tensory: Tensory - Modern Tensor Operations for R

Provides dense and decomposed tensor classes for R together with operations that use an API similar to the MATLAB Tensor Toolbox, including matricized products, contractions, reshaping helpers, and symmetry utilities. Selected kernels can delegate to `xtensor`/`Rcpp` backends for performance.

tensory provides dense and decomposed tensor classes together with an API similar to the MATLAB Tensor Toolbox for R users.

Details

The tensory package provides tensor algebra tools for R with an API similar to the MATLAB Tensor Toolbox. It includes dense tensors, matricized tensors, and decomposed tensor representations, with selected operations able to delegate to optimized C++ backends.

Key Features

R6-based tensor classes for dense and decomposed representations
Operations with names and semantics similar to the MATLAB Tensor Toolbox such as ttm(), ttt(), ttv(), permute(), and contract()
Dense helpers for unfolding, vectorization, symmetry checks, and matricized tensor products
Optional C++ acceleration hooks via xtensor and Rcpp

Main Classes

Tensor: The core R6 class for tensor operations
Tenmat: Matricized tensor representation
KTensor: Kruskal tensor representation
TTensor: Tucker tensor representation

Main Functions

tensor: Create tensor objects from R data structures
zeros: Create tensors filled with zeros
ones: Create tensors filled with ones
ttm: Tensor times matrix or vector
ttt: Tensor-times-tensor products
permute: Reorder tensor dimensions
mttkrp: Matricized tensor times Khatri-Rao product

Performance

The package is structured so dense operations can use optimized C++ implementations when available while preserving an R-friendly interface and predictable tensor shapes.

Getting Started


# Create a tensor from a matrix
t <- tensor(matrix(1:6, nrow=2, ncol=3))
print(t)

# Create a 3D tensor
t3d <- tensor(1:24, c(2, 3, 4))

# Mathematical operations
result <- t3d$clone_tensor()$add(t3d)$multiply(2)

Author

Dayu Sun