Curve

Curve is an open-source tool to help label anomalies on time-series data. The labeled data (also known as the ground truth) is necessary for evaluating time-series anomaly detection methods. Otherwise, one can not easily choose a detection method, or say method A is better than method B. The labeled data can also be used as the training set if one wants to develop supervised learning methods for detection.

Curve is designed to support plugin, so one can equip Curve with customized and powerful functions to help label effectively. For example, a plugin to identify anomalies which are similar to the one you labeled, so you don't have to search them through all the data.

Curve is originally developed by Baidu and Tsinghua NetMan Lab.

Getting Started

Install dependencies and build*

Make sure bash, python2.7+ with pip and nodejs with npm are already installed. Then run the build script. It will install all the dependenceis and build Curve.

./build.sh

Run and stop

Simply use control.sh to start or stop Curve.

./control.sh start
./control.sh stop

Server will blind 8080 by default, you can change it in ./api/uwsgi.ini.

If you pull updates from github, make sure to rebuild first

Data format

You can load a CSV file into Curve. The CSV should have the following format

• First column is the timestamp
• Second column is the value
• Third column (optional) is the label. 0 for normal and 1 for abnormal.

The header of CSV is optinal, like timestamp,value,label.

Some examples of valid CSV

• With a header and the label column

timestamp	value	label
1476460800	2566.35	0
1476460860	2704.65	0
1476460920	2700.05	0

• Without the header

1476460800	2566.35	0
1476460860	2704.65	0
1476460920	2700.05	0

• Without the header and the label colum

1476460800	2566.35
1476460860	2704.65
1476460920	2700.05

• Timestamp in human-readable format

20161015000000	2566.35
20161015000100	2704.65
20161015000200	2700.05

Test

cd api && pytest

Plugin dir

api/curve/v1/plugins