Right now, there are two ways to load a term-topic probability matrix: By calling either MatrixState.importMatrix() or MatrixState.importEntries().
The former loads a 2D array of numbers and the latter loads an 1D array of non-zero entries. For matrices with a large number of zeros (in our case), the latter is the more efficient representation. However, the client code currently passes the matrix from MatrixState to MatrixModel using a full matrix representation, nullifying any advantage we get from loading a sparse matrix.
Right now, there are two ways to load a term-topic probability matrix: By calling either MatrixState.importMatrix() or MatrixState.importEntries().
The former loads a 2D array of numbers and the latter loads an 1D array of non-zero entries. For matrices with a large number of zeros (in our case), the latter is the more efficient representation. However, the client code currently passes the matrix from MatrixState to MatrixModel using a full matrix representation, nullifying any advantage we get from loading a sparse matrix.