GradVis AI is a hands-on educational Python project that teaches machine learning foundations through manual implementations of optimization algorithms.
- Linear Regression using gradient descent and MSE
- Logistic Regression using gradient descent and cross-entropy
- Polynomial Regression (optional non-linear extension)
- Optional L1/L2 regularization for optimization intuition
- Interactive Streamlit dashboard with:
- algorithm switching
- synthetic or CSV-based datasets
- hyperparameter controls
- loss curves, parameter trajectories, and fitted boundaries/lines
- calculus and statistics explanations
This project emphasizes:
- Calculus: explicit partial derivatives + update rules
- Statistics: loss, variance/bias intuition, and model diagnostics
- Optimization: convergence behavior with learning-rate and regularization controls
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py- Include at least two numeric columns
- All numeric columns except the last are treated as features (
X) - The final numeric column is treated as target (
y)
app.py
src/gradvis_ai/models/
src/gradvis_ai/utils/
tests/
- No scikit-learn or high-level ML model libraries are used for training algorithms.
- NumPy is used for vectorized math and gradient calculations.