irhdab/TransLSTM-Predictor
A quantitative trading model that combines CNN, Bi-LSTM, and Transformer architectures to achieve state-of-the-art (SOTA) accuracy in stock movement forecasting.
What's novel
A quantitative trading model that combines CNN, Bi-LSTM, and Transformer architectures to achieve state-of-the-art (SOTA) accuracy in stock movement forecasting.
Code Analysis
10 files read · 3 roundsA CNN-LSTM-Transformer hybrid model that predicts 30-day percentage returns from OHLCV data with technical indicators, using a 3-model ensemble and a naive buy/sell backtest strategy.
Strengths
The model architecture is a genuine multi-branch hybrid (CNN, Transformer, LSTM) with proper positional encoding, and the dual-scaler approach for features vs. targets shows awareness of data leakage. Feature engineering covers a reasonable set of technical indicators.
Weaknesses
The walk-forward validation is non-standard and convoluted, the ensemble evaluation uses a monkey-patch hack, two modules (predictor.py, visualizer.py) are dead code, and the single test just calls main() without any meaningful assertions. The backtester ignores transaction costs and uses a questionable win-rate metric.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
11
Contributors
1
Files
26
Active weeks
3
Repository
Language
Python
Stars
3
Forks
0
License
MIT