def test_lcnn_reload():
# needs change
current_dir = tempfile.mkdtemp()
download_url(url=URL, dest_dir=current_dir)
untargz_file(path.join(current_dir, 'lcnn_data_feature.tar.gz'), current_dir)
tasks, datasets, transformers = load_dataset_from_disk(
path.join(current_dir, 'lcnn_data'))
train, valid, test = datasets
model_dir = tempfile.mkdtemp()
model = LCNNModel(
mode='regression', batch_size=8, learning_rate=0.001, model_dir=model_dir)
model.fit(train, nb_epoch=10)
# check predict shape
valid_preds = model.predict_on_batch(valid.X)
assert valid_preds.shape == (65, 1)
test_preds = model.predict(test)
assert test_preds.shape == (65, 1)
# check overfit
regression_metric = Metric(mae_score)
scores = model.evaluate(test, [regression_metric], transformers)
assert scores[regression_metric.name] < 0.6
# reload
reloaded_model = LCNNModel(
mode='regression', batch_size=8, learning_rate=0.001, model_dir=model_dir)
reloaded_model.restore()
original_pred = model.predict(test)
reload_pred = reloaded_model.predict(test)
assert np.all(np.abs(original_pred - reload_pred) < 0.0000001)
def __init__(self,
pretrain_model_path: Optional[str] = None,
radius: int = 1,
unseen: str = 'UNK',
gather_method: str = 'sum'):
"""
Paremeters
----------
pretrain_file: str, optional
The path for pretrained model. If this value is None, we use the model which is put on
github repository (https://github.com/samoturk/mol2vec/tree/master/examples/models).
The model is trained on 20 million compounds downloaded from ZINC.
radius: int, optional (default 1)
The fingerprint radius. The default value was used to train the model which is put on
github repository.
unseen: str, optional (default 'UNK')
The string to used to replace uncommon words/identifiers while training.
gather_method: str, optional (default 'sum')
How to aggregate vectors of identifiers are extracted from Mol2vec.
'sum' or 'mean' is supported.
"""
try:
from gensim.models import word2vec
from mol2vec.features import mol2alt_sentence, sentences2vec
except ModuleNotFoundError:
raise ValueError("This class requires mol2vec to be installed.")
self.radius = radius
self.unseen = unseen
self.gather_method = gather_method
self.sentences2vec = sentences2vec
self.mol2alt_sentence = mol2alt_sentence
if pretrain_model_path is None:
data_dir = get_data_dir()
pretrain_model_path = path.join(data_dir,
'mol2vec_model_300dim.pkl')
if not path.exists(pretrain_model_path):
targz_file = path.join(data_dir, 'mol2vec_model_300dim.tar.gz')
if not path.exists(targz_file):
download_url(DEFAULT_PRETRAINED_MODEL_URL, data_dir)
untargz_file(
path.join(data_dir, 'mol2vec_model_300dim.tar.gz'),
data_dir)
# load pretrained models
self.model = word2vec.Word2Vec.load(pretrain_model_path)