def test_synthetic_sample_and_remove(self):
synthetic = SyntheticDatabase(10, 10, number_features=3) # 100 records, 10 entities, 10 records each
synthetic2 = synthetic.sample_and_remove(50)
self.assertEqual(len(synthetic.database.records), 50)
self.assertEqual(len(synthetic2.database.records), 50)
self.assertEqual(set(synthetic.database.records.keys()), set(synthetic.labels.keys()))
self.assertEqual(set(synthetic2.database.records.keys()), set(synthetic2.labels.keys()))
self.assertEqual(len(set(synthetic.database.records.keys()) & set(synthetic2.database.records.keys())), 0)
self.assertEqual(len(set(synthetic.labels.keys()) & set(synthetic2.labels.keys())), 0)
def experiment_wrapper(dataset_name):
"""
Experiment wrapper, just takes the type of experiment, all parameters saved here
:param dataset_name: Name of the database to run on, either synthetic, restaurant, abt-buy, trafficking
"""
if dataset_name == 'synthetic':
number_entities = 100
records_per_entity = 10
train_database_size = 200
train_class_balance = 0.5
validation_database_size = 200
corruption = 0.001 #0.025
number_thresholds = 30
number_features = 10
synthetic_database = SyntheticDatabase(number_entities, records_per_entity, number_features=number_features)
corruption_array = corruption*np.random.normal(loc=0.0, scale=1.0, size=[validation_database_size,
synthetic_database.database.feature_descriptor.number])
synthetic_database.corrupt(corruption_array)
synthetic_train = synthetic_database.sample_and_remove(train_database_size)
synthetic_validation = synthetic_database.sample_and_remove(validation_database_size)
synthetic_test = synthetic_database
thresholds = np.linspace(0, 1, number_thresholds)
experiment = Experiment(synthetic_train.database, synthetic_validation.database, synthetic_test.database,
synthetic_train.labels, synthetic_validation.labels, synthetic_test.labels,
train_class_balance, thresholds)
experiment.plot()
else:
number_thresholds = 5
if dataset_name == 'restaurant': # 864 records, 112 matches
features_path = '../data/restaurant/merged.csv'
labels_path = '../data/restaurant/labels.csv'
train_database_size = 300
train_class_balance = .4
validation_database_size = 200
database = Database(annotation_path=features_path)
elif dataset_name == 'abt-buy': # ~4900 records, 1300 matches
features_path = '../data/Abt-Buy/merged.csv'
labels_path = '../data/Abt-Buy/labels.csv'
train_database_size = 300
train_class_balance = 0.4
validation_database_size = 300
database = Database(annotation_path=features_path)
elif dataset_name == 'trafficking':
features_path = '../data/trafficking/features.csv'
labels_path = '../data/trafficking/labels.csv'
train_database_size = 300
train_class_balance = 0.5
validation_database_size = 300
#database = Database(annotation_path=features_path)
else:
raise Exception('Invalid dataset name')
thresholds = np.linspace(0, 1, number_thresholds)
# labels = np.loadtxt(open(labels_path, 'rb'))
# database_train = database.sample_and_remove(train_database_size)
# database_validation = database.sample_and_remove(validation_database_size)
# database_test = database
# labels_train = dict()
# labels_validation = dict()
# labels_test = dict()
# for identifier, label in enumerate(labels):
# if identifier in database_train.records:
# labels_train[identifier] = label
# elif identifier in database_validation.records:
# labels_validation[identifier] = label
# elif identifier in database_test.records:
# labels_test[identifier] = label
# else:
# raise Exception('Record identifier ' + str(identifier) + ' not in either database')
###
database_train = Database('../data/trafficking/cluster_subsample0_10000.csv', header_path='../data/trafficking/cluster_subsample_header_LM.csv', max_records=5000)
database_validation = Database('../data/trafficking/cluster_subsample1_10000.csv', header_path='../data/trafficking/cluster_subsample_header_LM.csv', max_records=5000)
database_test = Database('../data/trafficking/cluster_subsample2_10000.csv', header_path='../data/trafficking/cluster_subsample_header_LM.csv', max_records=1000)
labels_train = fast_strong_cluster(database_train)
labels_validation = fast_strong_cluster(database_validation)
labels_test = fast_strong_cluster(database_test)
###
experiment = Experiment(database_train, database_validation, database_test,
labels_train, labels_validation, labels_test,
train_class_balance, thresholds)
#print 'Saving results'
#pickle.dump(experiment, open('experiment.p', 'wb'))
experiment.plot()
print 'Finished'
