andertadashi / vldb17-imr

Code release of "Time Series Data Cleaning: From Anomaly Detection to Anomaly Repairing" (VLDB 17)

vldb17-imr

Code release of "Time Series Data Cleaning: From Anomaly Detection to Anomaly Repairing" (VLDB 17). The description of code files are listed below:

• IMR.java: Algorithm 1 in the paper. Use IMR algorithm to repair time series with certain lables.
• IMRIC.java: Algorithm 2 in the paper. Use incremental compuation to speed up IMR algorithm.
• TimePoint.java: the class for TimePoint indicating a time point
• TimeSeries.java: the class for TimeSeries indicating a time sereis.

Datasets

We use following two datasets in the paper:

• GPS data with real errors, which will be made public later
• Intel Lab Data with synthetic errors.

The schema of the data file contains five columns,

• timestamp: the timestamp of the data
• dirty: the observation
• label: the value after label(if no labels, then it will be hold the observation value)
• truth: the truth
• isLabel(boolean): whether this point is labeled or not

Attention

• The example dataset is data/ild3k.data
• The first p data must be labeled

Parameters

The input and output of IMR/IMRIC algorithm is:

Method

mainIMR(p,delta,maxNumIterations,dirtySeries,labelSeries,labelList)
mainIMRIC(p,delta,maxNumIterations,dirtySeries,labelSeries,labelList)

Input:

int p = 3                     // order of the model
double delta = 0.1            // $\delta$
int maxNum = 100000           // maximum number of iterations
TimeSeries dirtySeries
TimeSeries labelSeries
ArrayList<Boolean> labelList

Output

Timeseries resultSeries

Library

jama.jar is used to implement the code

Citation

If you use this code for your research, please consider citing:

@article{DBLP:journals/pvldb/ZhangS0Y17,
  author    = {Aoqian Zhang and
               Shaoxu Song and
               Jianmin Wang and
               Philip S. Yu},
  title     = {Time Series Data Cleaning: From Anomaly Detection to Anomaly Repairing},
  journal   = {{PVLDB}},
  volume    = {10},
  number    = {10},
  pages     = {1046--1057},
  year      = {2017},
}