def optimize_model(dataset,
features=None,
padding=50,
model_cfg=None,
scorer=None,
n_samples=None,
max_evals=10):
if features is None:
features = []
if model_cfg is None:
model_cfg = {}
if scorer is None:
def scorer(**kwargs):
return 0.0
if n_samples is None:
n_samples = len(dataset[0][0])
n_samples = 6000 if n_samples >= 6000 else n_samples
ds_text, ds_labels = resample(dataset[0][0],
dataset[0][1],
n_samples=n_samples,
random_state=0)
text_train, text_test, y_train, y_test = train_test_split(ds_text,
ds_labels,
test_size=0.4)
embedding_extractor = make_embedding_pipeline(dataset, padding, features)
embedding_extractor = make_pipeline(*embedding_extractor)
X_train = embedding_extractor.fit_transform(text_train, y_train)
X_test = embedding_extractor.transform(text_test)
clf = CNNClassifier
opt = HyperoptOptimizer(clf, model_cfg, scorer,
max_evals).optimize(X_train, X_test, y_train,
y_test)
return opt
def build_model(dataset, features=None, model=None, **kwargs):
if features is None:
features = []
if model is None:
model = {}
embedding_extractor = make_embedding_pipeline(dataset, features)
clf = LSTMCNNClassifier(**model)
if 'early_stopping' in model:
clf.early_stopping(**model)
if 'validation_split' in model:
clf.validation_split(**model)
return make_pipeline(*embedding_extractor, clf).fit(dataset[0][0], dataset[0][1])