Handwritten Digit Classification with Keras and MNIST

This project implements a deep learning model using Keras (with TensorFlow backend) to classify handwritten digits from the MNIST dataset. The model is trained, evaluated, and saved for future inference. All steps — from data preprocessing to visualization of training metrics — are included.

---

📌 Overview

• Dataset: MNIST – 60,000 training and 10,000 test images of handwritten digits (28x28 grayscale pixels).
• Task: Multiclass image classification (digits 0–9).
• Framework: TensorFlow 2.x with Keras.
• Features:
  • Data normalization and efficient input pipeline using tf.data.
  • Deep neural network with dropout for regularization.
  • Training with Adam optimizer and evaluation on test set.
  • Model saving in Keras format.
  • Training visualization (loss & accuracy curves).
  • Integration with TensorBoard for monitoring.

---

🏗️ Model Architecture

A simple feedforward neural network (Multilayer Perceptron) built using tf.keras.Sequential:

model = tf.keras.Sequential([
    tf.keras.layers.Flatten(input_shape=(28, 28)),
    tf.keras.layers.Dense(512, activation='relu'),
    tf.keras.layers.Dropout(0.2),
    tf.keras.layers.Dense(10)
])

⚙️ Training Pipeline

The data pipeline uses tensorflow_datasets and tf.data for efficient loading and preprocessing:

• Normalization: Images converted from uint8 to float32 and scaled to [0, 1].
• Batching: 128 samples per batch.
• Caching & Prefetching: Improves training speed.
• Shuffling: Applied only on the training set.

---

🧪 Training & Evaluation

• Optimizer: Adam (learning_rate=0.001 by default).
• Loss: SparseCategoricalCrossentropy(from_logits=True).
• Metrics: Sparse Categorical Accuracy.
• Epochs: 10.
• Validation: Performance monitored on the test set.

After training, the model is evaluated on the test dataset to report final accuracy.

---

💾 Model Saving

The trained model is saved in the native Keras format (.keras) for persistence and future inference:

./deep_learning/classificatore_cifre/model_saved.keras

This format preserves the architecture, weights, and training configuration.

---

📈 Training Visualization

Two plots are generated at the end of training:

1. Training vs Validation Loss
2. Training vs Validation Accuracy

These help assess convergence and overfitting.

Additionally, the model architecture is visualized and saved as:

./deep_learning/classificatore_cifre/model.png

Using keras.utils.plot_model with shapes and layer names.

---

📊 TensorBoard Logging

Training logs are saved to:

./deep_learning/classificatore_cifre/logs

You can launch TensorBoard with:

tensorboard --logdir=./deep_learning/classificatore_cifre/logs

---

🖼️ Example Output

---

✅ Requirements

To run this project, install the required packages:

pip install tensorflow tensorflow-datasets matplotlib

---

🚀 How to Run

1. Clone the repository.
2. Run the script in a Python environment (e.g., Jupyter Notebook or .py file).
3. View results and plots.
4. (Optional) Launch TensorBoard to monitor training.

---

📁 Project Structure

deep_learning/
└── classificatore_cifre/
    ├── model_saved.keras      # Saved trained model
    ├── model.png              # Model architecture diagram
    ├── logs/                  # TensorBoard logs
    └── training_plots.png     # Accuracy and loss curves

---

📚 Resources

• TensorFlow Keras Tutorial
• MNIST Dataset on TensorFlow Datasets
• Keras Sequential Model

---