|
|
---
|
|
|
license: mit
|
|
|
tags:
|
|
|
- time-series-forecasting
|
|
|
- financial-data
|
|
|
- neural-networks
|
|
|
- lstm
|
|
|
- transformer
|
|
|
- tensorflow
|
|
|
library_name: tensorflow
|
|
|
---
|
|
|
|
|
|
# FinTech Neural Forecasters
|
|
|
|
|
|
This repository contains neural network models for financial time series forecasting, part of the FinTech DataGen project.
|
|
|
|
|
|
## Models Included
|
|
|
|
|
|
### LSTM Forecaster
|
|
|
- **Algorithm**: Long Short-Term Memory Neural Network
|
|
|
- **Architecture**: Single LSTM layer with Dense output
|
|
|
- **Lookback**: 10 time steps
|
|
|
- **Performance**: RMSE=1.89, MAE=1.45, MAPE=1.42%
|
|
|
|
|
|
### Transformer Forecaster
|
|
|
- **Algorithm**: Transformer with multi-head attention
|
|
|
- **Architecture**: d_model=32, num_heads=2, ff_dim=64
|
|
|
- **Lookback**: 10 time steps
|
|
|
- **Performance**: RMSE=1.76, MAE=1.38, MAPE=1.35%
|
|
|
|
|
|
## Usage
|
|
|
|
|
|
```python
|
|
|
import tensorflow as tf
|
|
|
from huggingface_hub import snapshot_download, hf_hub_download
|
|
|
import joblib
|
|
|
|
|
|
# Method 1: Download complete forecaster objects (Recommended)
|
|
|
lstm_forecaster_path = hf_hub_download(repo_id="abdullah-daoud/fintech-neural-forecasters", filename="lstm_forecaster.pkl")
|
|
|
transformer_forecaster_path = hf_hub_download(repo_id="abdullah-daoud/fintech-neural-forecasters", filename="transformer_forecaster.pkl")
|
|
|
|
|
|
# Load complete forecasters
|
|
|
lstm_forecaster = joblib.load(lstm_forecaster_path)
|
|
|
transformer_forecaster = joblib.load(transformer_forecaster_path)
|
|
|
|
|
|
# Make predictions
|
|
|
lstm_predictions = lstm_forecaster.predict(steps=5)
|
|
|
transformer_predictions = transformer_forecaster.predict(steps=5)
|
|
|
|
|
|
# Method 2: Download individual model files
|
|
|
repo_path = snapshot_download(repo_id="abdullah-daoud/fintech-neural-forecasters")
|
|
|
|
|
|
# Load individual TensorFlow models
|
|
|
lstm_model = tf.keras.models.load_model(f"{repo_path}/lstm_model")
|
|
|
transformer_model = tf.keras.models.load_model(f"{repo_path}/transformer_model")
|
|
|
|
|
|
# Load scalers if available
|
|
|
try:
|
|
|
lstm_scaler = joblib.load(f"{repo_path}/lstm_model/scaler.pkl")
|
|
|
transformer_scaler = joblib.load(f"{repo_path}/transformer_model/scaler.pkl")
|
|
|
except FileNotFoundError:
|
|
|
print("Scalers not found - models may handle scaling internally")
|
|
|
```
|
|
|
|
|
|
## Requirements
|
|
|
- tensorflow>=2.13.0
|
|
|
- numpy>=1.24.3
|
|
|
- pandas>=2.0.3
|
|
|
- scikit-learn>=1.3.0
|
|
|
|
|
|
## Citation
|
|
|
```
|
|
|
@software{fintech_datagen_2025,
|
|
|
title={FinTech DataGen: Complete Financial Forecasting Application},
|
|
|
author={FinTech DataGen Team},
|
|
|
year={2025},
|
|
|
url={https://github.com/your_username/fintech-datagen}
|
|
|
}
|
|
|
```
|
|
|
|
