def do(task, src_cfg, **kwargs):
ip_type = kwargs.get('ip_type', 'dense5.track15mbh')
suffix = kwargs.get('suffix', '')
nr_clusters = kwargs.get('nr_clusters')
dataset = Dataset(
src_cfg, ip_type=ip_type, suffix=suffix, nr_clusters=nr_clusters)
per_slice = kwargs.get('per_slice', False)
# Select the task based on opt_type argument.
if task == 'compute':
pass
elif task == 'check':
from fisher_vectors import data
data.check_given_dataset(dataset)
elif task == 'merge':
if per_slice:
from fisher_vectors.per_slice import features
class_idx = kwargs.get('class_idx')
features.merge_class_given_dataset(dataset, class_idx)
else:
from fisher_vectors import data
data.merge_given_dataset(dataset)
elif task == 'remove':
pass
elif task == 'evaluate':
from fisher_vectors.scripts.evaluate import evaluate_given_dataset
evaluate_given_dataset(dataset, **kwargs)
#with open(os.path.join(
# ROOT_PATH, dataset.DATASET, 'results', RESULTS_FN), 'a') as ff:
# ff.write('%s %2.3f\n' % (str(model), 100 * model.score()))
else:
raise Exception("Task is not defined.")
usage()
sys.exit(1)
def test_merge_class_given_dataset(self):
out_folder = "/tmp/"
out_fn = os.path.join(out_folder, "train_class_0.dat")
labels_fn = os.path.join(out_folder, "labels_train_class_0.info")
nr_pos = 30
nr_neg = 60
merge_class_given_dataset(
self.dataset, 0, out_folder=out_folder, nr_positive_samples=nr_pos, nr_negative_samples=nr_neg
)
# Load sufficient statistics.
sstats = np.fromfile(out_fn, dtype=np.float32).reshape((-1, self.sstats_len))
labels = pickle.load(open(labels_fn, "rb"))
# Check if the number is accurate.
assert sstats.shape[0] == len(labels), "Size mismatch."
def test_merge_class_given_dataset(self):
out_folder = '/tmp/'
out_fn = os.path.join(out_folder, 'train_class_0.dat')
labels_fn = os.path.join(out_folder, 'labels_train_class_0.info')
nr_pos = 30
nr_neg = 60
merge_class_given_dataset(self.dataset,
0,
out_folder=out_folder,
nr_positive_samples=nr_pos,
nr_negative_samples=nr_neg)
# Load sufficient statistics.
sstats = np.fromfile(out_fn, dtype=np.float32).reshape(
(-1, self.sstats_len))
labels = pickle.load(open(labels_fn, 'rb'))
# Check if the number is accurate.
assert sstats.shape[0] == len(labels), "Size mismatch."
