Skip to content

Latest commit

 

History

History
57 lines (45 loc) · 1.3 KB

README.md

File metadata and controls

57 lines (45 loc) · 1.3 KB

1. Introduction

M competition is a well-known competition focused on time series forecasting, held once a year. This year's M6 is a financial forecasting competition that primarily focuses on predicting stock prices (returns) and risks, providing new perspectives on the Efficient Market Hypothesis (EMH).
Our Team: OPPO XZ Lab
Final Global Rank: 4th

2. Implementation

cd m6_dea_dnn

Data Pulling

python ./code/feature/data_crawler.py --meta_path './eod_data/M6_Universe.csv' --api_key '[EOD API key]' --save_path './data/' --year_duration 5

(2) Feature Engieering

mkdir pp_data2
python ./code/feature/feature_engineer.py --data_path './data/' --self_data_path './pp_data2/'

(3) Forecast Modeling we support three models for forecasting.

# LightGBM
python ./code/lgb_classifier.py

# DNN
python ./code/dnn_classifier.py

# DNN with DAE
python ./code/autoencoder_dnn_classifier.py

(4) Decision Making

  • Data Preparation: data_preparation.ipynb
  • Traditional models: traditional_models.ipynb
  • GA and PSO: nonlinear_models.ipynb
  • DE: DE_experiments.ipynb

Experiment result see:

# Traditional models
traditional_models.ipynb

# GA
ga_weights.zip

# PSO
pso_weights.zip

# DE
de_weights_100_100.zip
de_weights_maxiter.zip
de_weights_sizepop.zip