tt_sketch.sketch

Interface for the streaming and orthogonal sketching algorithms

class tt_sketch.sketch.SketchedTensorTrain(sketch_: SketchContainer, left_drm: DRM, right_drm: DRM)

Container for storing the output of the streaming sketch

Stores the result of the sketch as well as the DRMs used for the sketching. Can be cheaply converted to a tensor train, or the sketch can be efficiently updated using the __add__ method.

C_cores(direction='auto') → List[ndarray[Any, dtype[float64]]]

property Omega_mats: List[ndarray[Any, dtype[float64]]]

property Psi_cores: List[ndarray[Any, dtype[float64]]]

property T: SketchedTensorTrain

Transpose of the tensor.

If a tensor has shape (n1,n2,…,nd), then transpose of the tensor has shape (nd,…,n2,n1).

dot(other: Tensor, reverse=False) → float

Dot product of two tensors

increase_rank(tensor: Tensor, new_left_rank: Union[int, Tuple[int, ...]], new_right_rank: Union[int, Tuple[int, ...]]) → SketchedTensorTrain

Increase the rank of the approximation by performing a new sketch.

Requires DRM with support for the CanIncreaseRank protocol, which currently is only supported by SparseGaussianDRM.

left_drm: DRM

property left_rank: Tuple[int, ...]

right_drm: DRM

property right_rank: Tuple[int, ...]

property size: int

Number of floating point elements used to store the tensor.

sketch_: SketchContainer

to_numpy() → ndarray[Any, dtype[float64]]

Converts the tensor to a (dense) numpy array of same shape.

to_tt() → TensorTrain

tt_sketch.sketch.assemble_sketched_tt(sketch: SketchContainer, direction='auto') → List[ndarray[Any, dtype[float64]]]

Reconstructs a TT from a sketch, using Psi and Omega matrices.

tt_sketch.sketch.blocked_stream_sketch(tensor: Tensor, left_drm: CanSlice, right_drm: CanSlice, left_rank_slices: List[Tuple[int, ...]], right_rank_slices: List[Tuple[int, ...]]) → SketchContainer

Do a blocked sketch.

It’s use is mainly theoretical, since this this would only be faster in a distributed setting (which isn’t properly supported).

tt_sketch.sketch.get_drm_capabilities()

List what all the DRMs are capable of

tt_sketch.sketch.hmt_sketch(tensor: Tensor, rank: Union[int, Tuple[int, ...]], seed: Optional[int] = None, drm_type: Optional[Type[DRM]] = None, drm: Optional[DRM] = None, return_drm: bool = False) → TensorTrain

Perform an orthogonal sketch of a tensor

tt_sketch.sketch.orthogonal_sketch(tensor: Tensor, left_rank: Union[int, Tuple[int, ...]], right_rank: Union[int, Tuple[int, ...]], seed: Optional[int] = None, left_drm_type: Optional[Type[DRM]] = None, right_drm_type: Optional[Type[DRM]] = None, left_drm: Optional[DRM] = None, right_drm: Optional[DRM] = None, return_drm: bool = False) → TensorTrain

Perform an orthogonal sketch of a tensor

tt_sketch.sketch.stream_sketch(tensor: Tensor, left_rank: Union[int, Tuple[int, ...]], right_rank: Union[int, Tuple[int, ...]], seed: Optional[int] = None, left_drm_type: Optional[Type[DRM]] = None, right_drm_type: Optional[Type[DRM]] = None, left_drm: Optional[DRM] = None, right_drm: Optional[DRM] = None, return_drm: bool = False) → SketchedTensorTrain

Perform a streaming sketch of a tensor

tt_sketch.sketch_container

class tt_sketch.sketch_container.SketchContainer(Psi_cores: List[ndarray[Any, dtype[float64]]], Omega_mats: List[ndarray[Any, dtype[float64]]], shape: Optional[Tuple[int, ...]] = None, left_rank: Optional[Tuple[int, ...]] = None, right_rank: Optional[Tuple[int, ...]] = None)

Container class for the Psi_cores and Omega_mats often used internally.

Omega_mats: List[ndarray[Any, dtype[float64]]]

Psi_cores: List[ndarray[Any, dtype[float64]]]

property T: SketchContainer

left_rank: Tuple[int, ...]

right_rank: Tuple[int, ...]

shape: Tuple[int, ...]

classmethod zero(shape: Tuple[int, ...], left_rank: Tuple[int, ...], right_rank: Tuple[int, ...]) → SketchContainer

tt_sketch.sketch_dispatch

Implements methods for dispatching sketching methods for tensors and DRMs.

class tt_sketch.sketch_dispatch.OrthogTTDRM(rank, tensor)

Represents the orthogonalized TT used as left-sketch for psi

class tt_sketch.sketch_dispatch.SketchMethod(value)

An enumeration.

tt_sketch.sketch_dispatch.general_sketch(tensor: Tensor, left_drm: Optional[DRM], right_drm: DRM, method: SketchMethod) → SketchContainer

General algorithm for sketching a tensor.

Does the heavy lifting for both the streaming and orthogonal sketching algorithms.

tt_sketch.sketch_dispatch.orth_step(Psi: ndarray[Any, dtype[float64]], Omega: Optional[ndarray[Any, dtype[float64]]]) → ndarray[Any, dtype[float64]]

Perform the orthogonalization step in the orthogonal sketching algorithm.