The A / B / C matrices are quite sparse (high density of 0 values), enabling use of efficient sparse matrix data structures such as [CSR](https://stackoverflow.com/questions/52299420/scipy-csr-matrix-understand-indptr). Nova has a commit here: https://github.com/microsoft/Nova/commit/93afdf3f017c5c20e852da1401525c7b144205da
The A / B / C matrices are quite sparse (high density of 0 values), enabling use of efficient sparse matrix data structures such as CSR.
Nova has a commit here: microsoft/Nova@93afdf3