MuyGPyS (v0.9.0) Reference

MuyGPyS is toolkit for training approximate Gaussian Process (GP) models using the MuyGPs (Muyskens, Goumiri, Priest, Schneider) algorithm.

Citation

If you use MuyGPyS in a research paper, please reference our article:

@article{muygps2021,
  title={MuyGPs: Scalable Gaussian Process Hyperparameter Estimation Using Local Cross-Validation},
  author={Muyskens, Amanda and Priest, Benjamin W. and Goumiri, Im{\`e}ne and Schneider, Michael},
  journal={arXiv preprint arXiv:2104.14581},
  year={2021}
}

Package Documentation:

• neighbors
  • NN_Wrapper
• gp
  • deformation
  • tensors
  • hyperparameter
  • kernels
  • metric
  • noise
  • MuyGPS
  • MultivariateMuyGPS
• optimize
  • batch
  • chassis
  • loss
  • objective
• examples
  • regress
  • fast_posterior_mean
  • classify
  • two-class classify with uq
  • muygps_torch
• torch
  • muygps_layer

Examples:

• Univariate Regression Tutorial
  • Sampling a Curve from a Conventional GP
  • Constructing Nearest Neighbor Lookups
  • Sampling Batches of Data
  • Setting and Optimizing Hyperparameters
  • Inference
• Illustrating MuyGPs Sparsification, Prediction, and Uncertainty Quantification
  • Sampling a 2D Surface from a Conventional GP
  • Nearest Neighbors Sparsification
  • Comparing MuyGPs to Conventional GP Posteriors
• Deep Kernels with MuyGPs in PyTorch Tutorial
  • Training a Deep Kernel MuyGPs Model
• Fast Posterior Mean Tutorial
  • Benchmarking Fast Prediction
  • Comparison with Conventional Prediction
• Anisotropic Metric Tutorial
  • Sampling a 2D Surface from a Conventional GP
  • Training an Anisotropic Model
  • Inference
  • Results comparison
• Loss Function Tutorial
  • Variance-free Loss Functions
  • Variance-Sensitive Loss Functions

Variable Name Conventions

We make use of several canonical variable names that refer to tensor shape dimensions. Here is a partial list of the major names and their meanings.

• train_count - the number of training observations.

• test_count - the number of test or prediction observations.

• batch_count - the number of elements to be predicted. Can coincide with train_count or test_count depending on usage. Sometimes also called data_count.

• feature_count - the number of features in the observations. Omitted for univariate feature spaces.

• response_count - the number of response variables. Omitted for univariate responses.

• nn_count - the number of nearest neighbors upon which predictions are conditioned.

• out_shape - a tuple referring to the shape associated with the output shape of the cross-covariance. For a univariate problem, in_shape = (nn_count,). For a multivariate problem, out_shape most likely refers to (nn_count, response_count).

• in_shape - a tuple referring to the shape associated with how the covariance is conditioned on observations. For a univariate problem, in_shape == (nn_count,). For a multivariate problem, in_shape might refer to (nn_count, response_count), but could instead have a different second element if the observations do not come from the same space as the predictions.

Resources:

• References

Indices and tables

• Index

• Module Index

• Search Page