Digit Classification

An ML / MLOps classification project on MNIST - Database of Handwritten Digits implemented using convolutional neural network (ConvNet / CNN) with the help of PyTorch and DVC libraries.

DAGsHub Link (for experimentation and pipelining): Click Here!

To build from source

1. Install Git and DVC.

2. Clone the repository.

   git clone https://github.com/swarajpande4/Digit-Classification.git
   
   cd Digit-Classification/
   

3. Set up virtual environment for python.

   pip install virtualenv
   
   virtualenv venv/
   
   source venv/bin/activate
   
   pip install -r requirements.txt
   

4. Run the following command to execute the pipeline after making changes to the code/ scripts.

   dvc repro
   

5. Deactivate the virtual environment.

   deactivate
   

Files and Directory Structure

.
├── .github/workflows
    └── cml.yaml                    // CML workflow for GitHub Actions
├── code
    ├── eval.py                     // Evaluation metrics script
    ├── featurization.py            // Featurization script
    ├── get_data.py                 // Fetches the datasets for CML container (GitHub Actions Job)
    ├── model_class.py              // Model Class script
    └── train_model.py              // Trains the model instance
├── data
    ├── model.pkl                   (DVC)
    ├── norms_params.json           (DVC)
    ├── processed_test_data.npy     (DVC)
    ├── processed_train_data.npy    (DVC)
    ├── test_data.csv               (DVC)
    ├── train_data.csv              (DVC)
    ├── train_data.csv.dvc
    └── test_data.csv.dvc        
├── metrics 
    ├── confmat.png                 // Confusion matrix displayed by GitHub Actions
    ├── eval.json                   // Evaluation metrics for pipeline 
    ├── metrics.txt                 // Evaluation metrics displayed by Github Actions                       
    └── train_metric.json           
├── notebook
    └── notebook.ipynb              // Jupyter Notebook 
├── dvc.lock
├── dvc.yaml
├── requirements.txt
└── README.md