dyval.DAG.describer¶

class promptbench.dyval.DAG.describer.BaseDAGDescriber(dag_obj, add_rand_desc=0, delete_desc=0)¶

Bases: object

Base class for creating descriptions of Directed Acyclic Graphs (DAGs).

This class provides methods to traverse and describe DAGs in various orders and to manipulate descriptions.

Parameters:¶

describe_question_node(node): Describes a node in the DAG for question formation.
generate_rand_description(): Generates random descriptions based on the DAG type.
topological_traversal(desc_func): Traverses the DAG in topological order and applies a description function.
reverse_topological_traversal(topo_desc): Reverses the order of topological descriptions.
random_traversal(topo_desc): Shuffles the topological descriptions randomly.
add_rand_description(desc): Adds random descriptions to the existing description list.
delete_description(desc): Deletes descriptions from the existing description list.
_describe_question(): Describes the DAG for question formation in various traversal orders.

class promptbench.dyval.DAG.describer.GeneralDAGDescriber(dag_obj, add_rand_desc=0, delete_desc=0)¶

A describer class for GeneralDAG instances.

Inherits from BaseDAGDescriber and provides specific implementations for describing GeneralDAG nodes.

dag_objGeneralDAG: The GeneralDAG instance to describe.
add_rand_descint, optional: The number of random descriptions to add (inherited).
delete_descint, optional: The number of descriptions to delete (inherited).

describe_question_node(node): Provides a description for a GeneralDAG node.
generate_rand_description(): Generates random descriptions specific to GeneralDAG.
describe_answer(): Describes the DAG for answer formation (not implemented yet).

class promptbench.dyval.DAG.describer.TreeDAGDescriber(dag_obj, add_rand_desc=0, delete_desc=0, trainset=False)¶

A describer class for TreeDAG instances.

Inherits from BaseDAGDescriber and provides specific implementations for describing TreeDAG nodes.

dag_objTreeDAG: The TreeDAG instance to describe.
add_rand_descint, optional: The number of random descriptions to add (inherited).
delete_descint, optional: The number of descriptions to delete (inherited).
trainsetbool: Indicates if the describer is used for training set generation.

describe_inference_node(node): Provides a description for a TreeDAG node for inference.
generate_rand_description(): Generates random descriptions specific to TreeDAG.
describe_inference_steps(): Describes the inference steps based on the DAG’s topology.
describe_answer(): Provides the answer based on the root value of the TreeDAG.
describe_question(): Describes the DAG for question formation in various traversal orders.