def test_dataset_featurize_no_vectorize(corpus, kb):
dataset = rel_ext.Dataset(corpus, kb)
kbts_by_rel, _ = dataset.build_dataset(sampling_rate=0.1)
def featurizer(kbt, corpus):
return utils.randvec(10)
dataset.featurize(kbts_by_rel, [featurizer], vectorize=False)
def test_find_new_relation_instances(corpus, kb, featurizer, vectorize):
dataset = rel_ext.Dataset(corpus, kb)
rel_ext.find_new_relation_instances(
dataset, [featurizer],
train_split='train',
test_split='dev',
model_factory=lambda: LogisticRegression(solver='liblinear'),
k=10,
vectorize=vectorize,
verbose=False)
def test_experiment(featurizer, vectorize, corpus, kb):
dataset = rel_ext.Dataset(corpus, kb)
splits = dataset.build_splits(
split_names=['tiny_train', 'tiny_dev', 'rest'],
split_fracs=[0.05, 0.05, 0.90],
seed=1)
results = rel_ext.experiment(splits,
train_split='tiny_train',
test_split='tiny_dev',
featurizers=[featurizer],
vectorize=vectorize,
verbose=False)
def test_dataset_featurize_vectorize(corpus, kb):
dataset = rel_ext.Dataset(corpus, kb)
kbts_by_rel, _ = dataset.build_dataset(sampling_rate=0.1)
featurizers = [lambda kbt, corpus, feature_counter: {"bias": 1}]
dataset.featurize(kbts_by_rel, featurizers)
def test_dataset_build_splits(corpus, kb):
dataset = rel_ext.Dataset(corpus, kb)
dat = dataset.build_splits(seed=1)
def test_dataset_build_dataset(corpus, kb):
dataset = rel_ext.Dataset(corpus, kb)
dat = dataset.build_dataset(include_positive=True, sampling_rate=0.1)
# In[3]:
rel_ext_data_home = os.path.join('data', 'rel_ext_data')
# In[4]:
corpus = rel_ext.Corpus(os.path.join(rel_ext_data_home, 'corpus.tsv.gz'))
# In[5]:
kb = rel_ext.KB(os.path.join(rel_ext_data_home, 'kb.tsv.gz'))
# In[6]:
dataset = rel_ext.Dataset(corpus, kb)
# You are not wedded to this set-up for splits. The bake-off will be conducted on a previously unseen test-set, so all of the data in `dataset` is fair game:
# In[7]:
splits = dataset.build_splits(split_names=['tiny', 'train', 'dev'],
split_fracs=[0.01, 0.79, 0.20],
seed=1)
# In[8]:
splits
# ## Baselines
