def test_load_yaml(self):
p = BaseEncoder.load_yaml(os.path.join(self.dirname, 'yaml', 'pipeline-multi-encoder.yml'))
self.assertRaises(RuntimeError, p.encode, 1)
p.train(1)
self.assertEqual(5, p.encode(1))
p = BaseEncoder.load_yaml(os.path.join(self.dirname, 'yaml', 'pipeline-multi-encoder.yml'))
self.assertRaises(RuntimeError, p.encode, 1)
def test_base(self):
a = BaseEncoder.load_yaml(self.yaml_path)
self.assertFalse(a.is_trained)
# simulate training
a.is_trained = True
a.dump()
os.path.exists(self.dump_path)
# load the dump from yaml
b = BaseEncoder.load_yaml(self.yaml_path)
self.assertTrue(b.is_trained)
def test_load_contrib(self):
os.environ['GNES_CONTRIB_MODULE'] = '%s:%s' % (cls_name, module_path)
from gnes.encoder.base import BaseEncoder, BaseTextEncoder
a = BaseEncoder.load_yaml(self.yaml_path)
self.assertIsInstance(a, BaseTextEncoder)
self.assertEqual(a.encode([]), 'hello 531')
a.dump()
a.dump_yaml(self.dump_yaml_path)
b = BaseEncoder.load_yaml(self.dump_yaml_path)
self.assertIsInstance(b, BaseTextEncoder)
self.assertEqual(b.encode([]), 'hello 531')
def test_base(self):
a = BaseEncoder.load_yaml(self.yaml_path)
self.assertFalse(a.is_trained)
for c in a.components:
c.is_trained = True
a.dump()
os.path.exists(self.dump_path)
# load the dump from yaml
b = BaseEncoder.load_yaml(self.yaml_path)
self.assertTrue(b.is_trained)
def test_name_warning(self):
d1 = DummyTFEncoder()
d2 = DummyTFEncoder()
d1.name = ''
d2.name = ''
d3 = PipelineEncoder()
d3.components = lambda: [d1, d2]
d3.name = 'dummy-pipeline'
d3.work_dir = './'
d3.dump()
d3.dump_yaml()
print('there should not be any warning after this line')
BaseEncoder.load_yaml(d3.yaml_full_path)
def test_mobilenet_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.mobilenet_yaml)
for test_img in self.test_img:
vec = self.encoder.encode(test_img)
print("the length of data now is:", len(test_img))
self.assertEqual(vec.shape[0], len(test_img))
self.assertEqual(vec.shape[1], 1280)
def test_inception_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.inception_yaml)
for test_img in self.test_img:
vec = self.encoder.encode(test_img)
print("the length of data now is:", len(test_img))
self.assertEqual(vec.shape[0], len(test_img))
self.assertEqual(vec.shape[1], 2048)
def test_inception_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.inception_yaml)
vec = self.encoder.encode(self.test_img)
self.assertEqual(len(vec), len(self.test_img))
self.assertEqual(len(vec[0].shape), 2)
self.assertEqual(vec[0].shape[0], self.test_img[0].shape[0])
self.assertEqual(vec[0].shape[1], 1152)
def test_dump_load(self):
self.encoder = BaseEncoder.load_yaml(self.yaml_path)
self.encoder.dump(self.dump_path)
encoder2 = BaseEncoder.load(self.dump_path)
vec = encoder2.encode(self.test_frames)
self.assertEqual(vec.shape[0], 3)
self.assertEqual(vec.shape[1], 19310)
def test_dump_load(self):
self.encoder = BaseEncoder.load_yaml(self.yaml_path)
self.encoder.dump(self.dump_path)
encoder2 = BaseEncoder.load(self.dump_path)
for test_img in self.test_img:
vec = encoder2.encode(test_img)
self.assertEqual(vec.shape[0], 2)
self.assertEqual(vec.shape[1], 2048)
def test_dump_load(self):
self.encoder = BaseEncoder.load_yaml(self.vgg_yaml)
self.encoder.dump(self.dump_path)
vgg_encoder2 = BaseEncoder.load(self.dump_path)
for test_img in self.test_img:
vec = vgg_encoder2.encode(test_img)
self.assertEqual(vec.shape[0], len(test_img))
self.assertEqual(vec.shape[1], 4096)
def test_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.yaml_path)
# train before encode to create pca_components
self.encoder.train(self.test_numeric)
vec = self.encoder.encode(self.test_numeric)
self.assertEqual(vec.shape, (1000, 300))
# dump after train with valied pca_components
self.encoder.dump(self.dump_path)
encoder2 = BaseEncoder.load(self.dump_path)
vec = encoder2.encode(self.test_numeric)
self.assertEqual(vec.shape, (1000, 300))
def test_pipeline_train(self):
p = PipelineEncoder()
p.components = lambda: [DummyEncoder(), DummyEncoder(), DummyEncoder()]
self.assertRaises(RuntimeError, p.encode, 1)
p.train(1)
self.assertEqual(4, p.encode(1))
p.name = 'dummy-pipeline'
p.dump()
p.dump_yaml()
a = BaseEncoder.load_yaml(p.yaml_full_path)
self.assertEqual(4, a.encode(1))
def setUp(self):
dirname = os.path.dirname(__file__)
self.dump_path = os.path.join(dirname, 'model.bin')
self.text_yaml = os.path.join(dirname, 'yaml', 'torch-transformers-encoder.yml')
self.tt_encoder = BaseEncoder.load_yaml(self.text_yaml)
self.test_str = []
with open(os.path.join(dirname, 'sonnets_small.txt')) as f:
for line in f:
line = line.strip()
if line:
self.test_str.append(line)
def test_vgg_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.yaml_path)
for test_img in self.test_img:
vec = self.encoder.encode(test_img)
self.assertEqual(vec.shape[0], 2)
self.assertEqual(vec.shape[1], 2048)
def test_load_contrib(self):
from gnes.encoder.base import BaseEncoder, BaseTextEncoder
a = BaseEncoder.load_yaml(self.yaml_path)
self.assertIsInstance(a, BaseTextEncoder)
self.assertEqual(a.encode([]), 'hello 531')
def test_yt8m_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.yt8m_yaml)
vec = self.encoder.encode(self.test_video)
self.assertEqual(vec.shape[0], len(self.test_video))
self.assertEqual(vec.shape[1], 19310)
def test_yt8m_encoding(self):
self.encoder = BaseEncoder.load_yaml(self.yaml_path)
vec = self.encoder.encode(self.test_frames)
self.assertEqual(vec.shape[0], 3)
self.assertEqual(vec.shape[1], 19310)