Conversion of ARFF to PAT files read by JavaNNS.
This converter is a rough converter from .arff to .pat format.
By Emil Kjer
Data Mining software Weka by The Unifersity of Waikato http://www.cs.waikato.ac.nz/ml/weka/
Neural Network simulator JavaNNS by Mathematich Naturwissenschaftliche Fakultat: http://www.ra.cs.uni-tuebingen.de/downloads/JavaNNS/
You'll need python on your system. To check, open a terminal and type python --version
If you don't have python already installed, download MiniConda and install.
Next you need to install the correct modules for python.
If you have conda, conda install click numpy scipy sklearn or otherwise, pip install click numpy scipy sklearn
Either git clone or download this repositiory and from the directory you can run the following:
chmod u+x arff2pat.py
./arff2pat.py or python arff2pat.py
Enter the names of input arff file, output pat file and float values for the test split (e.g. 0.33 is 33% of full set goes to test set) and validation split.
./arff2pat.py --arff=weather.numeric.arff --pat=weather.numeric.pat --testsize=0.33 --validationsize=0.1
If test size is set to 0.0 it will only convert the supplied file directly to an equivalent pat file.
If test size is > 0.0, train, validation (if validation size > 0.0) and test files will be generated.
If any rows contain missing data (? values) these will be discarded before the train/test/validation split (or before the output where no split required).
- Edited to take in command line arguments (instead of hard coded paths etc.)
- Uses numpy and sklearn train_test_split to split the data (stratified)