pyarrow.SparseCSFTensor#

class pyarrow.SparseCSFTensor#

Bases: _Weakrefable

A sparse CSF tensor.

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

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.

Examples

>>> import pyarrow as pa
>>> import numpy as np
>>> # Create a 3D sparse tensor
>>> dense_tensor = np.zeros((2, 3, 2), dtype=np.float32)
>>> dense_tensor[0, 1, 0] = 1.0
>>> dense_tensor[1, 2, 1] = 2.0
>>> sparse_csf = pa.SparseCSFTensor.from_dense_numpy(dense_tensor)
>>> sparse_csf
<pyarrow.SparseCSFTensor>
type: float
shape: (2, 3, 2)

__init__(*args, **kwargs)#

Methods

__init__(*args,**kwargs)
dim_name(self,i)	Returns the name of the i-th tensor dimension.
equals(self,SparseCSFTensorother)	Return true if sparse tensors contains exactly equal data
from_dense_numpy(cls,obj[,dim_names])	Convert numpy.ndarray to arrow::SparseCSFTensor
from_numpy(data,indptr,indices,shape[,...])	Create arrow::SparseCSFTensor from numpy.ndarrays
from_tensor(obj)	Convert arrow::Tensor to arrow::SparseCSFTensor
to_numpy(self)	Convert arrow::SparseCSFTensor to numpy.ndarrays with zero copy
to_tensor(self)	Convert arrow::SparseCSFTensor to arrow::Tensor

Attributes

dim_names
is_mutable
ndim
non_zero_length
shape
size
type

dim_name(self, i)#

Returns the name of the i-th tensor dimension.

Parameters:

iint

The physical index of the tensor dimension.

Returns:

str

dim_names#

equals(self, SparseCSFTensor other)#

Return true if sparse tensors contains exactly equal data

Parameters:

otherSparseCSFTensor

The other tensor to compare for equality.

classmethod from_dense_numpy(cls, obj, dim_names=None)#

Convert numpy.ndarray to arrow::SparseCSFTensor

Parameters:

objnumpy.ndarray

Data used to populate the rows.

dim_nameslist[str], optional

Names of the dimensions.

Returns:

pyarrow.SparseCSFTensor

static from_numpy(data, indptr, indices, shape, axis_order=None, dim_names=None)#

Create arrow::SparseCSFTensor from numpy.ndarrays

Parameters:

datanumpy.ndarray

Data used to populate the sparse tensor.

indptrnumpy.ndarray

The sparsity structure. Each two consecutive dimensions in a tensor correspond to a buffer in indices. A pair of consecutive values at indptr[dim][i] indptr[dim][i + 1] signify a range of nodes in indices[dim + 1] who are children of indices[dim][i] node.

indicesnumpy.ndarray

Stores values of nodes. Each tensor dimension corresponds to a buffer in indptr.

shapetuple

Shape of the matrix.

axis_orderlist, optional

the sequence in which dimensions were traversed to produce the prefix tree.

dim_nameslist, optional

Names of the dimensions.

Examples

>>> import pyarrow as pa
>>> import numpy as np
>>> data = np.array([1.0, 2.0], dtype=np.float32)
>>> indptr = [
...     np.array([0, 1, 2], dtype=np.int64),
...     np.array([0, 1, 2], dtype=np.int64),
... ]
>>> indices = [
...     np.array([0, 1], dtype=np.int64),
...     np.array([1, 2], dtype=np.int64),
...     np.array([0, 1], dtype=np.int64),
... ]
>>> sparse_csf = pa.SparseCSFTensor.from_numpy(
...     data, indptr, indices, shape=(2, 3, 2))
>>> sparse_csf
<pyarrow.SparseCSFTensor>
type: float
shape: (2, 3, 2)

static from_tensor(obj)#

Convert arrow::Tensor to arrow::SparseCSFTensor

Parameters:

objTensor

The dense tensor that should be converted.

is_mutable#

ndim#

non_zero_length#

shape#

size#

to_numpy(self)#

Convert arrow::SparseCSFTensor to numpy.ndarrays with zero copy

to_tensor(self)#

Convert arrow::SparseCSFTensor to arrow::Tensor

type#