Numina Pipeline Example

This guide is intended as an introductory overview of the creation of instrument reduction pipelines with Numina. For detailed reference documentation of the functions and classes contained in the package, see the Numina Reference.

Warning

This “Pipeline Creation Guide” is still a work in progress; some of the material is not organized, and several aspects of Numina are not yet covered sufficient detail.

Execution environment of the Recipes

Recipes have different execution environments. Some recipes are designed to process observing modes required for the observation. These modes are related to visualization, acquisition and focusing. The Recipes are integrated in the GTC environment. We call these recipes the Data Factory Pipeline, (DFP).

Other group of recipes are devoted to scientific observing modes: imaging, spectroscopy and auxiliary calibrations. These Recipes constitute the Data Reduction Pipeline, (DRP). The software is meant to be standalone, users shall download the software and run it in their own computers, with reduction parameters and calibrations provided by the instrument team.

Users of the DRP will use the simple Numina CLI. Users of the DFP shall interact with the software through the GTC Inspector.

Instrument Reduction Pipeline Example

In the following sections we create an Instrument Reduction Pipeline for an instrument name CLODIA.

In order to make a new Instrument Reduction Pipeline visible to Numina and the GTC Control System you have to create a full Python package that will contain the reduction recipes, data types and other processing code.

The creation of Python packages is described in detail (for example) in the Python Packaging User Guide.

Then, we create a Python package called clodiadrp with the following structure (we ignore files such as README or LICENSE as they are not relevant here):

clodiadrp
|-- clodiadrp
|   |-- __init__.py
|-- setup.py

From here the steps are:

  1. Create a configuration yaml file.
  2. Create a loader file.
  3. Link the entry_point option in the setup.py with the loader file.
  4. Create the Pipeline’s Recipes.

In the following we will continue with the same example as previously.

Configuration File

The configuration file contains basic information such as:

  • the list of modes of the instrument
  • the list of recipes of the instrument
  • the mapping between recipes and modes.

In this example, we assume that CLODIA has three modes: Bias, Flat and Image. The first two modes are used for pedestal and flat-field illumination correction. The third is the main scientific mode of the instrument.

Create a new yaml file in the root folder named drp.yaml.

name: CLODIA
configurations:
  default: {}
modes:
 -key: bias
  name: Bias
  summary: Bias mode
  description: >
    Full description of the Bias mode
 -key: flat
  name: Flat
  summary: Flat mode
  description: >
    Full description of the Flat mode
 -key: image
  name: Image
  summary: Image mode
  description: >
    Full description of the Image mode
pipelines:
  default:
    version: 1
    recipes:
      bias: clodiadrp.recipes.recipe
      flat: clodiadrp.recipes.recipe
      image: clodiadrp.recipes.recipe

The entry modes contains a list of the observing modes of the instrument. There are three: Bias, Flat and Image. Each entry contains information about the mode. A name, a short summary and a multi-line description. The field key is used to map the observing modes and the recipes, so key has to be unique and equal to only one value in each recipes block under pipelines.

The entry pipelines contains only one pipeline, called default by convention. The pipeline contains recipes, each related to one observing mode by means of the filed key. For the moment we haven’t developed any recipe, so the value of each key (clodiadrp.recipes.recipe) doesn’t exist yet.

Note

This file has to be included in package_data inside setup.py to be distributed with the package, see Installing Package Data for details.

Loader File

Create a new loader file in the root folder named loader.py with the following information:

import numina.core

def drp_load():
    """Entry point to load CLODIA DRP."""
    return numina.core.drp_load('clodiadrp', 'drp.yaml')

Create entry point

Once we have created the loader.py file, the only thing we have to do is to make CLODIA visible to Numina/GCS. To do so, just modify the setup.py file to add an entry point.

from setuptools import setup

setup(name='clodiadrp',
      entry_points = {
        'numina.pipeline.1': ['CLODIA = clodiadrp.loader:drp_load'],
        },
)

Both the Numina CLI tool and GCS check this particular entry point. They call the function provided by the entry point. The function drp_load() reads and parses the YAML file and creates an object of class InstrumentDRP for each recipes it founds. These objects are used by Numina CLI and GCS to discover the available Instrument Reduction Pipelines.

At this stage, the file layout is as follows:

clodiadrp
|-- clodiadrp
|   |-- __init__.py
|   |-- loader.py
|   |-- drp.yaml
|-- setup.py

Note

In fact, it is not necessary to use a YAML file to contain the Instrument information. The only strict requirement is that the function in the entry point ‘numina.pipeline.1’ must return a valid InstrumentDRP object. The use of a YAML file and the drp_load() function is only a matter of convenience.

Recipes Creation

We haven’t created any reduction recipe yet. As a matter of organization, we suggest to create a dedicated subpackage for recipes clodiadrp.recipes and a module for each recipe. The file layout is:

clodiadrp
|-- clodiadrp
|   |-- __init__.py
|   |-- loader.py
|   |-- drp.yaml
|   |-- recipes
|   |   |-- __init__.py
|   |   |-- bias.py
|   |   |-- flat.py
|   |   |-- image.py
|-- setup.py

Recipes must provide three things: 1) a description of the inputs of the recipe; 2) a description of the products of the recipe and 3) a run method which is in charge of executing the proccessing. Additionally, all Recipes must inherit from BaseRecipe.

We start with a simple Bias recipe. Its purpose is to process images previously taken in Bias mode, that is, a series of pedestal images. The recipe will receive the result of the observation and return a master bias image.

from numina.core import Result, Requirement
from numina.core import DataFrameType
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe

class Bias(BaseRecipe):                            (1)

    obresult = Requirement(ObservationResultType, "Observation Result")  (2)
    master_bias = Result(DataFrameType)           (3)

    def run(self, rinput):                         (4)

        # Here the raw images are processed
        # and a final image myframe is created

        result = self.create_result(master_bias=myframe)  (5)
        return result
  1. Each recipe must be a class derived from BaseRecipe
  2. This recipe only requires the result of the observation. Each requirement is an object of the Requirement class or any subclass of it. The type of the requirement is ObservationResultType, representing the result of the observation.
  3. This recipe only produces one result. Each product is an object of Result class. The type of the product is given by DataFrameType, representing an image.
  4. Each recipe must provide a run method. The method has only one argument that collects the values of all inputs declared by the recipe. In this case, rinput has a member named obresult and can be accessed through rinput.obresult which belongs to ObservationResult class.
  5. The recipe must return an object that collects all the declared products of the recipe, of RecipeResult class. This is accomplished internally by the create_result method. It will raise a run time exception if any of the declared products are not provided.

We can now create the Flat recipe (inside flat.py). This recipe has two requirements, the observation result and a master bias image (flat-field images require bias subtraction).

from numina.core import Result, Requirement
from numina.core import DataFrameType
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe

class Flat(BaseRecipe):

    obresult = Requirement(ObservationResultType, "Observation Result")  (1)
    master_bias = Requirement(DataFrameType, "Master Bias")      (2)
    master_flat = Result(DataFrameType)

    def run(self, rinput):                          (3)

        # Here the raw images are processed
        # and a final image myframe is created

        result = self.create_result(master_flat=myframe)  (4)
        return result
  1. This recipe only requires the result of the observation. Each requirement is an object of the Requirement class or any subclass of it. The type of the requirement is ObservationResultType, representing the result of the observation.
  2. It also requires a master bias image which belongs to DataFrameType class (represents an image).
  3. In this case, rinput has two members: 1) rinput.obresult of ObservationResult class and 2) a rinput.master_bias of DataFrame class
  4. The arguments of create_result must be the same names used in the product definition.

Finally, the recipe for Image mode reduction (inside image.py) has three requirements, the observation result, a master bias and a master flat images

from numina.core import Result, Requirement
from numina.core import DataFrameType
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe

class Image(BaseRecipe):

    obresult = Requirement(ObservationResultType)
    master_bias = Requirement(DataFrameType)
    master_flat = Requirement(DataFrameType)
    final = Result(DataFrameType)

    def run(self, rinput):                          (1)

        # Here the raw images are processed
        # and a final image myframe is created

        result = self.create_result(final=myframe)
        return result
  1. In this case, rinput will have three members rinput.obresult of ObservationResult class, rinput.master_bias of DataFrame class and rinput.master_flat of DataFrame class.

Note

It is not strictly required that the requirements and products names are consistent between recipes, although it is highly recommended.

Now we must update drp.yaml to insert the full name of the recipes (package and class), as follows

name: CLODIA
configurations:
  default: {}
modes:
 -key: bias
  name: Bias
  summary: Bias mode
  description: >
    Full description of the Bias mode
 -key: flat
  name: Flat
  summary: Flat mode
  description: >
    Full description of the Flat mode
 -key: image
  name: Image
  summary: Image mode
  description: >
    Full description of the Image mode
pipelines:
  default:
    version: 1
    recipes:
      bias: clodiadrp.recipes.bias.Bias
      flat: clodiadrp.recipes.flat.Flat
      image: clodiadrp.recipes.image.Image

Specialized data products

There is some information that is missing of our current setup. The products of some recipes are the inputs of others. The master bias created by Bias is the input that Flat and Image require. To represent this situation we use specialized data products. We start by adding a new module products:

clodiadrp
|-- clodiadrp
|   |-- __init__.py
|   |-- loader.py
|   |-- products.py
|   |-- drp.yaml
|   |-- recipes
|   |   |-- __init__.py
|   |   |-- bias.py
|   |   |-- flat.py
|   |   |-- image.py
|-- setup.py

We have two types of images that are products of recipes that can be required by other recipes: master bias and master flat. We represent this by creating two new types derived from DataFrameType (because the new types are images) and DataProductMixin (because the new types are products that must be handled by both Numina CLI and GTC Control system) classes.

from numina.types.frame import DataFrameType
from numina.types.product import DataProductMixin

class MasterBias(DataFrameType, DataProductMixin):
    pass


class MasterFlat(DataFrameType, DataProductMixin):
    pass

Now we must modify our recipes as follows. First Bias

from numina.core import Result, Requirement
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe
from clodiadrp.products import MasterBias   (1)

class Bias(BaseRecipe):

    obresult = Requirement(ObservationResultType)
    master_bias = Result(MasterBias)        (2)

    ...                                       (3)
  1. Import the new type MasterBias.
  2. Declare that our recipe produces MasterBias images.
  3. run method remains unchanged.

Then Flat:

from numina.core import Result, Requirement
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe
from clodiadrp.products import MasterBias, MasterFlat

class Flat(BaseRecipe):

    obresult = Requirement(ObservationResultType)
    master_bias = Requirement(MasterBias)         (1)
    master_flat = Result(MasterFlat)             (2)

    ...                                       (3)
  1. MasterBias is used as a requirement. This guaranties that the images provided here are those created by Bias and no other.
  2. Declare that our recipe produces MasterFlat images.
  3. run method remains unchanged.

And finally Image:

from numina.core import Result, Requirement
from numina.core import DataFrameType
from numina.types.obsresult import ObservationResultType
from numina.core.recipes import BaseRecipe
from clodiadrp.products import MasterBias, MasterFlat

class Image(BaseRecipe):

    obresult = Requirement(ObservationResultType)
    master_bias = Requirement(MasterBias)       (1)
    master_flat = Requirement(MasterFlat)       (2)
    final = Result(DataFrameType)              (3)

    ...                                       (4)
  1. MasterBias is used as a requirement. This guaranties that the images provided here are those created by Bias and no other.
  2. MasterFlat is used as a requirement. This guaranties that the images provided here are those created by Flat and no other.
  3. Declare that our recipe produces Image images.
  4. run method remains unchanged.