def testGraphReloading(self, predictor_str):
"""Reloading the graph should not affect values."""
predictor = self._get_predictor(predictor_str)
if predictor_str == DEFAULT:
new_predictor = holparam_predictor.HolparamPredictor(
self.default_ckpt)
elif predictor_str == TAC_DEP:
new_predictor = holparam_predictor.TacDependentPredictor(
self.tac_dep_ckpt)
else:
raise ValueError('Unknown predictor string: %s' % predictor_str)
self.assertAllEqual(
predictor.batch_goal_embedding(self.formulas),
new_predictor.batch_goal_embedding(self.formulas),
'Reloading the graph should not change goal embeddings.')
self.assertAllEqual(
predictor.batch_thm_embedding(self.formulas),
new_predictor.batch_thm_embedding(self.formulas),
'Reloading the graph should not change theorem embeddings.')
self.assertAllEqual(
predictor.batch_tactic_scores(self.embeddings),
new_predictor.batch_tactic_scores(self.embeddings),
'Reloading the graph should not change tactic scores.')
self.assertAllEqual(
predictor.batch_thm_scores(self.embeddings[0],
self.embeddings,
tactic_id=self.tactic_id),
new_predictor.batch_thm_scores(self.embeddings[0],
self.embeddings,
tactic_id=self.tactic_id),
'Reloading the graph should not change theorem scores.')
def setUpClass(cls):
"""Restoring the graph takes a lot of time, so we do it only once here."""
cls.default_ckpt = test_util.test_src_dir_path(DEFAULT_TEST_PATH)
cls.default_predictions = holparam_predictor.HolparamPredictor(
cls.default_ckpt)
cls.tac_dep_ckpt = test_util.test_src_dir_path(TAC_DEP_TEST_PATH)
cls.tac_dep_predictions = holparam_predictor.TacDependentPredictor(
cls.tac_dep_ckpt)
def setUpClass(cls):
"""Restoring the graph takes a lot of memory, so we do it only once here."""
cls.predictor = holparam_predictor.HolparamPredictor(
PREDICTIONS_MODEL_PREFIX)
cls.model_architecture = deephol_pb2.ProverOptions.PAIR_DEFAULT
cls.theorem_database = proof_assistant_pb2.TheoremDatabase()
cls.theorem_database.theorems.add(name='EQ_REFL', conclusion=EQ_REFL)
cls.theorem_database.theorems.add(name='EQ_SYM', conclusion=EQ_SYM)
cls.test_goal_index = 1
cls.test_theorem = cls.theorem_database.theorems[cls.test_goal_index]
cls.test_goal = cls.theorem_database.theorems[cls.test_goal_index]
cls.test_premise_set = prover_util.make_premise_set(
cls.test_theorem, 'default')
cls.options = deephol_pb2.ActionGeneratorOptions()
def get_predictor(
options: deephol_pb2.ProverOptions) -> predictions.Predictions:
"""Returns appropriate predictor based on prover options."""
model_arch = options.model_architecture
if model_arch == deephol_pb2.ProverOptions.PAIR_DEFAULT:
return holparam_predictor.HolparamPredictor(
str(options.path_model_prefix))
if model_arch == deephol_pb2.ProverOptions.PARAMETERS_CONDITIONED_ON_TAC:
return holparam_predictor.TacDependentPredictor(
str(options.path_model_prefix))
if (model_arch == deephol_pb2.ProverOptions.HIST_AVG
or model_arch == deephol_pb2.ProverOptions.HIST_CONV
or model_arch == deephol_pb2.ProverOptions.HIST_ATT):
raise NotImplementedError(
'History-dependent model %s is not supported in the prover.' %
model_arch)
raise AttributeError('Unknown model architecture in prover options: %s' %
model_arch)
def _get_new_predictor(self): return holparam_predictor.HolparamPredictor(self.checkpoint)
def setUpClass(cls):
"""Restoring the graph takes a lot of time, so we do it only once here."""
super(HolparamPredictorTest, cls).setUpClass()
cls.checkpoint = test_util.test_src_dir_path(DEFAULT_TEST_PATH)
cls.predictor = holparam_predictor.HolparamPredictor(cls.checkpoint)
def start_bundle(self):
logging.info('Initializing the batching predictor...')
self.predictor = holparam_predictor.HolparamPredictor(
self.chkpt, max_embedding_batch_size=self.batch_size)
logging.info('Initializing the embedding store...')
self.emb_store = embedding_store.TheoremEmbeddingStore(self.predictor)