Progressive Learning Technique for Multi-class Classification

This repository contains the python implementation of progressive multi-class classifier of our paper "A Novel Progressive Learning Technique for Multi-class Classification". The paper is available at https://arxiv.org/pdf/1609.00085.pdf

Folder Description

1. src - Contains the python source codes
2. datasets - Contains the dataset csv file
3. results - Contains the results file generated during execution
4. logs - Contains the logs of each execution

Dependencies This code is developed in Python 3.6 (Ubuntu 16.04) and has following package dependencies:

• numpy 1.14.0
• pandas 0.21.1
• scikit-learn 0.19.1

To run the code: Move to src folder from command line and

$ python main.py

Arguements:

• Filename of the dataset

   -f FILENAME, --filename FILENAME    (default: ../datasets/iris_plt.csv)
  

• Label location in the dataset

   -l LABEL_LOCATION, --label LABEL_LOCATION     [can take "last" or "first" or None](default: last) 
  

• Scaling Type

   -s SCALE_TYPE, --scale SCALE_TYPE   [can take "minmax" or "std"](default: minmax)
  

• Testing Ratio

   -t TEST_RATIO, --testratio TEST_RATIO     [can take values in range 0 to 1](default: 0.1)
  

Hyperparameters of the model:

• Number of Hidden layer neurons

   -n HIDDEN_NEURONS, --neurons HIDDEN_NEURONS     (default: 10)
  

• Number of samples in initial block

   -i INIT_BLOCK_SIZE, --initial INIT_BLOCK_SIZE     (default=30)
  

• Batch size for training

   -b BATCH_SIZE, --batch BATCH_SIZE   (default=1)
  

Contact For queries, please email [email protected]

Citation

• Please consider the following paper for citing this work:

   @article{venkatesan2016novel,
   title={A novel progressive learning technique for multi-class classification},
   author={Venkatesan, Rajasekar and Er, Meng Joo},
   journal={Neurocomputing},
   volume={207},
   pages={310--321},
   year={2016},
   publisher={Elsevier}
   }