mlq4st is a Python package for conditional distribution modeling and simulation of spatio-temporal processes. It combines machine-learning quantile regression (to learn non-Gaussian, covariate-dependent marginals) with a latent Gaussian random field (GRF) (to enforce coherent space-time dependence).

• HAL preprint: https://hal.science/hal-05441043/

• Conditional marginals Y | X via quantile regression:
  • KNN-based conditional CDF (knn)
  • Quantile Regression Forests (qrf)
  • Quantile Regression Neural Networks (qrnn, via quantnn)
• Latent Gaussian mapping (Gaussian copula):
  • U = F_{Y|X}(y)
  • Z = Phi^{-1}(U) (Phi = standard normal CDF)
• Spatio-temporal dependence in latent space with GRFs (e.g., Matérn–Gneiting)
• Optional hyperparameter selection via time-series cross-validation (depending on method)

pip install "git+https://github.com/sobakrim/mlq4st.git@main"