TrainSum.decomposition(*, decomposition: T, ncores: int = 2, optimizer: Literal['optimal', 'dp', 'greedy', 'random-greedy', 'random-greedy-128', 'branch-all', 'branch-2', 'auto', 'auto-hq'] = DEFAULT_OPTIMIZER, direction: Direction = Direction.TO_RIGHT) → DecompositionOptions[T]

TrainSum.decomposition(*, max_rank: int, cutoff: float = 0.0, ncores: int = 2, optimizer: Literal['optimal', 'dp', 'greedy', 'random-greedy', 'random-greedy-128', 'branch-all', 'branch-2', 'auto', 'auto-hq'] = DEFAULT_OPTIMIZER, direction: Direction = Direction.TO_RIGHT) → DecompositionOptions[SVDecomposition[NDArray]]

Einsum operations based on matrix decompositions. One can provide either a decomposition object or paramters for a singular value decomposition (max_rank, cutoff). ncores specifies the number of cores to be used for the sweeping strategy.

Parameters:

• max_rank (int)

• decomposition (Any | None)

• cutoff (float)

• ncores (int)

• optimizer (Literal['optimal', 'dp', 'greedy', 'random-greedy', 'random-greedy-128', 'branch-all', 'branch-2', 'auto', 'auto-hq'])

• direction (Direction)

Return type:

DecompositionOptions[Any]