MCPcopy Create free account
hub / github.com/google-research/language / construct_membership

Method construct_membership

language/emql/data_loader.py:271–305  ·  view source on GitHub ↗

Construct sets from KB. A natural set is defined as a set of entities that share the same property, e.g. movies directed by Nolan, or actors appear in Inception. Args: train_split: percentage of data for training Returns: a list of train sets and test sets

(
      self, train_split)

Source from the content-addressed store, hash-verified

269 self.construct_query2box(name)
270
271 def construct_membership(
272 self, train_split):
273 """Construct sets from KB.
274
275 A natural set is defined as a set of entities that share the same
276 property, e.g. movies directed by Nolan, or actors appear in Inception.
277
278 Args:
279 train_split: percentage of data for training
280
281 Returns:
282 a list of train sets and test sets
283 """
284 train_data = list()
285 test_data = list()
286
287 for unused_subj_id, subj in self.id2entity.items():
288 all_objs = set()
289 for _, objs in self.kb[subj].items():
290 all_objs.update(objs)
291 # Group entities by graph structure. Entities within distance one
292 # are considered as an element in the set.
293 if len(all_objs) > 1:
294 one_data = all_objs
295 else:
296 continue
297
298 if random.random() < train_split:
299 train_data.append(one_data)
300 else:
301 test_data.append(one_data)
302
303 tf.logging.info('set examples: %d train, %d test', len(train_data),
304 len(test_data))
305 return train_data, test_data
306
307 def construct_set_pairs(
308 self, train_split):

Callers 1

construct_datasetMethod · 0.95

Calls

no outgoing calls

Tested by

no test coverage detected