embedded-adaptive-stacking-keras 0.5

Biblioteca para detecção de outliers baseada em Stacking com Keras

📌 EAS - Embedded Adaptive Stacking

EAS (Embedded Adaptive Stacking) is a library for pattern detection in time series using model stacking with LSTM, GRU, BiLSTM, and BiGRU.

🚀 About the Project

EAS (Embedded Adaptive Stacking) is a library for time series analysis using model stacking based on recurrent neural networks.

🔹 Key Features

✔ Smart Stacking: Combines LSTM, GRU, BiLSTM, and BiGRU to improve predictions.
✔ Dynamic Optimization: Includes the LossAdaptiveOptimizer (LORO), which automatically adjusts the learning rate.
✔ Cost-Sensitive Loss Function: CustomLossWithRegression allows fine-tuning penalties for extreme events.
✔ Results Visualization: Clear comparison between predictions and actual values.

---

📀 Installation

To install the library directly from PyPI, use:

pip install adaptive-stacking-keras

Or to manually install the latest version from the repository:

git clone https://github.com/your-username/adaptive-stacking-keras.git
cd adaptive-stacking-keras
pip install .

---

📀 How to Use

🔹 Usage Example

Here is a simple example of how to use the library to train a model with Stacking and dynamic optimization.

import tensorflow as tf
from adaptive_stacking_keras import (
    StackingModel,
    CustomLossWithRegression,
    LORO,
    plot_time_series_comparison,
)

# Creating the model with multiple hidden layers
hidden_dims = [64, 128]
model = StackingModel(input_dim=10, hidden_dims=hidden_dims, embedding_dim=32, output_dim=1)

# Creating LORO optimizer
optimizer = LORO(learning_rate=0.001)

# Generating synthetic data
tf.random.set_seed(42)
x_train = tf.random.normal((100, 10, 10))
y_train = tf.random.normal((100, 1))

# Training the model
for epoch in range(5):  # Only 5 epochs for demonstration
    with tf.GradientTape() as tape:
        y_pred, (threshold, alpha) = model(x_train)
        loss = CustomLossWithRegression(model.threshold_alpha_layer)(y_train, y_pred)
    gradients = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(gradients, model.trainable_variables))
    print(f"Epoch [{epoch+1}/5] - Loss: {loss.numpy():.4f} - Threshold: {threshold.numpy():.4f} - Alpha: {alpha.numpy():.4f}")

# Testing with data
x_test = tf.random.normal((50, 10, 10))
y_test = tf.random.normal((50, 1))
y_pred, _ = model(x_test)

# Visualization of results
plot_time_series_comparison(y_test, y_pred, time_range=(10, 40), title="Model Results")

---

📝 Documentation

Check out the full documentation on the GitHub repository.

💎 Contributions

Contributions are welcome! To contribute:

1. Fork this repository.
2. Create a branch with your feature (git checkout -b my-feature).
3. Commit your changes (git commit -m 'Adding new feature').
4. Push to the repository (git push origin my-feature).
5. Open a Pull Request.

🌟 License

This project is licensed under the MIT License - see the LICENSE file for more details.

---

💪 Built with dedication for developers and researchers!