icde19-iim

Code release of "Learning Individual Models for Imputation"

Code release of "Learning Individual Models for Imputation" (ICDE 19). The description of code files are listed below:

IIM.java: Algorithm 1,2,3 in the paper. Use IIM algorithm to impute missing values.
Database.java: the class for Relation(database instance) indicating a relation.
LocalCluster.java: the class for Individual model for a certain tuple with a given $\ell$.

Datasets

We use following nine datasets in the paper (details are in Experiment Section):

ASF, CCS, CCPP and SN in UCI
CA, DA, MAM and HEP in KEEL
Siemens

The schema of the data file is shown in data/asf.csv

Attention

The example dataset is data/asf.csv which is clean and data/asf1_0.1miss.csv which exists 150 injected missing on Attribute A2.
Due to the resolution of computing, using incremental algorithm may have a little difference.

Parameters

The input and output of IIM algorithm is:

Method

setParams(lparams, K);
mainIIM(isInc, equalSigma, isRKNN);

Input:

int[] lparams = {1, 101, 5}; // LBEGIN, LMAX, INTERVAL(STEP), the range of adaptive learning on the number of learning neighbors(Algorithm 3)
int K = 10;  // the number of learning and imputing neighbors

isInc = True; // use incremental computing
equalSigma = True; // ignore the difference sigma when do combination (impute)
isRKNN = True; // use weighted distance when do combination (impute)

Output

Map<Position, Cell> repairedCells

Library

jama.jar is used to implement the code. You can get it by maven.

Citation

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

@inproceedings{DBLP:conf/icde/ZhangSSW19,
  author    = {Aoqian Zhang and
               Shaoxu Song and
               Yu Sun and
               Jianmin Wang},
  title     = {Learning Individual Models for Imputation},
  booktitle = {{ICDE}},
  pages     = {160--171},
  publisher = {{IEEE}},
  year      = {2019}
}