PipelineTask

class lsst.pipe.base.PipelineTask(*, config=None, log=None, initInputs=None, **kwargs)

Bases: lsst.pipe.base.Task

Base class for all pipeline tasks.

This is an abstract base class for PipelineTasks which represents an algorithm executed by framework(s) on data which comes from data butler, resulting data is also stored in a data butler.

PipelineTask inherits from a pipe.base.Task and uses the same configuration mechanism based on pex.config. PipelineTask sub-class typically implements run() method which receives Python-domain data objects and returns pipe.base.Struct object with resulting data. run() method is not supposed to perform any I/O, it operates entirely on in-memory objects. runQuantum() is the method (can be re-implemented in sub-class) where all necessary I/O is performed, it reads all input data from data butler into memory, calls run() method with that data, examines returned Struct object and saves some or all of that data back to data butler. runQuantum() method receives daf.butler.Quantum instance which defines all input and output datasets for a single invocation of PipelineTask.

Subclasses must be constructable with exactly the arguments taken by the PipelineTask base class constructor, but may support other signatures as well.

Parameters:
config : pex.config.Config, optional

Configuration for this task (an instance of self.ConfigClass, which is a task-specific subclass of PipelineTaskConfig). If not specified then it defaults to self.ConfigClass().

log : lsst.log.Log, optional

Logger instance whose name is used as a log name prefix, or None for no prefix.

initInputs : dict, optional

A dictionary of objects needed to construct this PipelineTask, with keys matching the keys of the dictionary returned by getInitInputDatasetTypes and values equivalent to what would be obtained by calling Butler.get with those DatasetTypes and no data IDs. While it is optional for the base class, subclasses are permitted to require this argument.

Attributes:
canMultiprocess : bool, True by default (class attribute)

This class attribute is checked by execution framework, sub-classes can set it to False in case task does not support multiprocessing.

Attributes Summary

canMultiprocess

Methods Summary

adaptArgsAndRun(inputData, inputDataIds, …) Run task algorithm on in-memory data.
emptyMetadata() Empty (clear) the metadata for this Task and all sub-Tasks.
getAllSchemaCatalogs() Get schema catalogs for all tasks in the hierarchy, combining the results into a single dict.
getDatasetTypes(config, configClass) Return dataset type descriptors defined in task configuration.
getFullMetadata() Get metadata for all tasks.
getFullName() Get the task name as a hierarchical name including parent task names.
getInitInputDatasetTypes(config) Return dataset type descriptors that can be used to retrieve the initInputs constructor argument.
getInitOutputDatasetTypes(config) Return dataset type descriptors that can be used to write the objects returned by getOutputDatasets.
getInitOutputDatasets() Return persistable outputs that are available immediately after the task has been constructed.
getInputDatasetTypes(config) Return input dataset type descriptors for this task.
getName() Get the name of the task.
getOutputDatasetTypes(config) Return output dataset type descriptors for this task.
getResourceConfig() Return resource configuration for this task.
getSchemaCatalogs() Get the schemas generated by this task.
getTaskDict() Get a dictionary of all tasks as a shallow copy.
makeField(doc) Make a lsst.pex.config.ConfigurableField for this task.
makeSubtask(name, **keyArgs) Create a subtask as a new instance as the name attribute of this task.
run(**kwargs) Run task algorithm on in-memory data.
runQuantum(quantum, butler) Execute PipelineTask algorithm on single quantum of data.
saveStruct(struct, outputDataRefs, butler) Save data in butler.
timer(name[, logLevel]) Context manager to log performance data for an arbitrary block of code.

Attributes Documentation

canMultiprocess = True

Methods Documentation

adaptArgsAndRun(inputData, inputDataIds, outputDataIds)

Run task algorithm on in-memory data.

This method is called by runQuantum to operate on input in-memory data and produce coressponding output in-memory data. It receives arguments which are dictionaries with input data and input/output DataIds. Many simple tasks do not need to know DataIds so default implementation of this method calls run method passing input data objects as keyword arguments. Most simple tasks will implement run method, more complex tasks that need to know about output DataIds will override this method instead.

All three arguments to this method are dictionaries with keys equal to the name of the configuration fields for dataset type. If dataset type is configured with scalar fiels set to True then it is expected that only one dataset appears on input or output for that dataset type and dictionary value will be a single data object or DataId. Otherwise if scalar is False (default) then value will be a list (even if only one item is in the list).

The method returns Struct instance with attributes matching the configuration fields for output dataset types. Values stored in returned struct are single object if scalar is True or list of objects otherwise. If tasks produces more than one object for some dataset type then data objects returned in struct must match in count and order corresponding DataIds in outputDataIds.

Parameters:
inputData : dict

Dictionary whose keys are the names of the configuration fields describing input dataset types and values are Python-domain data objects (or lists of objects) retrieved from data butler.

inputDataIds : dict

Dictionary whose keys are the names of the configuration fields describing input dataset types and values are DataIds (or lists of DataIds) that task consumes for corresponding dataset type. DataIds are guaranteed to match data objects in inputData

outputDataIds : dict

Dictionary whose keys are the names of the configuration fields describing output dataset types and values are DataIds (or lists of DataIds) that task is to produce for corresponding dataset type.

Returns:
struct : Struct

Standard convention is that this method should return Struct instance containing all output data. Struct attribute names should correspond to the names of the configuration fields describing task output dataset types. If something different is returned then saveStruct method has to be re-implemented accordingly.

emptyMetadata()

Empty (clear) the metadata for this Task and all sub-Tasks.

getAllSchemaCatalogs()

Get schema catalogs for all tasks in the hierarchy, combining the results into a single dict.

Returns:
schemacatalogs : dict

Keys are butler dataset type, values are a empty catalog (an instance of the appropriate lsst.afw.table Catalog type) for all tasks in the hierarchy, from the top-level task down through all subtasks.

Notes

This method may be called on any task in the hierarchy; it will return the same answer, regardless.

The default implementation should always suffice. If your subtask uses schemas the override Task.getSchemaCatalogs, not this method.

classmethod getDatasetTypes(config, configClass)

Return dataset type descriptors defined in task configuration.

This method can be used by other methods that need to extract dataset types from task configuration (e.g. getInputDatasetTypes or sub-class methods).

Parameters:
config : Config

Configuration for this task. Typically datasets are defined in a task configuration.

configClass : type

Class of the configuration object which defines dataset type.

Returns:
Dictionary where key is the name (arbitrary) of the output dataset
and value is the `DatasetTypeDescriptor` instance. Default
implementation uses configuration field name as dictionary key.
Returns empty dict if configuration has no fields with the specified
``configClass``.
getFullMetadata()

Get metadata for all tasks.

Returns:
metadata : lsst.daf.base.PropertySet

The PropertySet keys are the full task name. Values are metadata for the top-level task and all subtasks, sub-subtasks, etc..

Notes

The returned metadata includes timing information (if @timer.timeMethod is used) and any metadata set by the task. The name of each item consists of the full task name with . replaced by :, followed by . and the name of the item, e.g.:

topLevelTaskName:subtaskName:subsubtaskName.itemName

using : in the full task name disambiguates the rare situation that a task has a subtask and a metadata item with the same name.

getFullName()

Get the task name as a hierarchical name including parent task names.

Returns:
fullName : str

The full name consists of the name of the parent task and each subtask separated by periods. For example:

  • The full name of top-level task “top” is simply “top”.
  • The full name of subtask “sub” of top-level task “top” is “top.sub”.
  • The full name of subtask “sub2” of subtask “sub” of top-level task “top” is “top.sub.sub2”.
classmethod getInitInputDatasetTypes(config)

Return dataset type descriptors that can be used to retrieve the initInputs constructor argument.

Datasets used in initialization may not be associated with any Dimension (i.e. their data IDs must be empty dictionaries).

Default implementation finds all fields of type InitInputInputDatasetConfig in configuration (non-recursively) and uses them for constructing DatasetTypeDescriptor instances. The names of these fields are used as keys in returned dictionary. Subclasses can override this behavior.

Parameters:
config : Config

Configuration for this task. Typically datasets are defined in a task configuration.

Returns:
Dictionary where key is the name (arbitrary) of the input dataset
and value is the `DatasetTypeDescriptor` instance. Default
implementation uses configuration field name as dictionary key.
When the task requires no initialization inputs, should return an
empty dict.
classmethod getInitOutputDatasetTypes(config)

Return dataset type descriptors that can be used to write the objects returned by getOutputDatasets.

Datasets used in initialization may not be associated with any Dimension (i.e. their data IDs must be empty dictionaries).

Default implementation finds all fields of type InitOutputDatasetConfig in configuration (non-recursively) and uses them for constructing DatasetTypeDescriptor instances. The names of these fields are used as keys in returned dictionary. Subclasses can override this behavior.

Parameters:
config : Config

Configuration for this task. Typically datasets are defined in a task configuration.

Returns:
Dictionary where key is the name (arbitrary) of the output dataset
and value is the `DatasetTypeDescriptor` instance. Default
implementation uses configuration field name as dictionary key.
When the task produces no initialization outputs, should return an
empty dict.
getInitOutputDatasets()

Return persistable outputs that are available immediately after the task has been constructed.

Subclasses that operate on catalogs should override this method to return the schema(s) of the catalog(s) they produce.

It is not necessary to return the PipelineTask’s configuration or other provenance information in order for it to be persisted; that is the responsibility of the execution system.

Returns:
datasets : dict

Dictionary with keys that match those of the dict returned by getInitOutputDatasetTypes values that can be written by calling Butler.put with those DatasetTypes and no data IDs. An empty dict should be returned by tasks that produce no initialization outputs.

classmethod getInputDatasetTypes(config)

Return input dataset type descriptors for this task.

Default implementation finds all fields of type InputDatasetConfig in configuration (non-recursively) and uses them for constructing DatasetTypeDescriptor instances. The names of these fields are used as keys in returned dictionary. Subclasses can override this behavior.

Parameters:
config : Config

Configuration for this task. Typically datasets are defined in a task configuration.

Returns:
Dictionary where key is the name (arbitrary) of the input dataset
and value is the `DatasetTypeDescriptor` instance. Default
implementation uses configuration field name as dictionary key.
getName()

Get the name of the task.

Returns:
taskName : str

Name of the task.

See also

getFullName

classmethod getOutputDatasetTypes(config)

Return output dataset type descriptors for this task.

Default implementation finds all fields of type OutputDatasetConfig in configuration (non-recursively) and uses them for constructing DatasetTypeDescriptor instances. The keys of these fields are used as keys in returned dictionary. Subclasses can override this behavior.

Parameters:
config : Config

Configuration for this task. Typically datasets are defined in a task configuration.

Returns:
Dictionary where key is the name (arbitrary) of the output dataset
and value is the `DatasetTypeDescriptor` instance. Default
implementation uses configuration field name as dictionary key.
getResourceConfig()

Return resource configuration for this task.

Returns:
Object of type `~config.ResourceConfig` or ``None`` if resource
configuration is not defined for this task.
getSchemaCatalogs()

Get the schemas generated by this task.

Returns:
schemaCatalogs : dict

Keys are butler dataset type, values are an empty catalog (an instance of the appropriate lsst.afw.table Catalog type) for this task.

Notes

Warning

Subclasses that use schemas must override this method. The default implemenation returns an empty dict.

This method may be called at any time after the Task is constructed, which means that all task schemas should be computed at construction time, not when data is actually processed. This reflects the philosophy that the schema should not depend on the data.

Returning catalogs rather than just schemas allows us to save e.g. slots for SourceCatalog as well.

getTaskDict()

Get a dictionary of all tasks as a shallow copy.

Returns:
taskDict : dict

Dictionary containing full task name: task object for the top-level task and all subtasks, sub-subtasks, etc..

classmethod makeField(doc)

Make a lsst.pex.config.ConfigurableField for this task.

Parameters:
doc : str

Help text for the field.

Returns:
configurableField : lsst.pex.config.ConfigurableField

A ConfigurableField for this task.

Examples

Provides a convenient way to specify this task is a subtask of another task.

Here is an example of use:

class OtherTaskConfig(lsst.pex.config.Config)
    aSubtask = ATaskClass.makeField("a brief description of what this task does")
makeSubtask(name, **keyArgs)

Create a subtask as a new instance as the name attribute of this task.

Parameters:
name : str

Brief name of the subtask.

keyArgs

Extra keyword arguments used to construct the task. The following arguments are automatically provided and cannot be overridden:

  • “config”.
  • “parentTask”.

Notes

The subtask must be defined by Task.config.name, an instance of pex_config ConfigurableField or RegistryField.

run(**kwargs)

Run task algorithm on in-memory data.

This method should be implemented in a subclass unless tasks overrides adaptArgsAndRun to do something different from its default implementation. With default implementation of adaptArgsAndRun this method will receive keyword arguments whose names will be the same as names of configuration fields describing input dataset types. Argument values will be data objects retrieved from data butler. If a dataset type is configured with scalar field set to True then argument value will be a single object, otherwise it will be a list of objects.

If the task needs to know its input or output DataIds then it has to override adaptArgsAndRun method instead.

Returns:
struct : Struct

See description of adaptArgsAndRun method.

Examples

Typical implementation of this method may look like:

def run(self, input, calib):
    # "input", "calib", and "output" are the names of the config fields

    # Assuming that input/calib datasets are `scalar` they are simple objects,
    # do something with inputs and calibs, produce output image.
    image = self.makeImage(input, calib)

    # If output dataset is `scalar` then return object, not list
    return Struct(output=image)
runQuantum(quantum, butler)

Execute PipelineTask algorithm on single quantum of data.

Typical implementation of this method will use inputs from quantum to retrieve Python-domain objects from data butler and call adaptArgsAndRun method on that data. On return from adaptArgsAndRun this method will extract data from returned Struct instance and save that data to butler.

The Struct returned from adaptArgsAndRun is expected to contain data attributes with the names equal to the names of the configuration fields defining output dataset types. The values of the data attributes must be data objects corresponding to the DataIds of output dataset types. All data objects will be saved in butler using DataRefs from Quantum’s output dictionary.

This method does not return anything to the caller, on errors corresponding exception is raised.

Parameters:
quantum : Quantum

Object describing input and output corresponding to this invocation of PipelineTask instance.

butler : object

Data butler instance.

Raises:
`ScalarError` if a dataset type is configured as scalar but receives
multiple DataIds in `quantum`. Any exceptions that happen in data
butler or in `adaptArgsAndRun` method.
saveStruct(struct, outputDataRefs, butler)

Save data in butler.

Convention is that struct returned from run() method has data field(s) with the same names as the config fields defining output DatasetTypes. Subclasses may override this method to implement different convention for Struct content or in case any post-processing of data may be needed.

Parameters:
struct : Struct

Data produced by the task packed into Struct instance

outputDataRefs : dict

Dictionary whose keys are the names of the configuration fields describing output dataset types and values are lists of DataRefs. DataRefs must match corresponding data objects in struct in number and order.

butler : object

Data butler instance.

timer(name, logLevel=10000)

Context manager to log performance data for an arbitrary block of code.

Parameters:
name : str

Name of code being timed; data will be logged using item name: Start and End.

logLevel

A lsst.log level constant.

See also

timer.logInfo

Examples

Creating a timer context:

with self.timer("someCodeToTime"):
    pass  # code to time