ResNet Hyperparameter Tuning with KerasTuner

This project demonstrates how to build and tune a ResNet-inspired deep learning model for the Fashion MNIST dataset using TensorFlow and KerasTuner. The code leverages hyperparameter tuning to optimize the model architecture and training process.

Table of Contents

• Features
• Requirements
• Dataset
• Code Overview
• Running the Code
• Results
• Acknowledgments

---

Features

• Implements a ResNet-inspired architecture with residual blocks.
• Uses KerasTuner for hyperparameter optimization, including:
  • Number of filters in the initial convolutional layer.
  • Number of residual blocks.
  • Filters in each residual block.
  • Learning rate for the Adam optimizer.
• Performs hyperparameter tuning to maximize validation accuracy.
• Evaluates the best model on the test dataset.

Requirements

To run this project, you need the following libraries installed:

• TensorFlow 2.0+
• KerasTuner

You can install the required libraries using the following command:

pip install -r requirements.txt

Dataset

The project uses the Fashion MNIST dataset, which contains 70,000 grayscale images of 28x28 pixels, categorized into 10 classes. The dataset is split into:

• 60,000 training samples
• 10,000 test samples

The dataset is automatically loaded using TensorFlow's keras.datasets module.

Code Overview

1. Data Preprocessing:

   • Load the Fashion MNIST dataset.
   • Normalize the image pixel values to the range [0, 1].
   • Convert labels to one-hot encoded format.

2. Residual Block Function:

   • Implements a residual block with two convolutional layers and a shortcut connection.

3. HyperModel Class:

   • Defines a custom ResNetModel class that builds the model with hyperparameter tuning for the number of filters, residual blocks, and learning rate.

4. KerasTuner:

   • Uses the RandomSearch tuner to explore hyperparameter combinations.
   • Optimizes the model for validation accuracy.

5. Training and Evaluation:

   • Trains the model with the best hyperparameters.
   • Evaluates the final model on the test dataset.

Running the Code

1. Clone the repository or copy the code to your local environment.

2. Ensure all dependencies are installed (see Requirements).

3. Run the script:

   python resnet.py

4. The tuning process will output the best hyperparameters and save the results in the specified directory.

Results

After running the hyperparameter tuning, the best model's performance on the test dataset is evaluated and displayed. The test loss and accuracy are printed as:

test loss: 0.3063, test accuracy: 0.8876 # Your results may vary

Acknowledgments

• TensorFlow: For providing the framework to build and train deep learning models.
• KerasTuner: For facilitating hyperparameter optimization.
• Fashion MNIST: For the dataset used in this project.