MLP Diabetes Classifier

A simple project to train and evaluate a multilayer perceptron on the Pima Indians Diabetes Dataset using TensorFlow, SciKeras, and Scikit-Learn.

---

Installation

1. Clone the repo

git clone https://github.com/yourusername/mlp-diabetes-classifier.git
cd mlp-diabetes-classifier

2. Set up the Conda environment

It is included an environment.yml for Conda users:

conda env create -f environment.yml
conda activate mlp-diabetes

Dependencies

• Python 3.7+
• numpy, scikt-learn, tensorflow, scikeras, PyYAML, matplotlib.

You can also install them with:

pip install -r requirements.txt

Usage

1. Verify the dataset

The Pima Indians Diabetes Dataset from Kaggle is already included under data/diabetes.csv.

2. Adjust settings

Open config.yamland tweak any values you like (seed, test_size_hyperparameters list, etc.)

3. Run the full pipeline

python run_all.py

This will:

• Train de MLP with randomized hyperparameter search
• Save the best model to models/best_mlp.h5
• Print train/test accuracy and predictions
• Save evaluation plots to reports/

Project Structure

mlp-diabetes-classifier/
│
├── config.yaml          # Experiment settings
├── environment.yml      # Conda environment spec
├── requirements.txt     # Pinned pip dependencies (for Docker)
├── docker-compose.yml   # Docker Compose setup
├── Dockerfile           # Image build definition
├── .dockerignore       
├── .gitignore           
├── pytest.ini           
│
├── data/
│   └── diabetes.csv     # Pima Indians Diabetes Dataset
│
├── models/              # (Auto-created) Trained model & params
│
├── reports/
│   └── figures          # (Auto-created) Plots
│
├── tests/               # Test suite
│   ├── unit
│   ├── integration
│   └── e2e  
│       
├── src/
│   ├── config.py        # Loads config.yaml
│   ├── data_loader.py   # Reads & splits data
│   ├── model_builder.py # Defines the Keras MLP
│   ├── train.py         # Hyperparameter search & model saving
│   └── evaluate.py      # Loads model & prints metrics
│
└── run_all.py           # Runs train.py then evaluate.py

Dockerized Support

This project is fully containerized for portability and reproducibility.

Docker Dependencies

Before using Docker, you need to have the following installed locally on your system:

• Docker Engine
• Docker Compose

✅ Note: These tools are required only if you want to run the project in a containerized environment. If you're using Conda, Docker is optional.

How to Run

To build the image and run the project inside a container:

docker-compose up --build

This will:

• Build the Docker image using the included Dockerfile
• Run the run_all.py pipeline (training + evaluation)
• Save the best trained model in the models/ directory
• Save plots and metrics in the reports/ directory

✅ Note: Both models/and reports/ are mounted to your host machine, so your outputs are preserved outside the container.

Testing

The project includes a complete test suite using pytest. Tests use temporary directories, mock inputs, and validate expected outputs including saved models and plots.

Run all tests

pytest

This will automatically discover and run:

• Unit tests (tests/unit/)
• Integration tests (tests/integration/)
• End-to-End tests (tests/e2e/)

Run a specific group

pytest tests/unit/