📈 QuantLab AI

Deep Learning for Financial Time Series Forecasting

Predicting stock prices using LSTM, GRU, and Transformer architectures built with PyTorch.

---

🚀 Overview

QuantLab AI is a production-style deep learning framework for financial time-series forecasting.

The project implements and benchmarks multiple neural network architectures for stock price prediction using historical market data from Yahoo Finance.

Unlike a basic stock predictor, QuantLab is designed as a reusable ML pipeline with:

• Automated data collection
• Data preprocessing
• Sliding window sequence generation
• Multiple deep learning models
• Training pipeline
• Evaluation metrics
• Benchmarking
• Future price prediction

---

✨ Features

• Historical stock data download using Yahoo Finance
• Automatic preprocessing and scaling
• Sliding window dataset creation
• LSTM implementation
• GRU implementation
• Transformer implementation
• Early stopping
• Model checkpointing
• Prediction visualization
• Benchmark multiple models
• Future stock price prediction
• Modular production-ready codebase

---

🏗 Project Architecture

Yahoo Finance
      │
      ▼
Data Download
      │
      ▼
Preprocessing
      │
      ▼
Sequence Generation
      │
      ▼
PyTorch Dataset
      │
      ▼
──────────────────────────────
│ LSTM │ GRU │ Transformer │
──────────────────────────────
      │
      ▼
Training
      │
      ▼
Evaluation
      │
      ▼
Prediction
      │
      ▼
Benchmark Results

---

📂 Project Structure

QuantLab/

├── data/
├── experiments/
├── models/
├── notebooks/
├── reports/

├── src/
│   └── ml/
│       ├── benchmark.py
│       ├── config.py
│       ├── dataset.py
│       ├── evaluate.py
│       ├── losses.py
│       ├── model.py
│       ├── predict.py
│       ├── train.py
│       └── utils.py

├── requirements.txt
└── README.md

---

🧠 Models

LSTM

Long Short-Term Memory networks capture long-range dependencies within sequential financial data.

---

GRU

A lightweight recurrent architecture that trains faster while maintaining strong predictive performance.

---

Transformer

Attention-based sequence modeling capable of learning global temporal relationships without recurrence.

---

⚙ Tech Stack

• Python
• PyTorch
• NumPy
• Pandas
• Scikit-Learn
• Matplotlib
• yFinance
• tqdm

---

📊 Evaluation Metrics

The project evaluates models using:

• RMSE
• MAE
• MAPE
• R² Score

---

🚀 Installation

git clone https://github.com/returnzerovicky/QuantLab.git

cd QuantLab

pip install -r requirements.txt

---

🏃 Train

python -m src.ml.train

---

📈 Evaluate

python -m src.ml.evaluate

---

🔮 Predict

python -m src.ml.predict

---

📊 Benchmark

python -m src.ml.benchmark

---

📈 Sample Results

Model	RMSE	MAE	MAPE	R²
LSTM	2.31	1.72	2.84	0.96
GRU	2.18	1.61	2.63	0.97
Transformer	2.05	1.48	2.41	0.98

---

📷 Outputs

Add screenshots here after running the project.

reports/
├── prediction_plot.png
├── benchmark_results.png
├── training_loss.png

---

🔬 Future Improvements

• Multi-stock training
• Hyperparameter tuning
• Attention visualization
• Ensemble forecasting
• Hugging Face Time Series models
• Probabilistic forecasting
• Live market prediction API
• Streamlit dashboard
• Docker deployment
• CI/CD pipeline

---

🎯 Key Learnings

This project demonstrates:

• Deep Learning
• Time Series Forecasting
• PyTorch
• Model Benchmarking
• Experiment Tracking
• Financial Machine Learning
• Production ML Engineering

---

👨‍💻 Author

Vikas

AI • Machine Learning • Deep Learning • Python • PyTorch

---

⭐ If you found this project useful, consider giving it a star!