agatha.ml.hypothesis_predictor.predicate_util module

class agatha.ml.hypothesis_predictor.predicate_util.NegativePredicateGenerator(coded_terms, graph)

Bases: object

generate()

class agatha.ml.hypothesis_predictor.predicate_util.PredicateEmbeddings(subj: numpy.array, obj: numpy.array, subj_neigh: List[numpy.array], obj_neigh: List[numpy.array])

Bases: object

obj: np.array = None

obj_neigh: List[np.array] = None

subj: np.array = None

subj_neigh: List[np.array] = None

class agatha.ml.hypothesis_predictor.predicate_util.PredicateObservationGenerator(graph, embeddings, neighbor_sample_rate)

Bases: object

Converts predicate names to predicate observations

class agatha.ml.hypothesis_predictor.predicate_util.PredicateScrambleObservationGenerator(predicates, *args, **kwargs)

Bases: agatha.ml.hypothesis_predictor.predicate_util.PredicateObservationGenerator

Same as above, but the neighborhood comes from randomly selected predicates

agatha.ml.hypothesis_predictor.predicate_util.clean_coded_term(term)

If term is not formatted as an agatha coded term key, produces a coded term key. Otherwise, just returns the term.

Return type

str

agatha.ml.hypothesis_predictor.predicate_util.collate_predicate_embeddings(predicate_embeddings)

agatha.ml.hypothesis_predictor.predicate_util.collate_predicate_training_examples(examples)

Takes a list of results from PredicateExampleDataset and produces tensors for input into the agatha training model.

Return type

Dict[str, Any]

agatha.ml.hypothesis_predictor.predicate_util.is_valid_predicate_name(predicate_name)

Return type

bool

agatha.ml.hypothesis_predictor.predicate_util.parse_predicate_name(predicate_name)

Parses subject and object from predicate name strings.

Predicate names are formatted strings that follow this convention: p:{subj}:{verb}:{obj}. This function extracts the subject and object and returns coded-term names in the form: m:{entity}. Will raise an exception if the predicate name is improperly formatted.

Parameters

predicate_name (str) – Predicate name in form p:{subj}:{verb}:{obj}.

Return type

Tuple[str, str]

Returns

The subject and object formulated as coded-term names.

agatha.ml.hypothesis_predictor.predicate_util.to_predicate_name(subj, obj, verb='unknown')

Converts two names into a predicate of form p:t1:verb:t2

Assumes that terms are correct Agatha graph keys. This means that we expect input terms in the form of m:____. Allows for a custom verb type, but defaults to unknown. Output will always be set to lowercase.

Example usage:

` to_predicate_name(m:c1, m:c2) > p:c1:unknown:c2 to_predicate_name(m:c1, m:c2, "treats") > p:c1:treats:c2 to_predicate_name(m:c1, m:c2, "TREATS") > p:c1:treats:c2 `

Parameters

• subj (str) – Subject term. In the form of “m:_____”

• obj (str) – Object term. In the form of “m:_____”

• verb (str) – Optional verb term for resulting predicate.

Return type

str

Returns

Properly formatted predicate containing subject and object. Verb type will be set to “UNKNOWN”