def test_train_save_load_hybrid_self_attention(datasets):
model_save_path = "self_att_hybrid_model.pth"
model = MatchingModel(attr_summarizer=attr_summarizers.Hybrid(
word_contextualizer="self-attention"))
model.run_train(
datasets.train,
datasets.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3,
)
s1 = model.run_eval(datasets.test)
model2 = MatchingModel(attr_summarizer=attr_summarizers.Hybrid(
word_contextualizer="self-attention"))
model2.load_state(model_save_path)
s2 = model2.run_eval(datasets.test)
assert s1 == s2
if os.path.exists(model_save_path):
os.remove(model_save_path)
def test_process_unlabeled_1(self):
vectors_cache_dir = '.cache'
if os.path.exists(vectors_cache_dir):
shutil.rmtree(vectors_cache_dir)
data_cache_path = os.path.join(test_dir_path, 'test_datasets',
'cacheddata.pth')
if os.path.exists(data_cache_path):
os.remove(data_cache_path)
vec_dir = os.path.abspath(os.path.join(test_dir_path, 'test_datasets'))
filename = 'fasttext_sample.vec.zip'
url_base = urljoin('file:', pathname2url(vec_dir)) + os.path.sep
ft = FastText(filename, url_base=url_base, cache=vectors_cache_dir)
train, valid, test = process(path=os.path.join(test_dir_path,
'test_datasets'),
train='test_train.csv',
validation='test_valid.csv',
test='test_test.csv',
id_attr='id',
ignore_columns=('left_id', 'right_id'),
embeddings=ft,
embeddings_cache_path='',
pca=True)
model_save_path = 'sif_model.pth'
model = MatchingModel(attr_summarizer='sif')
model.run_train(train,
valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3)
test_unlabeled = process_unlabeled(
path=os.path.join(test_dir_path, 'test_datasets', 'test_test.csv'),
trained_model=model,
ignore_columns=('left_id', 'right_id'))
self.assertEqual(test_unlabeled.all_text_fields, test.all_text_fields)
if os.path.exists(model_save_path):
os.remove(model_save_path)
if os.path.exists(data_cache_path):
os.remove(data_cache_path)
if os.path.exists(vectors_cache_dir):
shutil.rmtree(vectors_cache_dir)
def test_process_unlabeled_1():
vectors_cache_dir = ".cache"
if os.path.exists(vectors_cache_dir):
shutil.rmtree(vectors_cache_dir)
data_cache_path = os.path.join(test_dir_path, "test_datasets",
"cacheddata.pth")
if os.path.exists(data_cache_path):
os.remove(data_cache_path)
train, valid, test = process(
path=os.path.join(test_dir_path, "test_datasets"),
train="test_train.csv",
validation="test_valid.csv",
test="test_test.csv",
id_attr="id",
ignore_columns=("left_id", "right_id"),
embeddings=embeddings,
embeddings_cache_path="",
pca=True,
)
model_save_path = "sif_model.pth"
model = MatchingModel(attr_summarizer="sif")
model.run_train(
train,
valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3,
)
test_unlabeled = process_unlabeled(
path=os.path.join(test_dir_path, "test_datasets", "test_test.csv"),
trained_model=model,
ignore_columns=("left_id", "right_id"),
)
assert test_unlabeled.all_text_fields == test.all_text_fields
if os.path.exists(model_save_path):
os.remove(model_save_path)
if os.path.exists(data_cache_path):
os.remove(data_cache_path)
if os.path.exists(vectors_cache_dir):
shutil.rmtree(vectors_cache_dir)
def test_rnn(self):
model_save_path = 'rnn_model.pth'
model = MatchingModel(attr_summarizer='rnn')
model.run_train(self.train,
self.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3)
s1 = model.run_eval(self.test)
model2 = MatchingModel(attr_summarizer='rnn')
model2.load_state(model_save_path)
s2 = model2.run_eval(self.test)
self.assertEqual(s1, s2)
if os.path.exists(model_save_path):
os.remove(model_save_path)
def test_train_save_load_rnn(datasets):
model_save_path = "rnn_model.pth"
model = MatchingModel(attr_summarizer="rnn")
model.run_train(
datasets.train,
datasets.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3,
)
s1 = model.run_eval(datasets.test)
model2 = MatchingModel(attr_summarizer="rnn")
model2.load_state(model_save_path)
s2 = model2.run_eval(datasets.test)
assert s1 == s2
if os.path.exists(model_save_path):
os.remove(model_save_path)
def test_predict_unlabeled_hybrid(datasets):
model_save_path = "hybrid_model.pth"
model = MatchingModel(attr_summarizer="hybrid")
model.run_train(
datasets.train,
datasets.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3,
)
unlabeled = process_unlabeled(
path=os.path.join(test_dir_path, "test_datasets",
"test_unlabeled.csv"),
trained_model=model,
ignore_columns=("left_id", "right_id"),
)
pred_test = model.run_eval(datasets.test, return_predictions=True)
pred_unlabeled = model.run_prediction(unlabeled)
assert sorted(tup[1] for tup in pred_test) == sorted(
list(pred_unlabeled["match_score"]))
if os.path.exists(model_save_path):
os.remove(model_save_path)
def test_hybrid_self_attention(self):
model_save_path = 'self_att_hybrid_model.pth'
model = MatchingModel(attr_summarizer=attr_summarizers.Hybrid(
word_contextualizer='self-attention'))
model.run_train(self.train,
self.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3)
s1 = model.run_eval(self.test)
model2 = MatchingModel(attr_summarizer=attr_summarizers.Hybrid(
word_contextualizer='self-attention'))
model2.load_state(model_save_path)
s2 = model2.run_eval(self.test)
self.assertEqual(s1, s2)
if os.path.exists(model_save_path):
os.remove(model_save_path)
def test_hybrid(self):
model_save_path = 'hybrid_model.pth'
model = MatchingModel(attr_summarizer='hybrid')
model.run_train(self.train,
self.valid,
epochs=1,
batch_size=8,
best_save_path=model_save_path,
pos_neg_ratio=3)
unlabeled = process_unlabeled(path=os.path.join(
test_dir_path, 'test_datasets', 'test_unlabeled.csv'),
trained_model=model,
ignore_columns=('left_id', 'right_id'))
pred_test = model.run_eval(self.test, return_predictions=True)
pred_unlabeled = model.run_prediction(unlabeled)
self.assertEqual(sorted([tup[1] for tup in pred_test]),
sorted(list(pred_unlabeled['match_score'])))
if os.path.exists(model_save_path):
os.remove(model_save_path)