Accurate Multivariate Stock Movement Prediction via Data-Axis Transformer with Multi-Level Contexts Implementation

This is a Keras implementation of the stock movement prediction model in "Accurate Multivariate Stock Movement Prediction via Data-Axis Transformer with Multi-Level Contexts" (Jaemin Yoo, Yejun Soun, Yong-chan Park, and U Kang, KDD, 2021).

The model is consisted of 3 stages

• time axis attention
• context aggregation
• data axis attention

The model presented compares the performance with two other models

• https://www.ijcai.org/proceedings/2019/0810.pdf
• Stocknet: https://github.com/yumoxu/stocknet-code

how to run

see hyperparameters

installation

brew install graphviz pip install -r requirements.txt

add kernel

''' ipython kernel install --user --name=ml '''

start NNI UI

nnictl create --config config.yaml

stop NNI

nnictl stop --all

NNI 관련 파일

1. config.yaml (nni 설정파일)
2. pred_nni.py (모델 생성및 nni hyper parameter 지정)

NNI run

• threshold 사용하여 (하락, threshold 구간 = 0) , (상승 = 1) 결과 = VmMI9