nuglifeleoji/Factor-Research

Implements a complete quantitative research pipeline with vectorized alpha factor generation (100+ technical and statistical factors using time-series operators), multi-stage feature selection combining IC filtering, correlation analysis, and ML-based importance ranking to identify 85 independent factors. The training framework uses proper time-series cross-validation with Optuna hyperparameter optimization across six model architectures (Linear, DNN, DeepNet, AlexNet, LSTM+ResNet, Transformer), evaluating performance via Information Coefficient to avoid look-ahead bias in financial predictions.

388 stars. No commits in the last 6 months.