Planning

[geek-yang]

• Submission (Nov 10th)
• Baseline (Nov 17th)
  • Model (Top-down search, sensitivity test on each feature) @geek-yang
  • Feature (Bottom-up search, purely feature EDA) @kurtosis111
  • Deal with missing value
  • Deal with tags (e.g. groupby? Feature aggregation, like mean, std, etc.)
  • De-noise (e.g. low-pass filter)
  • Cross-validation (K-Fold)
• Model selection (Dec 8th)
  • Mono-model
  • Tree model (e.g. LightGBM, XGBoost, etc.) @kurtosis111
  • Neural network (RNN, transformer) @geek-yang
  • Multi-model
  • Ensemble
• Hyper-parameter tuning (Dec 31th)

[geek-yang]

For lag features, we will play with as many lags as possible.