The purpose of the data model is to abstract away the peculiarities of the underlying file format. The same data model may be used for data created from scratch in memory, or loaded from FITS or ASDF files or some future file format.
Hierarchy of models¶
There are different data model classes for different kinds of data.
One model instance, many arrays¶
Each model instance generally has many arrays that are associated with
it. For example, the
ImageModel class has the following arrays
associated with it:
data: The science data
dq: The data quality array
err: The error array
The shape of these arrays must be broadcast-compatible. If you try to assign an array to one of these members that is not broadcast-compatible with the data array, an exception is raised.
Working with models¶
Creating a data model from scratch¶
To create a new
ImageModel, just call its constructor. To create a
new model where all of the arrays will have default values, simply
provide a shape as the first argument:
from jwst.datamodels import ImageModel with ImageModel((1024, 1024)) as im: ...
In this form, the memory for the arrays will not be allocated until the arrays are accessed. This is useful if, for example, you don’t need a data quality array – the memory for such an array will not be consumed:
# Print out the data array. It is allocated here on first access # and defaults to being filled with zeros. print(im.data)
If you already have data in a numpy array, you can also create a model using that array by passing it in as a data keyword argument:
data = np.empty((50, 50)) dq = np.empty((50, 50)) with ImageModel(data=data, dq=dq) as im: ...
Creating a data model from a file¶
jwst.datamodels.open function is a convenient way to create a
model from a file on disk. It may be passed any of the following:
- a path to a FITS file
- a path to an ASDF file
- a readable file-like object
The file will be opened, and based on the nature of the data in the
file, the correct data model class will be returned. For example, if
the file contains 2-dimensional data, an
ImageModel instance will be
returned. You will generally want to instantiate a model using a
with statement so that the file will be closed automatically when
from jwst import datamodels with datamodels.open("myimage.fits") as im: assert isinstance(im, datamodels.ImageModel)
If you know the type of data stored in the file, or you want to ensure that what is being loaded is of a particular type, use the constructor of the desired concrete class. For example, if you want to ensure that the file being opened contains 2-dimensional image data:
from jwst.datamodels import ImageModel with ImageModel("myimage.fits") as im: # raises exception if myimage.fits is not an image file pass
This will raise an exception if the file contains data of the wrong shape.
Saving a data model to a file¶
Simply call the
save method on the model instance. The format to
save into will either be deduced from the filename (if provided) or
format keyword argument:
save always clobbers the output file.
It also accepts a writable file-like object (opened in binary mode). In that case, a format must be specified:
with open("myimage.fits", "wb") as fd: im.save(fd, format="fits")
Copying a model¶
To create a new model based on another model, simply use its
method. This will perform a deep-copy: that is, no changes to the
original model will propagate to the new model:
new_model = old_model.copy()
It is also possible to copy all of the known metadata from one model into a new one using the update method:
Models contain a list of history records, accessed through the
history attribute. This is just an ordered list of strings –
nothing more sophisticated.
To get to the history:
To add an entry to the history:
model.history.append("Processed through the frobulator step")
These history entries are stored in
HISTORY keywords when saving
to FITS format.
This section describes how to port code that uses
Opening a file¶
from jwst.datamodels import ImageModel with ImageModel("myfile.fits") as model: ...
In place of
ImageModel, use the type of data one expects to find in
the file. For example, if spectrographic data is expected, use
SpecModel. If it doesn’t matter (perhaps the application is only
sorting FITS files into categories) use the base class
An alternative is to use:
from jwst import datamodels with datamodels.open("myfile.fits") as model: ...
datamodels.open() method checks if the
DATAMODL FITS keyword has
been set, which records the DataModel that was used to create the file.
If the keyword is not set, then
datamodels.open() does its best to
guess the best DataModel to use.
Data should be accessed through one of the pre-defined data members on
the model (
err). There is no longer a need to hunt
through the HDU list to find the data.
The data model hides direct access to FITS header keywords. Instead, use the Metadata tree.
There is a convenience method,
find_fits_keyword to find where a
FITS keyword is used in the metadata tree:
>>> from jwst.datamodels import DataModel # First, create a model of the desired type >>> model = DataModel() >>> model.find_fits_keyword('DATE-OBS') [u'meta.observation.date']
This information shows that instead of:
Extra FITS keywords¶
When loading arbitrary FITS files, there may be keywords that are not
listed in the schema for that data model. These “extra” FITS keywords
are put under the model in the
_extra_fits namespace is a section for each header data
unit, and under those are the extra FITS keywords. For example, if
the FITS file contains a keyword
FOO in the primary header, its
value can be obtained using:
This feature is useful to retain any extra keywords from input files to output products.
To get a list of everything in
returns a dictionary of of the instance at the model._extra_fits node.
_instance can be used at any node in the tree to return a dictionary
of rest of the tree structure at that node.