Skip to content

Documentation of exe_kg_lib.classes.tasks package

Overview

This package contains classes that correspond to entities of type owl:class that are either rdfs:subClassOf ds:Task or rdfs:subClassOf ds:AtomicTask in the KG. In either case, these entities are in the top-level of the Task hierarchy for each of the three KG schemata: ML, Statistics and Visualization.

This package's classes implement the abstract run_method() to perform the following steps:

  1. The input data are taken:
    • Either from outputs of previous Tasks (parameter: other_task_output_dict) of the ExeKG
    • Or a given dataframe (parameter: input_data) that holds the input data for the ExeKG
  2. An algorithm is executed. The algorithm can be related to ML, Statistics or Visualization, depending on the Python file's prefix (i.e. ml, statistic, visual). The algorithm can:
    • Either be implemented as part of this library
    • Or belong to an external module. In this case, the module is determined using classes.tasks.task.Task.resolve_module() based on the Task's method_module_chain. See section Naming conventions for more info on method_module_chain.
  3. The output of the algorithm is returned as a dictionary with pairs of output name and value

Naming conventions

The below naming conventions are necessary for automatically mapping KG's tasks (with methods and properties) to Python objects while parsing the ExeKG.

  • Each class name in this package is the name of an owl:class that is either rdfs:subClassOf ds:Task or rdfs:subClassOf ds:AtomicTask.
  • The method_params_dict and method_inherited_params_dict fields inherited from classes.tasks.task.Task contain parameters for the algorithm to be executed.
    • Their keys are produced by applying utils.string_utils.property_iri_to_field_name() to the datatype property names of the Task's linked ds:AtomicMethod instance in the ExeKG. E.g. a key named split_ratio corresponds to hasParamSplitRatio property in the KG.
    • Their values are produced by applying classes.exe_kg_mixins.exe_kg_execution_mixin.ExeKGExecutionMixin._literal_to_field_value() to the literal values of the datatype properties in the ExeKG. E.g. a value of 0.6 corresponds to "0.2"^^xsd:float literal value in the KG.

  • The method_module_chain field inherited from classes.tasks.task.Task contains a hierarchy list of Python module names from top to bottom.

    • The module hierarchy is determined by utils.query_utils.get_module_hierarchy_chain() starting from the Task's linked ds:AtomicMethod instance in the ExeKG, and proceeding via the rdfs:subClassOf+ ds:Module property path.
    • Each item in the hierarchy list (except for the last one) comes from the name of a owl:class that is a rdfs:subClassOf+ ds:Module, after conversion by utils.string_utils.class_name_to_module_name().
    • The last item of the list comes from the type of the Task's linked ds:AtomicMethod instance, after conversion by utils.string_utils.class_name_to_method_name().

    The below example shows the module chain SVCMethod -> SvmModule -> SklearnModule which leads to method_module_chain = ["sklearn", "svm", "SVC"]. 

    #############################
### START: ExeKG fragment ###
#############################
ml:BinaryClassification1 a ml:BinaryClassification ;
                          ds:hasNextTask ml:Test1 ;
                          ml:hasBinaryClassificationMethod ml:SVCMethod1 ;
                          ml:hasTrainInput ml:DataInTrainX_BinaryClassification1_1,
                                          ml:DataInTrainY_BinaryClassification1_1 ;
                                          ml:hasTrainOutput ml:DataOutTrainModelSVCMethod .
###########################
### END: ExeKG fragment ###
###########################

#################################
### START: KG schema fragment ###
#################################
ml:SVCMethod a owl:Class ;
              rdfs:subClassOf ds:AtomicMethod,
                              ml:SvmModule,
                              ml:TrainMethod .

ml:SvmModule a owl:Class ;
              rdfs:subClassOf ml:SklearnModule .

ml:SklearnModule a owl:Class ;
                  rdfs:subClassOf ds:Module .
###############################
### END: KG schema fragment ###
###############################

  • The inputs and outputs fields inherited from classes.tasks.task.Task contain a list of classes.data_entity.DataEntity objects. In the case of inputs, the objects can also be of type classes.method.Method. The objects are generated by invoking the method _property_value_to_field_value() from the ExeKGExecutionMixin class. This method is applied to instances that are linked through a subclass of either ds:hasInput or ds:hasOutput to the Task in the ExeKG.

    • The field names of DataEntity and Method objects are filled by applying utils.string_utils.property_iri_to_field_name() to the properties of the Task's linked ds:DataEntity or ds:Method instances.
    • In the case of input DataEntities, the object's fields are filled using the properties of the ds:DataEntity instances that are referenced by the Task's linked ds:DataEntity instances.

    The below example shows a LinePlotting1 task instance that has DataInToPlot_LinePlotting1_1 as input. DataInToPlot_LinePlotting1_1 references ds:feature_1. So, in this case, the inputs field of the corresponding Task Python object will contain a DataEntity object with fields: source = "feature_1", reference = IRI(ds:feature_1). The fields data_semantics and data_structure are mainly used during pipeline construction. 

    ######################
### ExeKG fragment ###
######################

visu:LinePlotting1
    a                          visu:LinePlotting ;
    ds:hasNextTask             visu:LinePlotting2 ;
    visu:hasLinePlottingMethod visu:PlotMethod1 ;
    visu:hasPlottingInput      visu:DataInToPlot_LinePlotting1_1 .

visu:DataInToPlot_LinePlotting1_1
    a               visu:DataInToPlot ;
    ds:hasReference ds:feature_1 .

ds:feature_1
    a            ds:DataEntity, ds:Numerical, ds:Vector ;
    ds:hasSource "feature_1"^^xsd:string .