def test_create_hashing_vectorizer_nb():
model = create_model('hashing_vectorizer-nb')
vec = model.steps[0][1]
clf = model.steps[1][1]
assert isinstance(model, Pipeline)
assert isinstance(vec, HashingVectorizer)
assert isinstance(clf, OneVsRestClassifier)
def test_create_binaryrelevance_tfidf_knn():
model = create_model('binaryrelevance-tfidf-knn')
vec = model.steps[0][1]
br = model.steps[1][1]
# clf = br.get_params()['classifier']
assert isinstance(model, Pipeline)
assert isinstance(vec, TfidfVectorizer)
assert isinstance(br, BinaryRelevance)
def test_create_labelpowerset_tfidf_svm():
model = create_model('labelpowerset-tfidf-svm')
vec = model.steps[0][1]
lp = model.steps[1][1]
clf = lp.get_params()['classifier']
assert isinstance(model, Pipeline)
assert isinstance(vec, TfidfVectorizer)
assert isinstance(lp, LabelPowerset)
assert isinstance(clf, SVC)
def test_create_hashing_vectorizer_svm():
model = create_model('hashing_vectorizer-svm')
vec = model.steps[0][1]
ovr = model.steps[1][1]
clf = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(vec, HashingVectorizer)
assert isinstance(ovr, OneVsRestClassifier)
assert isinstance(clf, SGDClassifier)
def test_create_sent2vec_sgd():
model = create_model('sent2vec-sgd')
vec = model.steps[0][1]
ovr = model.steps[1][1]
clf = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(vec, Sent2VecVectorizer)
assert isinstance(ovr, OneVsRestClassifier)
assert isinstance(clf, SGDClassifier)
def test_create_tfidf_gboost():
model = create_model('tfidf-gboost')
vec = model.steps[0][1]
ovr = model.steps[1][1]
clf = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(vec, TfidfVectorizer)
assert isinstance(ovr, OneVsRestClassifier)
assert isinstance(clf, GradientBoostingClassifier)
def test_create_classifierchain_tfidf_svm():
model = create_model('classifierchain-tfidf-svm')
vec = model.steps[0][1]
cc = model.steps[1][1]
clf = cc.get_params()['classifier']
assert isinstance(model, Pipeline)
assert isinstance(vec, TfidfVectorizer)
assert isinstance(cc, ClassifierChain)
assert isinstance(clf, SVC)
def model_path(tmp_path, label_binarizer_path):
model = create_model("mesh-cnn")
label_binarizer = load_pickle(label_binarizer_path)
Y_vec = label_binarizer.transform(Y)
model.fit(X, Y_vec)
model_path = os.path.join(tmp_path, "model")
model.save(model_path)
return model_path
def test_create_scibert_svm():
model = create_model('scibert-svm')
vec = model.steps[0][1]
ovr = model.steps[1][1]
clf = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(vec, BertVectorizer)
assert isinstance(ovr, OneVsRestClassifier)
assert isinstance(clf, SVC)
assert vec.pretrained == 'scibert'
def test_create_sent2vec_tfidf_sgd():
model = create_model('sent2vec-tfidf-sgd')
vec = model.steps[0][1]
ovr = model.steps[1][1]
sent2vec = vec.transformer_list[0][1].steps[0][1]
tfidf = vec.transformer_list[1][1].steps[0][1]
sgd = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(sent2vec, Sent2VecVectorizer)
assert isinstance(tfidf, TfidfVectorizer)
assert isinstance(sgd, SGDClassifier)
def test_create_tfidf_onehot_scheme_svm():
model = create_model('tfidf+onehot_team-svm')
vec = model.steps[0][1]
ovr = model.steps[1][1]
tfidf = vec.transformer_list[0][1].steps[1][1]
onehot = vec.transformer_list[1][1].steps[1][1]
sgd = ovr.get_params()['estimator']
assert isinstance(model, Pipeline)
assert isinstance(onehot, OneHotEncoder)
assert isinstance(tfidf, TfidfVectorizer)
assert isinstance(sgd, SVC)
def train(
train_data_path,
label_binarizer_path,
approach,
parameters=None,
model_path=None,
threshold=None,
sparse_labels=False,
cache_path=None,
data_format="list",
verbose=True,
):
"""
train_data_path: path. path to JSONL data that contains "text" and "tags" fields.
label_binarizer_path: path. path to load or store label_binarizer.
approach: str. approach to use for modelling e.g. tfidf-svm or bert.
parameters: str. a stringified dict that contains params that get passed to the model.
model_path: path. path to save the model.
threshold: float, default 0.5. Probability on top of which a tag is assigned.
sparse_labels: bool, default False. whether tags (labels) would be sparse for memory efficiency.
cache_path: path, default None. path to use for caching data transformations for speed.
data_format: str, default list. one of list, generator. generator used for memory efficiency.
"""
if os.path.exists(label_binarizer_path):
print(f"{label_binarizer_path} exists. Loading existing")
with open(label_binarizer_path, "rb") as f:
label_binarizer = pickle.loads(f.read())
else:
label_binarizer = create_label_binarizer(
train_data_path, label_binarizer_path, sparse_labels
)
model = create_model(approach, parameters)
# X can be (numpy arrays, lists) or generators
X_train, _, Y_train, _ = load_train_test_data(
train_data_path, label_binarizer, data_format=data_format
)
model.fit(X_train, Y_train)
if model_path:
if str(model_path).endswith("pkl") or str(model_path).endswith("pickle"):
with open(model_path, "wb") as f:
f.write(pickle.dumps(model))
else:
if not os.path.exists(model_path):
Path(model_path).mkdir(parents=True, exist_ok=True)
model.save(model_path)
def test_create_mesh_xlinear():
model = create_model('mesh-xlinear')
assert isinstance(model, MeshXLinear)
def test_create_scibert():
model = create_model('scibert')
assert isinstance(model, BertClassifier)
assert model.pretrained == 'scibert'
def test_create_tfidf_transformers_svm():
model = create_model('tfidf-transformers-svm')
assert isinstance(model, TfidfTransformersSVM)
def test_create_spacy_textclassifier():
model = create_model("spacy-textclassifier")
assert isinstance(model, SpacyClassifier)
def test_create_mesh_cnn():
model = create_model('mesh-cnn')
assert isinstance(model, MeshCNN)
def test_create_cnn():
model = create_model('cnn')
vec = model.steps[0][1]
clf = model.steps[1][1]
assert isinstance(vec, KerasVectorizer)
assert isinstance(clf, CNNClassifier)
def test_create_bilstm():
model = create_model('bilstm')
vec = model.steps[0][1]
clf = model.steps[1][1]
assert isinstance(vec, KerasVectorizer)
assert isinstance(clf, BiLSTMClassifier)
def test_create_mesh_tfidf_svm():
model = create_model('mesh-tfidf-svm')
assert isinstance(model, MeshTfidfSVM)
def test_create_bert():
model = create_model('bert')
assert isinstance(model, BertClassifier)
assert model.pretrained == 'bert-base-uncased'
