AutoGraph

Contents

ingestion/: The code and libraries used on Codalab to run your submmission.

scoring/: The code and libraries used on Codalab to score your submmission.

code_submission/: An example of code submission you can use as template.

data/: Some sample data to test your code before you submit it.

run_local_test.py: A python script to simulate the runtime in codalab

Local development and testing

1. To make your own submission to AutoGraph challenge, you need to modify the file model.py in code_submission/, which implements your algorithm.
2. Test the algorithm on your local computer using Docker, in the exact same environment as on the CodaLab challenge platform. Advanced users can also run local test without Docker, if they install all the required packages.
3. If you are new to docker, install docker from https://docs.docker.com/get-started/. Then, at the shell, run:

cd path/to/autograph_starting_kit/
docker run --gpus=0 -it --rm -v "$(pwd):/app/autograph" -w /app/autograph nehzux/kddcup2020:v2

The option -v "$(pwd):/app/autograph" mounts current directory (autograph_starting_kit/) as /app/autograph. If you want to mount other directories on your disk, please replace $(pwd) by your own directory.

The Docker image

nehzux/kddcup2020:v2

4. You will then be able to run the ingestion program (to produce predictions) and the scoring program (to evaluate your predictions) on toy sample data. In the AutoGraph challenge, both two programs will run in parallel to give feedback. So we provide a Python script to simulate this behavior. To test locally, run:

python run_local_test.py

If the program exits without any errors, you can find the final score from the terminal's stdout of your solution. Also you can view the score by opening the scoring_output/scores.txt.

The full usage is

python run_local_test.py --dataset_dir=./data/demo --code_dir=./code_submission

You can change the argument dataset_dir to other datasets (e.g. the two practice datasets we provide). On the other hand, you can also modify the directory containing your other sample code.

Solution

The GCN assumed the edge link two similar nodes, which is usually not incorrect. Different nodes create links like men and women, teachers and students, etc.
This solution implies a network to calculate classification problem in this type of graph by using single structure and shows great stability in the competition with 4th in the public leaderboard and 3rd in the private leaderboard.

Contributor

Zhenzhe Ying, whitebird827@163.com