Data science monitoring and developer experience

Overview

This project is designed to train a machine learning model using the uv tool and monitor models using weights and biases (wand). To demontrate these tools, we look at King Rook and King data set and compared Fast Fourier Neural Network and RandomSearch algorithms such as xgboost, decision trees, and bagging. The goal is to see which metrics and evaluations can be used to compare these models. WandB provides model observability and the ability to create dashboards. UV tools is an alternative to pip and other python tools which so far has proven to provide a better developer experience.

Link to current dashboard on weights and biases

Dashboard

Project Structure

krko/
.
├── data                    # Data storage directory
│   ├── external            # External data sources
│   ├── interim             # Intermediate data
│   ├── processed           # Processed data ready for modeling
│   └── raw                 # Raw, unprocessed data
│       └── krkopt.data     # Original dataset file
├── Makefile                # Makefile for automating tasks
├── models                  # Trained models storage
│   └── model_0_0.h5        # Example trained model file
├── notebooks               # Jupyter notebooks for experiments
├── pyproject.toml          # Project dependencies and configuration
├── README.md               # Project overview and instructions
├── reports                 # Generated reports and figures
│   └── figures             # Figures for reports
├── src                     # Source code for the project
│   ├── config.py           # Configuration settings
│   ├── dataset.py          # Data loading and preprocessing
│   ├── features.py         # Feature engineering scripts
│   ├── __init__.py         # Package initialization
│   ├── modeling            # Modeling scripts
│   │   ├── __init__.py     # Package initialization for modeling
│   │   ├── predict.py      # Prediction script
│   │   └── train.py        # Training script
│   ├── models              # Model architectures
│   │   ├── ffnn.py         # Feedforward neural network model
│   │   └── __init__.py     # Package initialization for models
│   └── plots.py            # Plotting utilities
└── uv.loc                  # UV tool configuration file

How to Use uv

To train the model, use the following command:

uv run -m src.modeling.train

To add dependecies

uv add numpy

Make sure you have all the necessary dependencies installed before running the command.

Dependencies

• uv tool from https://docs.astral.sh/uv/
• Python 3.x
• Other dependencies as listed in pyproject.toml

License

This project is licensed under the MIT License.