compressed-sparse-fiber

CSF is a generalization of compressed sparse row (CSR) index. See smith2017knl

CSF index recursively compresses each dimension of a tensor into a set of prefix trees. Each path from a root to leaf forms one tensor non-zero index. CSF is implemented with two arrays of buffers and one arrays of integers.

Example usage

let rows = vec![
    (vec![1, 1, 1, 2], 1.0),
    (vec![1, 1, 1, 3], 2.0),
    (vec![1, 2, 1, 1], 3.0),
    (vec![1, 2, 1, 3], 4.0),
    (vec![1, 2, 2, 1], 5.0),
    (vec![2, 2, 2, 1], 6.0),
    (vec![2, 2, 2, 2], 7.0),
    (vec![2, 2, 2, 3], 8.0),
];

let csf: CompressedSparseFiber<_,_> = rows.into_iter().collect();                       

The above could be represented by a structure like this:

CompressedSparseFiber { 
    fptr: [[0, 2, 3], [0, 1, 3, 4], [0, 2, 4, 5, 8]], 
    fids: [[1, 2], [1, 2, 2], [1, 1, 2, 2], [2, 3, 1, 3, 1, 1, 2, 3]], 
    vals: [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0] 
}

Retreiving in uncompressed form

for t in csf {
    println!("{:?}", t);
}

References

• Apache Arrow
• smith2017knl
• Tensor-Matrix Products with a Compressed Sparse Tensor