alphanetworks 0.3.4

Custom deep learning layers and hybrid models for image classification.

AlphaNetworks

AlphaNetworks is a Python package designed to train advanced image classification models using hybrid architectures like ResNet50V2 and DenseNet169. It provides custom deep learning layers and blocks, such as the MoEBlock, MLABlock, and AlphaBlocks, to enhance flexibility and performance.

---

Table of Contents

• Introduction
• Features
• Installation
  • Prerequisites
  • Installation via Pip
• Usage
  • Command-Line Interface
    • Basic Usage
    • Options
    • Help
  • Examples
  • Programmatic Usage
• Troubleshooting
• Project Structure
• Documentation
• Contributing
  • Reporting Issues
• License
• Contact
• Acknowledgments

---

Introduction

AlphaNetworks combines state-of-the-art architectures with custom deep learning layers to offer superior performance in image classification tasks. By leveraging pre-trained weights, advanced optimization techniques, and flexible blocks, it ensures robust feature extraction and better generalization.

---

Features

• Custom Layers: Includes MoEBlock, MLABlock, and AlphaBlocks for advanced deep learning models.
• Hybrid Architecture: Integrates ResNet50V2 and DenseNet169.
• Data Augmentation: Enhances model robustness.
• CLI Support: Train and configure models via the command line.
• Customizable Hyperparameters: Adjust learning rate, batch size, etc.
• Download Pretrained Weights: Automatically fetch required weights.

---

Installation

Prerequisites

• Python: Version 3.6 or higher.
• TensorFlow: Version 2.x or later.
• pip: Latest version recommended.

Installation via Pip

Install AlphaNetworks directly from PyPI:

pip install alphanetworks

---

Usage

Command-Line Interface

Basic Usage

alphanetworks-train --train_dir TRAIN --val_dir VAL [OPTIONS]

Options

• --train_dir or -t: Path to the training dataset.
• --val_dir or -v: Path to the validation dataset.
• --epochs or -e: Number of training epochs (default: 50).
• --batch_size or -b: Batch size for training and validation (default: 8).
• --learning_rate or -lr: Initial learning rate for the Adam optimizer (default: 0.0001).
• --num_classes or -nc: Number of target classes for classification (default: 10).
• --output_dir or -o: Directory to save model weights and reports (default: ./output).

Help

For a detailed list of options, run:

alphanetworks-train --help

---

Examples

Example 1: Basic Training

alphanetworks-train --train_dir ./data/train --val_dir ./data/val --epochs 20

Example 2: Custom Parameters

alphanetworks-train --train_dir ./data/train --val_dir ./data/val --epochs 50   --batch_size 16 --learning_rate 0.0005 --num_classes 20 --output_dir ./models

---

Programmatic Usage

from alphanetworks import build_model

# Build the model
model = build_model(input_shape=(224, 224, 3), num_classes=10)

# Compile the model
model.compile(
    optimizer="adam",
    loss="categorical_crossentropy",
    metrics=["accuracy"]
)

# Train the model
model.fit(train_data, validation_data=val_data, epochs=30)

---

Troubleshooting

Download Errors

During runtime, pre-trained weights for ResNet50V2 or DenseNet169 are automatically downloaded. Ensure you have a stable internet connection.

---

Project Structure

alphanetworks/
├── alphanetworks/
│   ├── __init__.py
│   ├── layers.py
│   ├── models.py
├── scripts/
│   └── train_model.py
├── setup.py
├── LICENSE
├── README.md

---

Documentation

For detailed documentation, visit the GitHub repository.

---

Contributing

Contributions are welcome! Please submit a pull request or open an issue.

---

License

This project is licensed under the MIT License.

---

Contact

For any queries, reach out to the author at mail.

---

Acknowledgments

• TensorFlow and Keras Teams
• Researchers of ResNet and DenseNet