https://github.com/jpaillard/kernelchallenge

Data Challenge - Kernel methods
https://github.com/jpaillard/kernelchallenge

graphs kernel-logistic-regression kernel-methods kernel-support-vector-machine molecule weisfeiler-lehman

Last synced: 4 days ago
JSON representation

Data Challenge - Kernel methods

• Host: GitHub
• URL: https://github.com/jpaillard/kernelchallenge
• Owner: jpaillard
• Created: 2023-03-16T23:37:13.000Z (over 1 year ago)
• Default Branch: master
• Last Pushed: 2023-04-15T12:38:39.000Z (over 1 year ago)
• Last Synced: 2024-10-17T02:29:29.751Z (20 days ago)
• Topics: graphs, kernel-logistic-regression, kernel-methods, kernel-support-vector-machine, molecule, weisfeiler-lehman
• Language: Python
• Homepage:
• Size: 804 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# Data Challenge - Kernel methods

[![codecov](https://codecov.io/github/jpaillard/KernelChallenge/branch/master/graph/badge.svg?token=TJZSQ80QCV)](https://codecov.io/github/jpaillard/KernelChallenge)

[https://www.kaggle.com/competitions/data-challenge-kernel-methods-2022-2023](https://www.kaggle.com/competitions/data-challenge-kernel-methods-2022-2023)

Please refer to the `report.pdf` file to learn more about the project and the methods.

## Install

```

git clone 

```

Requirements and conda env

```

conda env create -f environment.yml

```

Install the package itself in developer mode

```

pip install -e .

```

## Usage

```

usage: main.py [-h] [--data_path DATA_PATH] [--n N] [--h_iter H_ITER] [--c C] [--method METHOD] [--edges]

               [--submit]

options:

  -h, --help            show this help message and exit

  --data_path DATA_PATH

                        Path to folder that contains the dataset (.pkl files)

  --n N                 Number of samples from the dataset to use for training

  --h_iter H_ITER       Number of iterations (depth) for the WL algorithm

  --c C                 Regularization parameter for the classifier

  --method METHOD       Classifier to use (SVC or KLR)

  --edges               Use edge embedding in the WL algorithm (see report for more details)

  --submit              create submission file for the challenge

```

### The best results have been obtained using the following command line:

` 

python main.py --method SVC --n 6000 --c 0.01 --submit --h_iter 1 --edges

`

## Description

### Architecture

```

.

├── KernelChallenge       # Main package

│   ├── kernels.py        # WL kernel implementation

│   ├── SVC.py            # SVC implementation

|   ├── KLR.py            # KLR implementation

|   ├── preprocessing.py  # preprocessing scripts before WL kernel

│   └── ...

├── tests                 # PyTest scripts

│   └── ...

├── environment.yml       # Conda environment

├── format_output.py      # script to format the prediction for kaggle competition   

├── main.py               # executable script

├── report.pdf            # Report presenting the project

└── ...

```

Implementation of the Weisfeiler-lehman kernel for graph classification. And two kernel methods classification algorithms: SVR and KLR. 

```BibTex

@article{shervashidze2011weisfeiler,

  title={Weisfeiler-lehman graph kernels.},

  author={Shervashidze, Nino and Schweitzer, Pascal and Van Leeuwen, Erik Jan and Mehlhorn, Kurt and Borgwardt, Karsten M},

  journal={Journal of Machine Learning Research},

  year={2011}

}

```