4. Using the Mappers API

The Mapper base class provides no functionality, rather its purpose is to define all the attributes for instances built from subclasses of Mapper. Though not strictly an abstract class, the Mapper base class cannot be used on its own.

The subclasses, PrimitiveMapper and RecipeMapper, do not override Mapper.__init__(). They implement mapping routines aiming to find the primitives and recipes that best match the input data.

The mappers require a Data Reduction Package itself containing Instrument Packages. We discuss these packages and how to build the basic directory structure for them in the Appendix.

Below, we use the geminidr data reduction package when we need an example. Note that geminidr is the default package unless it is overridden with the drpkg argument.

4.1. Selecting Primitives with PrimitiveMapper

Primitive classes are defined in a Data Reduction (DR) Package. For the Gemini instruments that is the geminidr package. The DR Package to use by the mapper is set by the drpkg argument to mapper.

These primitive classes define methods, referred to as primitives, that provide data processing functionality. Primitive classes in a DR Package are structured hierarchically and may employ multiple inheritance.

In Figure 4.1 below, we show an example of primitive class inheritance as done in geminidr, with GMOSImage the primitive class inheriting and mixing-in the other classes. In general, the class inheriting others will also have longer, more specific tagsets. This will ensure that the mapping for a GMOS image, if we follow the example in Figure 4.1, does return the GMOSImage class as the best match, not GMOS, not GEMINI.

The GMOSImage primitive class, by inheriting all of the classes above it and hence all their methods (aka primitives) becomes a complete set of primitives that can be applied to a GMOS IMAGE. It becomes a primitive set.

Generic primitive classes like CCD, Image, Photometry in Figure 4.1 normally should have empty tagsets.

_images/ExamplePrimSetInheritance.png

Figure 4.1: An example of primitive class inheritance. (A section of geminidr.)

Note

Hereafter, a Primitive class may be referred to as a “primitive set” or just “primitives”. Through inheritance, a primitive class collects many primitives from higher in the hierarchy. The primitive set implies the whole collection more clearly than just referring to the primitive class.

4.1.1. Mapping Data to Primitives

PrimitiveMapper requires as input a list of AstroData tags and the name of an instrument packages.

The mapping is done by matching the primitive set’s tagset attribute to the AstroData tags.

The primitive set is instantiated as follow, where ad is an AstroData object:

>>> from recipe_system.mappers.primitiveMapper import PrimitiveMapper
>>> tags = ad.tags
>>> instpkg = ad.instrument(generic=True).lower()
>>> pmapper = PrimitiveMapper(tags, instpkg)
>>> pclass = pmapper.get_applicable_primitives()
>>> p = pclass([ad])

This p can now be passed to the best match recipe.

4.1.1.1. Example

Let us discuss an example. Let us use a raw GMOS image.

 >>> ad = astrodata.open('S20161025S0111.fits')
 >>> ad.tags
set(['RAW', 'GMOS', 'GEMINI', 'SIDEREAL', 'UNPREPARED', 'IMAGE', 'SOUTH'])

The PrimitiveMapper uses these tags to search the geminidr packages, or the data reduction package specified in the mapper attribute drpkg. The first step in the search is to narrow it to the applicable instrument package, in this case, gmos. The mapper enforces the rule that the directory name for the instrument package be the lower case version of the AstroData descriptor instrument(). This speeds up the search but it is also a limitation.

The PrimitiveMapper instance stores the name of the package it will search in the dotpackage attribute. For example, using the GMOS ad:

>>> from recipe_system.mappers.primitiveMapper import PrimitiveMapper
 >>> pmapper = PrimitiveMapper([ad])
 >>> pmappmer.dotpackage
'geminidr.gmos'

The public method get_applicable_primitives(), when invoked, launches the search for the most appropriate primitive class. The search itself is focused on finding class objects with a tagset attribute.

>>> pclass = pmapper.get_applicable_primitives()

In the case of GMOS, the tagsets in primitive classes hierarchy (Figure 4.1 above) are:

class Gemini( ...  ):
   tagset = set(["GEMINI"])

class GMOS(Gemini, ... ):
   tagset = set(["GEMINI", "GMOS"])

class GMOSImage(GMOS, ... ):
   tagset = set(["GEMINI", "GMOS", "IMAGE"])

The function looks to match the tagset of the primitive class it finds to the AstroData tags. A successful match requires tagset to be a subset of the Astrodata tags. The match with the largest subset wins the contest. In our example, the class GMOSImage wins. Indeed:

>>> pclass
<class 'geminidr.gmos.primitives_gmos_image.GMOSImage'>

To use this selection of primitives, one then needs to instantiate it:

>>> p = pclass([ad])

This can be used by a recipe.

4.2. Selecting Recipes with RecipeMapper

Recipes are functions that receive a single argument: an instance of a primitive class. Recipe functions are not classes and do not (cannot) inherit. The recipe simply defines the primitives to be used and the order in which the primitive calls are made on the data.

Recipe functions are defined in python modules which we refer to as recipe libraries. The location of those modules in a data reduction package must obey some rules; they must be located in an instrument package under a subdirectory named recipes and therein in a subdirectory that matches the mode setting. The RecipeMapper searches only such directories. For example:

my_drpkg/instrumentX/recipes/modeY/recipes_IMAGING.py

4.2.1. Mapping Data to Recipes

At a minimum, RecipeMapper requires as input a list of AstroData tags and the name of an instrument package. The mapping is done by matching the recipe library’s recipe_tags module attribute to the AstroData tags of the first input element.

The recipe is obtained as follow, where ad is an AstroData object:

>>> from recipe_system.mappers.recipeMapper import RecipeMapper
>>> tags = ad.tags
>>> instpkg = ad.instrument(generic=True).lower()
>>> rmapper = RecipeMapper(tags, instpkg)
>>> recipe = rmapper.get_applicable_recipe()
>>> recipe.__name__
reduce

The recipe can then be run by passing it a primitive set (see the section above on primitive mapping):

>>> recipe(p)

4.2.1.1. Example

Let us discuss an example. Let us use a raw GMOS image.

 >>> ad = astrodata.open('S20161025S0111.fits')
 >>> ad.tags
set(['RAW', 'GMOS', 'GEMINI', 'SIDEREAL', 'UNPREPARED', 'IMAGE', 'SOUTH'])

The RecipeMapper uses these tags to search the geminidr packages, or the data reduction package specified in the mapper attribute drpkg. The first step in the search is to narrow it to the applicable instrument package, in this case, gmos. The mapper enforces the rule that the directory name for the instrument package be the lower case version of the AstroData descriptor instrument(). This speeds up the search but it is also a limitation.

Next, it narrows the search to the subdirectory recipes and then the subdirectory therein named after the mode. The default mode in RecipeMapper is sq. In our example, this means gmos/recipes/sq/

The RecipeMapper instance stores the name of the instrument, and thus the name of the instrument package in the pkg attribute, which is set upon instantiation from the .instrument() descriptor of the first AstroData object in the input list. For example, using the GMOS ad:

>>> from recipe_system.mappers.recipeMapper import RecipeMapper
>>> tags = ad.tags
>>> instpkg = ad.instrument(generic=True).lower()
>>> rmapper = RecipeMapper(tags, instpkg)
>>> recipe = rmapper.get_applicable_recipe()

The public method get_applicable_recipe(), when invoked, launches the search for the most appropriate recipe library. The search itself is focused on finding modules with a recipe_tags attribute.

>>> recipe = rmapper.get_applicable_recipe()

The function looks to match the recipe_tags of the recipe libraries it finds to the AstroData tags. A successful match requires recipe_tags to be a subset of the Astrodata tags. The match with the largest subset wins the contest. In our example, the recipe library recipes_IMAGE of mode sq of GMOS wins. In there, the default recipe is named reduce.

>>> recipe.__name__
reduce
>>> recipe.__module__
'geminidr.gmos.recipes.sq.recipes_IMAGE'

4.2.2. Selecting External (User) Recipes

It is possible to bypass the recipe mapping entirely and request the use of a user recipe library and recipe. It is simply a matter of setting the recipename to the path to the recipe library and to the name of the recipe when the RecipeMapper is instantiated. The RecipeMapper imports the file and returns the recipe function object.

While some users may have set their PYTHONPATH to include such arbitrary locations, which would allow the myrecipes module to be imported directly, most people will not have such paths in their PYTHONPATH, and would not be able to directly import their recipe file without modifying their environment. Using the RecipeMapper lets users avoid this hassle because it handles import transparently.

>>> rm = RecipeMapper(adinputs, recipename='/path/to/myrecipes.myreduce')
>>> recipefn = rm.get_applicable_recipe()
>>> recipefn.__name__
'myreduce'

Note that for user supplied recipe libraries and functions, the mode, the pkg, the drpkg, and the tags are irrelevant. Essentially, passing a user-defined recipe to the RecipeMapper tells the mapper, “do not search but use this.” In these cases, it is incumbent upon the users and developers to ensure that the external recipes specified are actually applicable to the datasets being processed.

4.2.3. Primitives and Recipes, Together at Last

To summarize the mapper usage described in this chapter, to launch a reduction one does:

>>> tags = ad.tags
>>> instpkg = ad.instrument(generic=True).lower()
>>> rmapper = RecipeMapper(tags, instpkg)
>>> pmapper = PrimitiveMapper(tags, instpkg)
>>> recipe = rmapper.get_applicable_recipe()
>>> pclass = pmapper.get_applicable_primitives()
>>> p = pclass([ad])
>>> recipe(p)

This is essentially what the Reduce class does.

4.3. Using the Primitives Directly

The primitves in the selected primitive set can be used directly. The recipe is just a function that calls some primitives from the set in some pre-defined order. This usage can be useful for testing, debugging or interactive use.

The primitive set is obtained as shown in the previous section, using the PrimitiveMapper and its get_applicable_primitives method. To retrieve the list of primitives from the primitive set, one can do this:

>>> import inspect
>>> for item in dir(p):
...    if not item.startswith('_') and \
...       inspect.ismethod(getattr(p, item)):
...       print(item)
...

The primitive set has been initialized with a list of AstroData objects. Running a primitive on them only requires calling the primitive as a method of the primitive set.

The exact call depends on the primitive itself. In geminidr, the primitives use a streams attribute to store the AstroData objects being processed in such a way that those objects do not need to be passed to the primitives when called. For example:

>>> p.prepare()
>>> p.addVAR()

The syntax above uses streams to pass the data along from one primitive to the next. The streams are attributes of the geminidr PrimitiveBASE class that all geminidr primitive classes inherit. When writing a new data reduction package, the use of that base class, and the parameter_override decorator in recipe_system.utils.decorators, is required to benefit from the streams system.

It is also recommended, and indeed the primitives in geminidr do that, to have the primitives methods return the modified output AstroData objects. When streams are used, return values are not necessary, but for debugging, testing, or exploration purposes it can be handy. For example, one can do this:

>>> intermediate_outputs = p.prepare()

A logger is currently required. The logger in gempy.utils called logutils is used by the recipe_system. The output will go to both stdout and a logfile. If the logfile is not defined, it leads double-printing of the logging on stdout. To avoid the double- printing the logfile name must be set to something. If you do not want to write a logfile to disk, on Unix systems you can set the file name to /dev/null.

>>> from gempy.utils import logutils
>>> logutils.config(file_name='/dev/null')

Here is what the stdout logging looks like when a primitive is run directly:

>>> p.prepare()
   PRIMITIVE: prepare
   ------------------
      PRIMITIVE: validateData
      -----------------------
      .
      PRIMITIVE: standardizeStructure
      -------------------------------
      .
      PRIMITIVE: standardizeHeaders
      -----------------------------
         PRIMITIVE: standardizeObservatoryHeaders
         ----------------------------------------
         Updating keywords that are common to all Gemini data
         .
         PRIMITIVE: standardizeInstrumentHeaders
         ---------------------------------------
         Updating keywords that are specific to GMOS
         .
      .
   .
   [<gemini_instruments.gmos.adclass.AstroDataGmos object at 0x11a12d650>]

Of interest when running the primitives in this way are the contents of the streams and params attributes of the primitive set.

The streams and params are features of the geminidr PrimitiveBASE class, the parameter_override decorator, and the pexconfig-based parameter system used in geminidr. For the time being, developers of new DR packages must use those systems.

The main stream that allows the data to be passed from one primitive to the other without the need of arguments is p.streams['main']. When the primitive set is instantiated by get_applicable_primitives, the input AstroData objects are stored in that stream. If for some reason you need to repopulate that stream, do something like this:

>>> new_inputs = [ad1, ad2, ... adn]
>>> p.streams['main'] = new_inputs

The input parameters to the primitives are stored in the primitive set in p.params. For example, to see the parameters to the prepare primitive and their current settings:

>>> p.params['prepare'].toDict()
OrderedDict([('suffix', '_prepared'), ('mdf', None), ('attach_mdf', True)])

Or, prettier

>>> print(*p.params['prepare'].toDict().items(), sep='\n')
('suffix', '_prepared')
('mdf', None)
('attach_mdf', True)

If you need guidance as to the recommended sequence of primitives, you can inspect the recipes returned by the RecipeMapper.

>>> from recipe_system.mappers.recipeMapper import RecipeMapper
>>> tags = ad.tags
>>> instpkg = ad.instrument(generic=True).lower()
>>> rmapper = RecipeMapper(tags, instpkg)
>>> recipe = rmapper.get_applicable_recipe()
>>> recipe.__name__
'reduce'
>>> import inspect
>>> print(inspect.getsource(recipe.__code__))
def reduce(p):
    """
    This recipe performs the standardization and corrections needed to
    convert the raw input science images into a stacked image.

    Parameters
    ----------
    p : PrimitivesBASE object
        A primitive set matching the recipe_tags.
    """

    p.prepare()
    p.addDQ()
    p.addVAR(read_noise=True)
    p.overscanCorrect()
    p.getProcessedBias()
    p.biasCorrect()
    p.ADUToElectrons()
    p.addVAR(poisson_noise=True)
    p.getProcessedFlat()
    p.flatCorrect()
    p.getProcessedFringe()
    p.fringeCorrect()
    p.mosaicDetectors()
    p.detectSources()
    p.adjustWCSToReference()
    p.resampleToCommonFrame()
    p.flagCosmicRaysByStacking()
    p.scaleByExposureTime()
    p.stackFrames(zero=True)
    p.storeProcessedScience()
    return