API reference¶
This page is the canonical entry point for the public API. Detailed signatures live in the per-module pages (Functional Module, VSA Models, Embeddings Module, Models Module, Memory Module, Utilities Module); this page is the high-level index.
PVSA — probabilistic primitives¶
|
A hypervector distributed as \(\mathcal{N}(\mu, \mathrm{diag}(\sigma^2))\). |
|
A Dirichlet distribution over the probability simplex \(\Delta_K\). |
|
A mixture-of-Gaussian posterior over hypervectors. |
|
Bind two independent Gaussian HVs under element-wise multiplication. |
|
Bundle a batch of independent Gaussian HVs by summation. |
|
Cyclically permute a Gaussian hypervector. |
|
Retrieve the entry in |
|
JIT-friendly cleanup against a stacked memory of Gaussian hypervectors. |
|
Approximate inverse of a Gaussian HV under element-wise product binding. |
|
Plug-in estimator of the expected cosine similarity under |
|
First-order variance of the dot product \(\langle X, Y \rangle\). |
|
KL divergence \(\mathrm{KL}(p \| q)\) for two diagonal Gaussians. |
|
Bind two Dirichlet HVs by element-wise mean product, re-normalised. |
|
Bundle a batch of Dirichlet HVs by summing concentrations. |
|
Closed-form KL divergence \(\mathrm{KL}(p \| q)\) for two Dirichlets. |
Bayesian classifiers¶
|
Gaussian-posterior classifier — one |
|
Streaming Gaussian-posterior classifier with Kalman updates. |
|
Bounded-memory streaming classifier with exponential decay. |
Uncertainty quantification¶
|
Temperature scaling for post-hoc classifier calibration. |
|
Split-conformal wrapper — prediction sets with coverage guarantee. |
|
Compare an observed statistic to its posterior-predictive distribution. |
|
Audit conformal-predictor coverage across a grid of \(\alpha\). |
Group-theoretic structure¶
|
Apply the \(\mathbb{Z}/d\) cyclic-shift action |
|
Compose two shifts in \(\mathbb{Z}/d\). |
|
The canonical single-argument shift-equivariant bilinear operator. |
|
Check whether |
|
Check whether |
|
Check whether |
Inference helpers¶
|
Evidence lower bound for a Gaussian-posterior PVSA model. |
|
Mean cosine similarity between posterior samples and the target. |
|
Multi-restart MCMC factorisation of a composite PVSA hypervector. |
Multi-device¶
|
Bind two sharded Gaussian hypervector batches across devices. |
|
Bundle a sharded batch of Gaussian hypervectors across devices. |
|
|
|
Reshape |
Classical VSA models¶
|
Multiply-Add-Permute (MAP) coding. |
|
Binary Spatter Codes (BSC). |
|
Holographic Reduced Representations (HRR). |
|
Fourier Holographic Reduced Representations (FHRR). |
|
Binary Sparse Block Codes (B-SBC). |
|
Cyclic Group Representation (CGR). |
|
Modular Composite Representation (MCR). |
|
Vector-Derived Transformation Binding (VTB). |
See Functional Module, VSA Models, Embeddings Module, Models Module, Memory Module, and Utilities Module for full per-module documentation including method signatures, return types, and docstrings.