Estimators

This section documents the estimator families available in xyz.

How to navigate this section

• Start with Theory and notation if you want the notation and the identities that connect entropy, mutual information, transfer entropy, and information storage.

• Read Gaussian and linear estimators first if you want the simplest and fastest family.

• Read kNN / KSG estimators if you want the most flexible continuous estimators and the closest conceptual match to ITS/TRENTOOL.

• Read Kernel estimators if you want a fixed-radius view of local neighborhoods.

• Read Discrete (binning) estimators if your data are symbolic or deliberately quantized.

• Read Univariate helper functions for quick helper functions and sanity checks.

• Read Workflows and meta-estimators for bootstrap confidence intervals, greedy source selection, and parallelization (n_jobs).

• Gaussian and linear estimators
  • Implemented classes
  • What these estimators compute
  • Why use the Gaussian family
  • When to prefer it
  • Typical use cases
  • How to use them
  • Parameter guidance
  • Practical advice
  • Interactive example
• kNN / KSG estimators
  • Implemented classes
  • Core mathematical idea
  • Why use KSG
  • When to prefer KSG
  • Typical use cases
  • How to use the KSG estimators
  • Parameter guidance
  • Important caveats
  • ITS parity note
  • Interactive example
• Kernel estimators
  • Implemented classes
  • Mathematical idea
  • Why use kernel estimators
  • When to use them
  • Typical use cases
  • How to use them
  • Choosing the radius r
  • Interactive example
• Discrete (binning) estimators
  • Implemented classes
  • Mathematics
  • Quantization
  • Why use discrete estimators
  • When to use them
  • Typical use cases
  • How to use them
  • Practical advice
  • Interactive example
• Univariate helper functions
  • Available functions
  • What they do
  • How to use them
  • When to use helper functions instead of estimators
  • Interactive example
• Workflows and meta-estimators
  • Bootstrap confidence intervals
  • Greedy source selection (multivariate TE)
  • Multivariate drivers in TE estimators
  • Parallelization (n_jobs)