README.md

# IPD — Interaction Preserving Discretisation

## How to Compile

You can compile the code yourself from scratch as follows
>cd /code
>javac *.java multibinning/business/*.java multibinning/data/*.java

You can then create the .jar file as follows
>jar cmf mainclass.txt ipd.jar *.class multibinning/business/*.class multibinning/data/*.java



## How to Run

You run IPD as follows
>java -jar ipd.jar <parameters>

where for the IPD methors you'll need to specify the following parameters:

| Parameter              | Meaning  
| ---------------------- | ----
| -FILE_INPUT            | name of input file  
| -FILE_CP_OUTPUT        | name of output file for cut points per dimension 
| -FILE_RUNTIME_OUTPUT   | name of output file for runtime in microseconds
| -FILE_DATA_OUTPUT      | name of output file for discretized data in .arff format
| -NUM_ROWS              | number of data points  
| -NUM_MEASURE_COLS      | number of numeric dimensions  
| -NUM_CAT_CONTEXT_COLS  | number of categorical dimensions  
| -MAX_VAL               | maximum value, used for normalization  
| -METHOD                | method used (0 for IPD_opt, 2 for IPD_gr)  

For example,

>java -jar ipd.jar -FILE_INPUT example/simple.csv  -FILE_CP_OUTPUT cuts.txt  
  -FILE_RUNTIME_OUTPUT runtime.txt  -FILE_DATA_OUTPUT out.txt  -NUM_ROWS 209  
  -NUM_MEASURE_COLS 2 -NUM_CAT_CONTEXT_COLS 0 -MAX_VAL 1.0  -METHOD 0

  the output files of which have been included in /example.




## Related Publications

Nguyen, H-V, Müller, E, Vreeken, J & Böhm, K Unsupervised Interaction-Preserving Discretization of Multivariate Data.  Data Mining and Knowledge Discovery vol.28(5), pp 1366-1397, Springer, 2014. 

All synthetic and benchmark data sets used in Nguyen et al. (2014) can be found in /data.
The benchmark datasets were drawn from the [UCI Machine Learning Repository](http://archive.ics.uci.edu/ml/).
The PAMAP data sets can be found at the [PAMAP website](http://www.pamap.org/).


## Unadvertised and Disabled Functionality

The IPD code contains a wide spectrum of discretisation methods that are currently not 
documented. Two such methods, Supervised Univariate Discretisation, and PCA-based 
Discretisation, require the Weka framework. 

While these are *enabled* in the provided jar file, to ease compilation and 
general use of the code, these have hence been *disabled* in the code. 


## Credits

Most of the IPD code was written by Hoang Vu Nguyen. 
Kailash Budhathoki fixed bugs and helped making the code easier to use.