About
This is an alternative Python implementation of graph construction method RGCLI (Robust Graph that Considers Labeled Instances), published by Berton and Lopes (2016) [1]. The implementation is based on kd-tree and Multi-threading and is faster than the original described in the paper, specially for large data sets.
Download
- You can download the RGCLI software in http://www.alanvalejo.com.br/software?name=rgcli
Usage
RGCLI execution
python main.py -f input/square.dat -l input/square.label -1 20 -2 2 -d output
Input: any numerical dataset with any delimiter for attributes
Output: a weighted undirected graph in the format: filename + '-gbili.ncol'
Parameters
| Option | Domain | Required | Default | Description |
|---|---|---|---|---|
| -f, --filename | string [FILE] | yes | - | dataset as input file |
| -o, --output | string [FILE] | no | ncol | output file |
| -l, --label | string [FILE] | yes | - | list of labels points used to construct RGCLI |
| -1, --ke | [1,n] Integer interval | no | 20 | ke for KNN |
| -2, --ki | [1,n] Integer interval | no | 2 | ki for RGCLI |
| -t, --threads | [0,n] Integer interval | no | 4 | number of threads |
| -e, --format | ['ncol', 'pajek'] | no | ncol | format output file |
| -c, --skip_last_column | bool | no | true | skip the last column |
| -c, --skip_rows | [1,n-1] | no | None | skip rows |
Instal
Pip
$ pip install -r /path/to/requirements.txt
Anaconda env
$ conda env create -f environment.yml
$ conda activate rgcli
Anaconda create
$ conda create --name rgcli python=3.7.2
$ conda activate rgcli
$ conda install -c anaconda numpy
$ conda install -c anaconda scipy
Known Bugs
- Please contact the author for problems and bug report
Contact
- Alan Valejo.
- Ph.D. candidate at University of São Paulo (USP), Brazil.
- [email protected].
License and credits
- The GNU General Public License v3.0
- Giving credit to the author by citing the papers [1]
References
[1] Berton, Lilian and Faleiros, Thiago P. and Valejo, Alan and Valverde-Rebaza, Jorge Lopes, A. A., Rgcli: robust graph that considers labeled instances for semi-supervised learning, in Neurocomputing, p. 238-248, vol. 226, 2016, doi: https://doi.org/10.1016/j.neucom.2016.11.053
@article{berton2016rgcli,
author = {Berton, Lilian and Faleiros, Thiago P. and Valejo, Alan and Valverde-Rebaza, Jorge Lopes, A. A.},
title = {Rgcli: robust graph that considers labeled instances for semi-supervised learning},
journal = {Neurocomputing},
year = {2016},
pages = {238-248},
volume = {226},
doi = {https://doi.org/10.1016/j.neucom.2016.11.053}
}