coaddDriver.py

Go to the documentation of this file.

import lsst.afw.image as afwImage
import lsst.afw.geom as afwGeom
from lsst.afw.fits import FitsError
import lsst.geom as geom
from lsst.ctrl.pool.parallel import BatchPoolTask
from lsst.ctrl.pool.pool import Pool, abortOnError, NODE
import lsst.sphgeom
from lsst.pex.config import Config, Field, ConfigurableField
from lsst.pipe.base import Struct, ArgumentParser, ConfigDatasetType
from lsst.pipe.tasks.coaddBase import CoaddTaskRunner
from lsst.pipe.tasks.makeCoaddTempExp import MakeCoaddTempExpTask
from lsst.pipe.tasks.multiBand import DetectCoaddSourcesTask
from lsst.pipe.tasks.selectImages import WcsSelectImagesTask
from lsst.pipe.tasks.assembleCoadd import SafeClipAssembleCoaddTask
from lsst.pipe.drivers.utils import getDataRef, NullSelectImagesTask, TractDataIdContainer
 
 
class CoaddDriverConfig(Config):
    coaddName = Field(dtype=str, default="deep", doc="Name for coadd")
    select = ConfigurableField(
        target=WcsSelectImagesTask, doc="Select images to process")
    makeCoaddTempExp = ConfigurableField(
        target=MakeCoaddTempExpTask, doc="Warp images to sky")
    doBackgroundReference = Field(
        dtype=bool, default=False, doc="Build background reference?")
    backgroundReference = ConfigurableField(
        target=NullSelectImagesTask, doc="Build background reference")
    assembleCoadd = ConfigurableField(
        target=SafeClipAssembleCoaddTask, doc="Assemble warps into coadd")
    doDetection = Field(dtype=bool, default=True,
                        doc="Run detection on the coaddition product")
    detectCoaddSources = ConfigurableField(
        target=DetectCoaddSourcesTask, doc="Detect sources on coadd")
    hasFakes = Field(dtype=bool, default=False,
                     doc="Should be set to True if fake sources were added to the data before processing.")
    calexpType = Field(dtype=str, default="calexp",
                       doc="Should be set to fakes_calexp if you want to process calexps with fakes in.")
 
    def setDefaults(self):
        self.makeCoaddTempExpmakeCoaddTempExp.select.retarget(NullSelectImagesTask)
        self.assembleCoaddassembleCoadd.select.retarget(NullSelectImagesTask)
        self.assembleCoaddassembleCoadd.doWrite = False
 
    def validate(self):
        if self.makeCoaddTempExpmakeCoaddTempExp.coaddName != self.coaddNamecoaddName:
            raise RuntimeError(
                "makeCoaddTempExp.coaddName and coaddName don't match")
        if self.assembleCoaddassembleCoadd.coaddName != self.coaddNamecoaddName:
            raise RuntimeError(
                "assembleCoadd.coaddName and coaddName don't match")
        if self.assembleCoaddassembleCoadd.matchingKernelSize != self.makeCoaddTempExpmakeCoaddTempExp.matchingKernelSize:
            message = ("assembleCoadd.matchingKernelSize (%s) and makeCoaddTempExp.matchingKernelSize (%s)"
                       " don't match" % (self.assembleCoaddassembleCoadd.matchingKernelSize,
                                         self.makeCoaddTempExpmakeCoaddTempExp.matchingKernelSize))
            raise RuntimeError(message)
 
 
class CoaddDriverTaskRunner(CoaddTaskRunner):
 
    def __init__(self, TaskClass, parsedCmd, doReturnResults=False):
        CoaddTaskRunner.__init__(self, TaskClass, parsedCmd, doReturnResults)
        self.reusereuse = parsedCmd.reuse
 
    def makeTask(self, parsedCmd=None, args=None):
        return self.TaskClass(config=self.config, log=self.log, reuse=self.reusereuse)
 
    @staticmethod
    def getTargetList(parsedCmd, **kwargs):
        """!Get bare butler into Task
 
        @param parsedCmd results of parsing command input
        """
        kwargs["butler"] = parsedCmd.butler
        kwargs["selectIdList"] = [
            ref.dataId for ref in parsedCmd.selectId.refList]
        return [(parsedCmd.id.refList, kwargs), ]
 
 
def unpickle(factory, args, kwargs):
    """Unpickle something by calling a factory"""
    return factory(*args, **kwargs)
 
 
class CoaddDriverTask(BatchPoolTask):
    ConfigClass = CoaddDriverConfig
    _DefaultName = "coaddDriver"
    RunnerClass = CoaddDriverTaskRunner
 
    def __init__(self, reuse=tuple(), **kwargs):
        BatchPoolTask.__init__(self, **kwargs)
        self.reusereuse = reuse
        self.makeSubtaskmakeSubtask("select")
        self.makeSubtaskmakeSubtask("makeCoaddTempExp", reuse=("makeCoaddTempExp" in self.reusereuse))
        self.makeSubtaskmakeSubtask("backgroundReference")
        self.makeSubtaskmakeSubtask("assembleCoadd")
        self.makeSubtaskmakeSubtask("detectCoaddSources")
        if self.configconfig.hasFakes:
            self.calexpTypecalexpType = "fakes_calexp"
        else:
            self.calexpTypecalexpType = "calexp"
 
    def __reduce__(self):
        """Pickler"""
        return unpickle, (self.__class__, [], dict(config=self.configconfig, name=self._name_name,
                                                   parentTask=self._parentTask_parentTask, log=self.loglog,
                                                   reuse=self.reusereuse))
 
    @classmethod
    def _makeArgumentParser(cls, **kwargs):
        """!Build argument parser
 
        Selection references are not cheap (reads Wcs), so are generated
        only if we're not doing a batch submission.
        """
        parser = ArgumentParser(name=cls._DefaultName_DefaultName)
        parser.add_id_argument("--id", "deepCoadd", help="data ID, e.g. --id tract=12345 patch=1,2",
                               ContainerClass=TractDataIdContainer)
        datasetType = ConfigDatasetType(name="calexpType")
        parser.add_id_argument("--selectId", datasetType=datasetType,
                               help="data ID, e.g. --selectId visit=6789 ccd=0..9")
        parser.addReuseOption(["makeCoaddTempExp", "assembleCoadd", "detectCoaddSources"])
        return parser
 
    @classmethod
    def batchWallTime(cls, time, parsedCmd, numCores):
        """!
        Return walltime request for batch job
 
        @param time: Requested time per iteration
        @param parsedCmd: Results of argument parsing
        @param numCores: Number of cores
        @return float walltime request length
        """
        numTargets = len(parsedCmd.selectId.refList)
        return time*numTargets/float(numCores)
 
    @abortOnError
    def runDataRef(self, tractPatchRefList, butler, selectIdList=[]):
        """!Determine which tracts are non-empty before processing
 
        @param tractPatchRefList: List of tracts and patches to include in the coaddition
        @param butler: butler reference object
        @param selectIdList: List of data Ids (i.e. visit, ccd) to consider when making the coadd
        @return list of references to sel.runTract function evaluation for each tractPatchRefList member
        """
        pool = Pool("tracts")
        pool.storeSet(butler=butler, skymap=butler.get(
            self.configconfig.coaddName + "Coadd_skyMap"))
        tractIdList = []
        for patchRefList in tractPatchRefList:
            tractSet = set([patchRef.dataId["tract"]
                            for patchRef in patchRefList])
            assert len(tractSet) == 1
            tractIdList.append(tractSet.pop())
 
        selectDataList = [data for data in pool.mapNoBalance(self.readSelectionreadSelection, selectIdList) if
                          data is not None]
        nonEmptyList = pool.mapNoBalance(
            self.checkTractcheckTract, tractIdList, selectDataList)
        tractPatchRefList = [patchRefList for patchRefList, nonEmpty in
                             zip(tractPatchRefList, nonEmptyList) if nonEmpty]
        self.loglog.info("Non-empty tracts (%d): %s" % (len(tractPatchRefList),
                                                     [patchRefList[0].dataId["tract"] for patchRefList in
                                                      tractPatchRefList]))
        # Install the dataRef in the selectDataList
        for data in selectDataList:
            data.dataRef = getDataRef(butler, data.dataId, self.calexpTypecalexpType)
 
        # Process the non-empty tracts
        return [self.runrun(patchRefList, butler, selectDataList) for patchRefList in tractPatchRefList]
 
    @abortOnError
    def run(self, patchRefList, butler, selectDataList=[]):
        """!Run stacking on a tract
 
        This method only runs on the master node.
 
        @param patchRefList: List of patch data references for tract
        @param butler: Data butler
        @param selectDataList: List of SelectStruct for inputs
        """
        pool = Pool("stacker")
        pool.cacheClear()
        pool.storeSet(butler=butler, warpType=self.configconfig.coaddName + "Coadd_directWarp",
                      coaddType=self.configconfig.coaddName + "Coadd")
        patchIdList = [patchRef.dataId for patchRef in patchRefList]
 
        selectedData = pool.map(self.warpwarp, patchIdList, selectDataList)
        if self.configconfig.doBackgroundReference:
            self.backgroundReference.runDataRef(patchRefList, selectDataList)
 
        def refNamer(patchRef):
            return tuple(map(int, patchRef.dataId["patch"].split(",")))
 
        lookup = dict(zip(map(refNamer, patchRefList), selectedData))
        coaddData = [Struct(patchId=patchRef.dataId, selectDataList=lookup[refNamer(patchRef)]) for
                     patchRef in patchRefList]
        pool.map(self.coaddcoadd, coaddData)
 
    def readSelection(self, cache, selectId):
        """!Read Wcs of selected inputs
 
        This method only runs on slave nodes.
        This method is similar to SelectDataIdContainer.makeDataRefList,
        creating a Struct like a SelectStruct, except with a dataId instead
        of a dataRef (to ease MPI).
 
        @param cache: Pool cache
        @param selectId: Data identifier for selected input
        @return a SelectStruct with a dataId instead of dataRef
        """
        try:
            ref = getDataRef(cache.butler, selectId, self.calexpTypecalexpType)
            self.loglog.info("Reading Wcs from %s" % (selectId,))
            md = ref.get("calexp_md", immediate=True)
            wcs = afwGeom.makeSkyWcs(md)
            data = Struct(dataId=selectId, wcs=wcs, bbox=afwImage.bboxFromMetadata(md))
        except FitsError:
            self.loglog.warn("Unable to construct Wcs from %s" % (selectId,))
            return None
        return data
 
    def checkTract(self, cache, tractId, selectIdList):
        """!Check whether a tract has any overlapping inputs
 
        This method only runs on slave nodes.
 
        @param cache: Pool cache
        @param tractId: Data identifier for tract
        @param selectDataList: List of selection data
        @return whether tract has any overlapping inputs
        """
        def makePolygon(wcs, bbox):
            """Return a polygon for the image, given Wcs and bounding box"""
            boxPixelCorners = geom.Box2D(bbox).getCorners()
            boxSkyCorners = wcs.pixelToSky(boxPixelCorners)
            return lsst.sphgeom.ConvexPolygon.convexHull([coord.getVector() for coord in boxSkyCorners])
 
        skymap = cache.skymap
        tract = skymap[tractId]
        tractWcs = tract.getWcs()
        tractPoly = makePolygon(tractWcs, tract.getBBox())
 
        for selectData in selectIdList:
            if not hasattr(selectData, "poly"):
                selectData.poly = makePolygon(selectData.wcs, selectData.bbox)
            if tractPoly.intersects(selectData.poly):
                return True
        return False
 
    def warp(self, cache, patchId, selectDataList):
        """!Warp all images for a patch
 
        Only slave nodes execute this method.
 
        Because only one argument may be passed, it is expected to
        contain multiple elements, which are:
 
        @param patchRef: data reference for patch
        @param selectDataList: List of SelectStruct for inputs
        @return selectDataList with non-overlapping elements removed
        """
        patchRef = getDataRef(cache.butler, patchId, cache.coaddType)
        selectDataList = self.selectExposuresselectExposures(patchRef, selectDataList)
        with self.logOperationlogOperation("warping %s" % (patchRef.dataId,), catch=True):
            self.makeCoaddTempExp.runDataRef(patchRef, selectDataList)
        return selectDataList
 
    def coadd(self, cache, data):
        """!Construct coadd for a patch and measure
 
        Only slave nodes execute this method.
 
        Because only one argument may be passed, it is expected to
        contain multiple elements, which are:
 
        @param patchRef: data reference for patch
        @param selectDataList: List of SelectStruct for inputs
        """
        patchRef = getDataRef(cache.butler, data.patchId, cache.coaddType)
        selectDataList = data.selectDataList
        coadd = None
 
        # We skip the assembleCoadd step if either the *Coadd dataset exists
        # or we aren't configured to write it, we're supposed to reuse
        # detectCoaddSources outputs too, and those outputs already exist.
        canSkipDetection = (
            "detectCoaddSources" in self.reusereuse and
            patchRef.datasetExists(self.detectCoaddSources.config.coaddName+"Coadd_det", write=True)
        )
        if "assembleCoadd" in self.reusereuse:
            if patchRef.datasetExists(cache.coaddType, write=True):
                self.loglog.info("%s: Skipping assembleCoadd for %s; outputs already exist." %
                              (NODE, patchRef.dataId))
                coadd = patchRef.get(cache.coaddType, immediate=True)
            elif not self.configconfig.assembleCoadd.doWrite and self.configconfig.doDetection and canSkipDetection:
                self.loglog.info(
                    "%s: Skipping assembleCoadd and detectCoaddSources for %s; outputs already exist." %
                    (NODE, patchRef.dataId)
                )
                return
        if coadd is None:
            with self.logOperationlogOperation("coadding %s" % (patchRef.dataId,), catch=True):
                coaddResults = self.assembleCoadd.runDataRef(patchRef, selectDataList)
                if coaddResults is not None:
                    coadd = coaddResults.coaddExposure
                    canSkipDetection = False  # can't skip it because coadd may have changed
        if coadd is None:
            return
 
        # The section of code below determines if the detection task should be
        # run. If detection is run, then the products are written out as
        # deepCoadd_calexp. If detection is not run, then the outputs of the
        # assemble task are written out as deepCoadd.
        if self.configconfig.doDetection:
            if canSkipDetection:
                self.loglog.info("%s: Skipping detectCoaddSources for %s; outputs already exist." %
                              (NODE, patchRef.dataId))
                return
            with self.logOperationlogOperation("detection on {}".format(patchRef.dataId),
                                   catch=True):
                idFactory = self.detectCoaddSources.makeIdFactory(patchRef)
                expId = int(patchRef.get(self.configconfig.coaddName + "CoaddId"))
                # This includes background subtraction, so do it before writing
                # the coadd
                detResults = self.detectCoaddSources.run(coadd, idFactory, expId=expId)
                self.detectCoaddSources.write(detResults, patchRef)
        else:
            if self.configconfig.hasFakes:
                patchRef.put(coadd, "fakes_" + self.assembleCoadd.config.coaddName + "Coadd")
            else:
                patchRef.put(coadd, self.assembleCoadd.config.coaddName + "Coadd")
 
    def selectExposures(self, patchRef, selectDataList):
        """!Select exposures to operate upon, via the SelectImagesTask
 
        This is very similar to CoaddBaseTask.selectExposures, except we return
        a list of SelectStruct (same as the input), so we can plug the results into
        future uses of SelectImagesTask.
 
        @param patchRef data reference to a particular patch
        @param selectDataList list of references to specific data products (i.e. visit, ccd)
        @return filtered list of SelectStruct
        """
        def key(dataRef):
            return tuple(dataRef.dataId[k] for k in sorted(dataRef.dataId))
        inputs = dict((key(select.dataRef), select)
                      for select in selectDataList)
        skyMap = patchRef.get(self.configconfig.coaddName + "Coadd_skyMap")
        tract = skyMap[patchRef.dataId["tract"]]
        patch = tract[(tuple(int(i)
                             for i in patchRef.dataId["patch"].split(",")))]
        bbox = patch.getOuterBBox()
        wcs = tract.getWcs()
        cornerPosList = geom.Box2D(bbox).getCorners()
        coordList = [wcs.pixelToSky(pos) for pos in cornerPosList]
        dataRefList = self.select.runDataRef(
            patchRef, coordList, selectDataList=selectDataList).dataRefList
        return [inputs[key(dataRef)] for dataRef in dataRefList]
 
    def writeMetadata(self, dataRef):
        pass