Kakao Arena - Melon Playlist Continuation (Solution by brkln)

This repository is for Melon Playlist Continuation, the 3rd machine learning challenge of Kakao Arena.

Environment

The codes are written based on Python 3.7.6. These are the requirements for running the codes:

• fire
• tqdm
• numpy
• pandas
• scipy
• implicit
• khaiii

How to run

1. Before running the codes, 2 json files should be prepared: train.json, test.json. The files can be downloaded from https://arena.kakao.com/
2. After downloading the json files, both files should be placed in a folder named res. The directory should look like this:

$> tree
.
├── README.md
├── arena_util.py
├── inference.py
├── requirements.txt
├── res
│   ├── test.json
│   └── train.json
└── train.py

3. train.py is run by the code below. After running the code, 6 files will be saved in the directory: songtag_length.pkl, popular_song_dict.pkl, popular_tag_dict.pkl, trainval_id_dict.pkl, songtag_matrix.npz, model.sav.

$> python train.py run \    
        --train_fname=res/train.json \
        --question_fname=res/test.json

4. inference.py is run by the code below. After running the code, the result file will be saved in the following directory: arena_data/results/results.json.

$> python inference.py run \
        --question_fname=res/test.json

Source

• arena_util.py is provided by Kakao Corp.
• Function get_token in train.py is provided in https://arena.kakao.com/forum/topics/226