def test_fb():
cvsm = CompositionalVectorAlgorithm("freebase",
"data/fb15k237/cvsm_entity",
entity_type2vec_filename=None,
pooling_method="sat",
attention_method="sat",
early_stopping_metric="map")
cvsm.train_and_test()
def test_wn():
cvsm = CompositionalVectorAlgorithm(
"wordnet",
experiment_dir="data/wn18rr/cvsm_entity",
entity_type2vec_filename="data/wn18rr/entity_type2vec.pkl",
pooling_method="sat",
attention_method="sat",
early_stopping_metric="map")
cvsm.train_and_test()
process_paths(
input_dir=os.path.join(CVSM_RET_DIR, "data/data_input"),
output_dir=os.path.join(CVSM_RET_DIR, "data/data_output"),
vocab_dir=os.path.join(CVSM_RET_DIR, "data/vocab"),
isOnlyRelation=False,
getOnlyRelation=False,
MAX_POSSIBLE_LENGTH_PATH=
8, # the max number of relations in a path + 1
NUM_ENTITY_TYPES_SLOTS=
15, # the number of types + 1 (the reason we +1 is to create a meaningless type for all entities)
pre_padding=True)
# 5. Run the model
# use $tensorboard --logdir runs to see the training progress
if run_step == 5:
cvsm = CompositionalVectorAlgorithm("wordnet", CVSM_RET_DIR,
ENTITY_TYPE2VEC_FILENAME)
# Not using pretrained word embeddings decreases performance
# cvsm = CompositionalVectorAlgorithm("wordnet", CVSM_RET_DIR, None)
cvsm.train_and_test()
# Uncomment if need to train only one relation
# cvsm.train("/home/weiyu/Research/ChainsOfReasoningWithAbstractEntities/data/wn18rr/cvsm_entity/data/data_output/member_of_domain_region")
if run_step == 6:
cvsm = CompositionalVectorAlgorithm(
"wordnet",
CVSM_RET_DIR,
ENTITY_TYPE2VEC_FILENAME,
pooling_method="sat",
from main.playground.model2.CompositionalVectorAlgorithm import CompositionalVectorAlgorithm
if __name__ == "__main__":
# 1. uncomment this part to train and test all relations for FB15k-237
cvsm = CompositionalVectorAlgorithm(
experiment_dir="data/freebase15k237/cvsm_entity",
entity_type2vec_filename=None)
cvsm.train_and_test()
# 2. uncomment this part to train and test all relations for WN18RR
#cvsm = CompositionalVectorAlgorithm(experiment_dir="data/wordnet18rr/cvsm_entity",
# entity_type2vec_filename="data/wordnet18rr/entity_type2vec.pkl")
#cvsm.train_and_test()
vocab_dir=
"/home/weiyu/Research/ChainsOfReasoningWithAbstractEntities/data/fb15k237/cvsm_entity/data/vocab",
isOnlyRelation=False,
getOnlyRelation=False,
MAX_POSSIBLE_LENGTH_PATH=
8, # the max number of relations in a path + 1
NUM_ENTITY_TYPES_SLOTS=
7, # the number of types + 1 (the reason we +1 is to create a meaningless type for all entities)
pre_padding=True)
# 13. Run the model
# use $tensorboard --logdir runs to see the training progress
if run_step == 14:
cvsm = CompositionalVectorAlgorithm("freebase",
CVSM_RET_DIR,
None,
attention_method="sat",
early_stopping_metric="map")
# cvsm.train_and_test()
#cvsm = CompositionalVectorAlgorithm(CVSM_RET_DIR, None, attention_method="specific", early_stopping_metric="map")
#cvsm.train_and_test()
#cvsm = CompositionalVectorAlgorithm(CVSM_RET_DIR, None, attention_method="abstract", early_stopping_metric="map")
#cvsm.train_and_test()
# Uncomment if need to train only one relation
# cvsm.train("/home/weiyu/Research/ChainsOfReasoningWithAbstractEntities/data/fb15k237/cvsm_entity/data/data_output/|food|food|nutrients.|food|nutrition_fact|nutrient")
if run_step == 15:
cvsm = CompositionalVectorAlgorithm("freebase",