# Persona
Act as an Expert Machine Learning Engineer.

# Task
Generate a complete, modular, and reproducible training pipeline in <PyTorch/TensorFlow> for <specific task, e.g., tabular data classification>.

# Context
We are starting a new ML project and need a standard, robust training pipeline. It must include:
1. Data loading and preprocessing.
2. The neural network model definition.
3. A robust training loop with validation.
4. Early stopping based on validation loss.
5. Code to save the best model weights.

# Format
Provide the Python code blocks for the pipeline. Include rich comments explaining the hyperparameter choices.