def fit(model_fn):
data, labels = sk_utils.prepare_data() # train
model = stack_model_gen()
t1 = time.time()
model.fit(None, labels)
t2 = time.time()
print('\nClassifier stack training time : %0.3f seconds.' % (t2 - t1))
_, test_labels = sk_utils.prepare_data(mode='test')
print('Begin testing stacked classifier ')
t1 = time.time()
preds = model.predict(pred_directory='test/*/', X_indices=None)
t2 = time.time()
print('Classifier stack finished predicting in %0.3f seconds.' % (t2 - t1))
print('\nEvaluationg stack model')
sk_utils.evaluate(test_labels, preds)
print()
joblib.dump(model, '%s.pkl' % (model_fn))
def calculate_score(model_dir='lstm/', base_dir='test/', dataset='full'):
basepath = base_dir + model_dir
path = basepath + "*.npy"
data, labels, texts, word_index = prepare_data(MAX_NB_WORDS, MAX_SEQUENCE_LENGTH, mode='test', dataset=dataset)
files = glob.glob(path)
model_predictions = np.load(files[0])
print('Loaded predictions. Shape = ', model_predictions.shape)
model_predictions = model_predictions.mean(axis=0)
preds = np.argmax(model_predictions, axis=1)
evaluate(labels, preds)
def fit_voting_classifier(dataset='full'):
np.random.seed(1000)
# print('Loading data')
data, labels = prepare_data(mode='test', dataset=dataset)
if dataset == 'full':
pred_dir = 'test/*/'
elif dataset == 'obama':
pred_dir = 'obama/*/'
else:
pred_dir = 'romney/*/'
preds = model.predict_proba_dir(pred_dir)
evaluate(labels, np.argmax(preds, axis=1))
def write_predictions(model_fn='stack_model/stack-model', dataset='full'):
basepath = model_fn
path = basepath + ".pkl"
data, labels = sk_utils.prepare_data(mode='test', dataset=dataset)
model = joblib.load(path) # type: StackedGeneralizer
if dataset == 'full':
pred_dir = 'test/*/'
elif dataset == 'obama':
pred_dir = 'obama/*/'
else:
pred_dir = 'romney/*/'
preds_proba = model.predict_proba(pred_dir)
preds = np.argmax(preds_proba, axis=1)
sk_utils.evaluate(labels, preds)
np.save("stack_model/stack_predictions-%s.npy" % (dataset), preds_proba)