguments (`dict`).
207 return_converted_columns : `bool`
208 Whether to return converted columns in the `coord1` and `coord2`
209 attributes, rather than keep the original values.
210 kwargs
212 Returns
213 -------
214 compcat_ref, compcat_target : `ComparableCatalog`
215 Comparable catalogs corresponding to the input reference and target.
217 """
218 convert_ref = self.coords_ref_to_convert
219 if convert_ref and not callable(radec_to_xy_func):
220 raise TypeError('radec_to_xy_func must be callable if converting ref coords')
222 # Set up objects with frequently-used attributes like selection bool array
223 extras_ref, extras_target = (
224 CatalogExtras(catalog, select=select, coordinate_factor=coord_factor)
225 for catalog, select, coord_factor in zip(
226 (catalog_ref, catalog_target),
227 (select_ref, select_target),
228 (self.coords_ref_factor, self.coords_target_factor),
229 )
230 )
232 compcats = []
234 # Retrieve coordinates and multiply them by scaling factors
235 for catalog, extras, (column1, column2), convert in (
236 (catalog_ref, extras_ref, (self.column_ref_coord1, self.column_ref_coord2), convert_ref),
237 (catalog_target, extras_target, (self.column_target_coord1, self.column_target_coord2), False),
238 ):
239 coord1, coord2 = (
240 _mul_column(catalog[column], extras.coordinate_factor)
241 for column in (column1, column2)
242 )
243 if convert:
244 xy_ref = radec_to_xy_func(coord1, coord2, self.coords_ref_factor, **kwargs)
245 for idx_coord, column_out in enumerate(self.coords_ref_to_convert.values()):
246 coord = np.array([xy[idx_coord] for xy in xy_ref])
247 catalog[column_out] = coord
248 if convert_ref and return_converted_columns:
249 column1, column2 = self.coords_ref_to_convert.values()
250 coord1, coord2 = catalog[column1], catalog[column2]
251 if isinstance(coord1, pd.Series):
252 coord1 = coord1.values
253 if isinstance(coord2, pd.Series):
254 coord2 = coord2.values
256 compcats.append(ComparableCatalog(
257 catalog=catalog, column_coord1=column1, column_coord2=column2,
258 coord1=coord1, coord2=coord2, extras=extras,
259 ))
261 return tuple(compcats)
264class MatchProbabilisticConfig(pexConfig.Config):
265 """Configuration for the MatchProbabilistic matcher.
266 """
267 column_ref_order = pexConfig.Field(
268 dtype=str,
269 default=None,
270 optional=True,
271 doc="Name of column in reference catalog specifying order for matching."
272 " Derived from columns_ref_flux if not set.",
273 )
275 @property
276 def columns_in_ref(self) -> Set[str]:
277 columns_all = [
278 self.coord_format.column_ref_coord1,
279 self.coord_format.column_ref_coord2,
280 ]
281 for columns in (
282 self.columns_ref_flux,
283 self.columns_ref_meas,
284 self.columns_ref_select_false,
285 self.columns_ref_select_true,
286 self.columns_ref_copy,
287 ):
288 columns_all.extend(columns)
290 return set(columns_all)
292 @property
293 def columns_in_target(self) -> Set[str]:
294 columns_all = [
295 self.coord_format.column_target_coord1,
296 self.coord_format.column_target_coord2,
297 ]
298 for columns in (
299 self.columns_target_meas,
300 self.columns_target_err,
301 self.columns_target_select_false,
302 self.columns_target_select_true,
303 self.columns_target_copy,
304 ):
305 columns_all.extend(columns)
306 return set(columns_all)
308 columns_ref_copy = pexConfig.ListField( 308 ↛ exitline 308 didn't jump to the function exit
309 dtype=str,
310 default=[],
311 listCheck=lambda x: len(set(x)) == len(x),
312 optional=True,
313 doc='Reference table columns to copy unchanged into both match tables',
314 )
315 columns_ref_flux = pexConfig.ListField(
316 dtype=str,
317 default=[],
318 optional=True,
319 doc="List of reference flux columns to nansum total magnitudes from if column_order is None",
320 )
321 columns_ref_meas = pexConfig.ListField(
322 dtype=str,
323 doc='The reference table columns to compute match likelihoods from '
324 '(usually centroids and fluxes/magnitudes)',
325 )
326 columns_ref_select_true = pexConfig.ListField(
327 dtype=str,
328 default=tuple(),
329 doc='Reference table columns to require to be True for selecting sources',
330 )
331 columns_ref_select_false = pexConfig.ListField(
332 dtype=str,
333 default=tuple(),
334 doc='Reference table columns to require to be False for selecting sources',
335 )
336 columns_target_copy = pexConfig.ListField( 336 ↛ exitline 336 didn't jump to the function exit
337 dtype=str,
338 default=[],
339 listCheck=lambda x: len(set(x)) == len(x),
340 optional=True,
341 doc='Target table columns to copy unchanged into both match tables',
342 )
343 columns_target_meas = pexConfig.ListField(
344 dtype=str,
345 doc='Target table columns with measurements corresponding to columns_ref_meas',
346 )
347 columns_target_err = pexConfig.ListField(
348 dtype=str,
349 doc='Target table columns with standard errors (sigma) corresponding to columns_ref_meas',
350 )
351 columns_target_select_true = pexConfig.ListField(
352 dtype=str,
353 default=('detect_isPrimary',),
354 doc='Target table columns to require to be True for selecting sources',
355 )
356 columns_target_select_false = pexConfig.ListField(
357 dtype=str,
358 default=('merge_peak_sky',),
359 doc='Target table columns to require to be False for selecting sources',
360 )
361 coord_format = pexConfig.ConfigField(
362 dtype=ConvertCatalogCoordinatesConfig,
363 doc="Configuration for coordinate conversion",
364 )
365 mag_brightest_ref = pexConfig.Field(
366 dtype=float,
367 default=-np.inf,
368 doc='Bright magnitude cutoff for selecting reference sources to match.'
369 ' Ignored if column_ref_order is None.'
370 )
371 mag_faintest_ref = pexConfig.Field(
372 dtype=float,
373 default=np.Inf,
374 doc='Faint magnitude cutoff for selecting reference sources to match.'
375 ' Ignored if column_ref_order is None.'
376 )
377 match_dist_max = pexConfig.Field(
378 dtype=float,
379 default=0.5,
380 doc='Maximum match distance. Units must be arcseconds if coords_spherical, '
381 'or else match those of column_*_coord[12] multiplied by coords_*_factor.',
382 )
383 match_n_max = pexConfig.Field(
384 dtype=int,
385 default=10,
386 optional=True,
387 doc='Maximum number of spatial matches to consider (in ascending distance order).',
388 )
389 match_n_finite_min = pexConfig.Field(
390 dtype=int,
391 default=3,
392 optional=True,
393 doc='Minimum number of columns with a finite value to measure match likelihood',
394 )
395 order_ascending = pexConfig.Field(
396 dtype=bool,
397 default=False,
398 optional=True,
399 doc='Whether to order reference match candidates in ascending order of column_ref_order '
400 '(should be False if the column is a flux and True if it is a magnitude.',
401 )
404def default_value(dtype):
405 if dtype == str:
406 return ''
407 elif dtype == np.signedinteger:
408 return np.Inf
409 elif dtype == np.unsignedinteger:
410 return -np.Inf
411 return None
414class MatcherProbabilistic:
415 """A probabilistic, greedy catalog matcher.
417 Parameters
418 ----------
419 config: `MatchProbabilisticConfig`
420 A configuration instance.
421 """
422 config: MatchProbabilisticConfig
424 def __init__(
425 self,
426 config: MatchProbabilisticConfig,
427 ):
428 self.config = config
430 def match(
431 self,
432 catalog_ref: pd.DataFrame,
433 catalog_target: pd.DataFrame,
434 select_ref: np.array = None,
435 select_target: np.array = None,
436 logger: logging.Logger = None,
437 logging_n_rows: int = None,
438 **kwargs
439 ):
440 """Match catalogs.
442 Parameters
443 ----------
444 catalog_ref : `pandas.DataFrame`
445 A reference catalog to match in order of a given column (i.e. greedily).
446 catalog_target : `pandas.DataFrame`
447 A target catalog for matching sources from `catalog_ref`. Must contain measurements with errors.
448 select_ref : `numpy.array`
449 A boolean array of the same length as `catalog_ref` selecting the sources that can be matched.
450 select_target : `numpy.array`
451 A boolean array of the same length as `catalog_target` selecting the sources that can be matched.
452 logger : `logging.Logger`
453 A Logger for logging.
454 logging_n_rows : `int`
455 The number of sources to match before printing a log message.
456 kwargs
457 Additional keyword arguments to pass to `format_catalogs`.
459 Returns
460 -------
461 catalog_out_ref : `pandas.DataFrame`
462 A catalog of identical length to `catalog_ref`, containing match information for rows selected by
463 `select_ref` (including the matching row index in `catalog_target`).
464 catalog_out_target : `pandas.DataFrame`
465 A catalog of identical length to `catalog_target`, containing the indices of matching rows in
466 `catalog_ref`.
467 exceptions : `dict` [`int`, `Exception`]
468 A dictionary keyed by `catalog_target` row number of the first exception caught when matching.
469 """
470 if logger is None:
471 logger = logger_default
473 config = self.config
475 # Transform any coordinates, if required
476 # Note: The returned objects contain the original catalogs, as well as
477 # transformed coordinates, and the selection of sources for matching.
478 # These might be identical to the arrays passed as kwargs, but that
479 # depends on config settings.
480 # For the rest of this function, the selection arrays will be used,
481 # but the indices of the original, unfiltered catalog will also be
482 # output, so some further indexing steps are needed.
483 ref, target = config.coord_format.format_catalogs(
484 catalog_ref=catalog_ref, catalog_target=catalog_target,
485 select_ref=select_ref, select_target=select_target,
486 **kwargs
487 )
489 # If no order is specified, take nansum of all flux columns for a 'total flux'
490 # Note: it won't actually be a total flux if bands overlap significantly
491 # (or it might define a filter with >100% efficiency
492 # Also, this is done on the original dataframe as it's harder to accomplish
493 # just with a recarray
494 column_order = (
495 catalog_ref.loc[ref.extras.select, config.column_ref_order]
496 if config.column_ref_order is not None else
497 np.nansum(catalog_ref.loc[ref.extras.select, config.columns_ref_flux], axis=1)
498 )
499 order = np.argsort(column_order if config.order_ascending else -column_order)
501 n_ref_select = len(ref.extras.indices)
503 match_dist_max = config.match_dist_max
504 coords_spherical = config.coord_format.coords_spherical
505 if coords_spherical:
506 match_dist_max = np.radians(match_dist_max / 3600.)
508 # Convert ra/dec sky coordinates to spherical vectors for accurate distances
509 func_convert = _radec_to_xyz if coords_spherical else np.vstack
510 vec_ref, vec_target = (
511 func_convert(cat.coord1[cat.extras.select], cat.coord2[cat.extras.select])
512 for cat in (ref, target)
513 )
515 # Generate K-d tree to compute distances
516 logger.info('Generating cKDTree with match_n_max=%d', config.match_n_max)
517 tree_obj = cKDTree(vec_target)
519 scores, idxs_target_select = tree_obj.query(
520 vec_ref,
521 distance_upper_bound=match_dist_max,
522 k=config.match_n_max,
523 )
525 n_target_select = len(target.extras.indices)
526 n_matches = np.sum(idxs_target_select != n_target_select, axis=1)
527 n_matched_max = np.sum(n_matches == config.match_n_max)
528 if n_matched_max > 0:
529 logger.warning(
530 '%d/%d (%.2f%%) selected true objects have n_matches=n_match_max(%d)',
531 n_matched_max, n_ref_select, 100.*n_matched_max/n_ref_select, config.match_n_max
532 )
534 # Pre-allocate outputs
535 target_row_match = np.full(target.extras.n, np.nan, dtype=np.int64)
536 ref_candidate_match = np.zeros(ref.extras.n, dtype=bool)
537 ref_row_match = np.full(ref.extras.n, np.nan, dtype=np.int64)
538 ref_match_count = np.zeros(ref.extras.n, dtype=np.int32)
539 ref_match_meas_finite = np.zeros(ref.extras.n, dtype=np.int32)
540 ref_chisq = np.full(ref.extras.n, np.nan, dtype=float)
542 # Need the original reference row indices for output
543 idx_orig_ref, idx_orig_target = (np.argwhere(cat.extras.select) for cat in (ref, target))
545 # Retrieve required columns, including any converted ones (default to original column name)
546 columns_convert = config.coord_format.coords_ref_to_convert
547 if columns_convert is None:
548 columns_convert = {}
549 data_ref = ref.catalog[
550 [columns_convert.get(column, column) for column in config.columns_ref_meas]
551 ].iloc[ref.extras.indices[order]]
552 data_target = target.catalog[config.columns_target_meas][target.extras.select]
553 errors_target = target.catalog[config.columns_target_err][target.extras.select]
555 exceptions = {}
556 # The kdTree uses len(inputs) as a sentinel value for no match
557 matched_target = {n_target_select, }
559 t_begin = time.process_time()
561 logger.info('Matching n_indices=%d/%d', len(order), len(ref.catalog))
562 for index_n, index_row_select in enumerate(order):
563 index_row = idx_orig_ref[index_row_select]
564 ref_candidate_match[index_row] = True
565 found = idxs_target_select[index_row_select, :]
566 # Select match candidates from nearby sources not already matched
567 # Note: set lookup is apparently fast enough that this is a few percent faster than:
568 # found = [x for x in found[found != n_target_select] if x not in matched_target]
569 # ... at least for ~1M sources
570 found = [x for x in found if x not in matched_target]
571 n_found = len(found)
572 if n_found > 0:
573 # This is an ndarray of n_found rows x len(data_ref/target) columns
574 chi = (
575 (data_target.iloc[found].values - data_ref.iloc[index_n].values)
576 / errors_target.iloc[found].values
577 )
578 finite = np.isfinite(chi)
579 n_finite = np.sum(finite, axis=1)
580 # Require some number of finite chi_sq to match
581 chisq_good = n_finite >= config.match_n_finite_min
582 if np.any(chisq_good):
583 try:
584 chisq_sum = np.zeros(n_found, dtype=float)
585 chisq_sum[chisq_good] = np.nansum(chi[chisq_good, :] ** 2, axis=1)
586 idx_chisq_min = np.nanargmin(chisq_sum / n_finite)
587 ref_match_meas_finite[index_row] = n_finite[idx_chisq_min]
588 ref_match_count[index_row] = len(chisq_good)
589 ref_chisq[index_row] = chisq_sum[idx_chisq_min]
590 idx_match_select = found[idx_chisq_min]
591 row_target = target.extras.indices[idx_match_select]
592 ref_row_match[index_row] = row_target
594 target_row_match[row_target] = index_row
595 matched_target.add(idx_match_select)
596 except Exception as error:
597 # Can't foresee any exceptions, but they shouldn't prevent
598 # matching subsequent sources
599 exceptions[index_row] = error
601 if logging_n_rows and ((index_n + 1) % logging_n_rows == 0):
602 t_elapsed = time.process_time() - t_begin
603 logger.info(
604 'Processed %d/%d in %.2fs at sort value=%.3f',
605 index_n + 1, n_ref_select, t_elapsed, column_order[order[index_n]],
606 )
608 data_ref = {
609 'match_candidate': ref_candidate_match,
610 'match_row': ref_row_match,
611 'match_count': ref_match_count,
612 'match_chisq': ref_chisq,
613 'match_n_chisq_finite': ref_match_meas_finite,
614 }
615 data_target = {
616 'match_candidate': target.extras.select if target.extras.select is not None else (
617 np.ones(target.extras.n, dtype=bool)),
618 'match_row': target_row_match,
619 }
621 for (columns, out_original, out_matched, in_original, in_matched, matches) in (
622 (
623 self.config.columns_ref_copy,
624 data_ref,
625 data_target,
626 ref,
627 target,
628 target_row_match,
629 ),
630 (
631 self.config.columns_target_copy,
632 data_target,
633 data_ref,
634 target,
635 ref,
636 ref_row_match,
637 ),
638 ):
639 matched = matches >= 0
640 idx_matched = matches[matched]
642 for column in columns:
643 values = in_original.catalog[column]
644 out_original[column] = values
645 dtype = in_original.catalog[column].dtype
647 # Pandas object columns can have mixed types - check for that
648 if dtype == object:
649 types = list(set((type(x) for x in values)))
650 if len(types) != 1:
651 raise RuntimeError(f'Column {column} dtype={dtype} has multiple types={types}')
652 dtype = types[0]
654 value_fill = default_value(dtype)
656 # Without this, the dtype would be '<U1' for an empty Unicode string
657 if dtype == str:
658 dtype = f'<U{max(len(x) for x in values)}'
660 column_match = np.full(in_matched.extras.n, value_fill, dtype=dtype)
661 column_match[matched] = in_original.catalog[column][idx_matched]
662 out_matched[f'match_{column}'] = column_match
664 catalog_out_ref = pd.DataFrame(data_ref)
665 catalog_out_target = pd.DataFrame(data_target)
667 return catalog_out_ref, catalog_out_target, exceptions