def stability(filtered_data, features_list, iteration):
"""
This function repeatedly sample the data and calculate the average feature size and stability score for each anomalies
:param filtered_data: list of data
:param features_list: list of features
:param iteration: number of iteration
:return:
stability matrix and feature list
"""
feature_list_result = []
for i in range(iteration):
new_data = random_data(filtered_data)
index_data = calculate_class_entropy(new_data, "stability")
new_data = select_segment(data, index_data)
data_segment_entropy = calculate_segment_entropy(new_data)
distance = calculate_D(data_segment_entropy, index_data['h_class'])
for j in range(len(distance)):
correlated_feature_index = remove_monotonic_feature(filtered_data[j], features_list)
Exstream_feature, Exstream_data = drop_features(distance[j, :], filtered_data[j], features_list,
correlated_feature_index)
if len(Exstream_feature) == 1:
feature_list_result.append(Exstream_data.columns[:-1].values)
else:
Exstream_cluster = remove_correlated_features(Exstream_data, Exstream_feature, features_list,
distance[j, :])
feature_list_result.append(Exstream_cluster.columns[:-1].values)
stability_matrix = np.zeros((2, len(distance)))
list = np.array(feature_list_result)
for i in range(len(distance)):
index = np.array(range(i, len(list), len(distance)))
temp = list[index]
avg_size, stability = stats(temp)
stability_matrix[:, i] = avg_size, stability
return stability_matrix, feature_list_result
path_truth = 'data/truth'
path_segment = 'data/aggregated'
file_clean_list = [
'batch146_17_clean.csv', 'batch146_19_clean.csv',
'batch146_20_clean.csv'
]
for file_clean in file_clean_list:
file_truth = file_clean.replace('clean', 'truth')
file_segment = file_clean.replace('clean', 'aggregated')
data = pd.read_csv(os.path.join(path_clean, file_clean))
# read index data
index_data = pd.read_csv(os.path.join(path_truth, file_truth))
index_data_mapped = mapping(index_data)
index_data_class_entropy = calculate_class_entropy(index_data_mapped)
filtered_data = select_segment(data, index_data_class_entropy)
aggregated_data = combine_data(filtered_data)
index_data = calculate_class_entropy(aggregated_data, "aggregate")
data_segment_entropy = pd.read_csv(
os.path.join(path_segment, file_segment))
#data_segment_entropy = calculate_segment_entropy(aggregated_data, "aggregate")
distance = calculate_D(data_segment_entropy, index_data['h_class'])
features_list = data_segment_entropy.columns
correlated_feature_index = remove_monotonic_feature(
aggregated_data, features_list)
Exstream_feature, Exstream_data = drop_features(
distance[0], aggregated_data, features_list,
correlated_feature_index)
Exstream_cluster = remove_correlated_features(Exstream_data,
Exstream_feature,
from sklearn.metrics import precision_score, recall_score, confusion_matrix, \
classification_report, accuracy_score, f1_score
pd.options.mode.chained_assignment = None
# shiqiGao [email protected]
# Thu Dao [email protected]
if __name__ == '__main__':
## read cleaned data
data = pd.read_csv('./data/clean/batch146_20_clean.csv')
# read index data
index_data = pd.read_csv('./data/truth/batch146_20_truth.csv')
## map index data and calculate class entropy
index_data_mapped = mapping(index_data)
index_data_class_entropy = calculate_class_entropy(index_data_mapped)
## calculate segment entropy
filtered_data = select_segment(data, index_data_class_entropy)
data_segment_entropy = pd.read_csv(
'./data/segment/batch146_20_segment.csv')
## 6x1 class entropy:
h_class = index_data_class_entropy['h_class']
## 6x19 class entropy:
h_segment = data_segment_entropy
# numpy array len(anomalies) x len(features)
distance = calculate_D(h_segment, h_class)
# adding reward for different features
# numpy array len(feature) x 1