sparse – Symbolic Sparse Matrices

In the tutorial section, you can find a sparse tutorial.

The sparse submodule is not loaded when we import Theano. You must import theano.sparse to enable it.

The sparse module provides the same functionality as the tensor module. The difference lies under the covers because sparse matrices do not store data in a contiguous array. Note that there are no GPU implementations for sparse matrices in Theano. The sparse module has been used in:

  • NLP: Dense linear transformations of sparse vectors.
  • Audio: Filterbank in the Fourier domain.

Compressed Sparse Format

This section tries to explain how information is stored for the two sparse formats of SciPy supported by Theano. There are more formats that can be used with SciPy and some documentation about them may be found here.

Theano supports two compressed sparse formats csc and csr, respectively based on columns and rows. They have both the same attributes: data, indices, indptr and shape.

  • The data attribute is a one-dimentionnal ndarray which contains all the non-zero elements of the sparse matrix.
  • The indices and indptr attributes are used to store the position of the data in the sparse matrix.
  • The shape attribute is exactly the same as the shape attribute of a dense (i.e. generic) matrix. It can be explicitly specified at the creation of a sparse matrix if it cannot be infered from the first three attributes.

CSC Matrix

In the Compressed Sparse Column format, indices stands for indexes inside the column vectors of the matrix and indptr tells where the column starts in the data and in the indices attributes. indptr can be thought of as giving the slice which must be applied to the other attribute in order to get each column of the matrix. In other words, slice(indptr[i], indptr[i+1]) corresponds to the slice needed to find the i-th column of the matrix in the data and indices fields.

The following example builds a matrix and returns its columns. It prints the i-th column, i.e. a list of indices in the column and their corresponding value in the second list.

>>> import numpy as np
>>> import scipy.sparse as sp
>>> data = np.asarray([7, 8, 9])
>>> indices = np.asarray([0, 1, 2])
>>> indptr = np.asarray([0, 2, 3, 3])
>>> m = sp.csc_matrix((data, indices, indptr), shape=(3, 3))
>>> m.toarray()
array([[7, 0, 0],
       [8, 0, 0],
       [0, 9, 0]])
>>> i = 0
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([0, 1], dtype=int32), array([7, 8]))
>>> i = 1
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([2], dtype=int32), array([9]))
>>> i = 2
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([], dtype=int32), array([], dtype=int64))

CSR Matrix

In the Compressed Sparse Row format, indices stands for indexes inside the row vectors of the matrix and indptr tells where the row starts in the data and in the indices attributes. indptr can be thought of as giving the slice which must be applied to the other attribute in order to get each row of the matrix. In other words, slice(indptr[i], indptr[i+1]) corresponds to the slice needed to find the i-th row of the matrix in the data and indices fields.

The following example builds a matrix and returns its rows. It prints the i-th row, i.e. a list of indices in the row and their corresponding value in the second list.

>>> import numpy as np
>>> import scipy.sparse as sp
>>> data = np.asarray([7, 8, 9])
>>> indices = np.asarray([0, 1, 2])
>>> indptr = np.asarray([0, 2, 3, 3])
>>> m = sp.csr_matrix((data, indices, indptr), shape=(3, 3))
>>> m.toarray()
array([[7, 8, 0],
       [0, 0, 9],
       [0, 0, 0]])
>>> i = 0
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([0, 1], dtype=int32), array([7, 8]))
>>> i = 1
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([2], dtype=int32), array([9]))
>>> i = 2
>>> m.indices[m.indptr[i]:m.indptr[i+1]], m.data[m.indptr[i]:m.indptr[i+1]]
(array([], dtype=int32), array([], dtype=int64))

List of Implemented Operations

  • Moving from and to sparse
    • dense_from_sparse. Both grads are implemented. Structured by default.
    • csr_from_dense, csc_from_dense. The grad implemented is structured.
    • Theano SparseVariable objects have a method toarray() that is the same as dense_from_sparse.
  • Construction of Sparses and their Properties
    • CSM and CSC, CSR to construct a matrix. The grad implemented is regular.
    • csm_properties. to get the properties of a sparse matrix. The grad implemented is regular.
    • csm_indices(x), csm_indptr(x), csm_data(x) and csm_shape(x) or x.shape.
    • sp_ones_like. The grad implemented is regular.
    • sp_zeros_like. The grad implemented is regular.
    • square_diagonal. The grad implemented is regular.
    • construct_sparse_from_list. The grad implemented is regular.
  • Cast
    • cast with bcast, wcast, icast, lcast, fcast, dcast, ccast, and zcast. The grad implemented is regular.
  • Transpose
    • transpose. The grad implemented is regular.
  • Basic Arithmetic
    • neg. The grad implemented is regular.
    • eq.
    • neq.
    • gt.
    • ge.
    • lt.
    • le.
    • add. The grad implemented is regular.
    • sub. The grad implemented is regular.
    • mul. The grad implemented is regular.
    • col_scale to multiply by a vector along the columns. The grad implemented is structured.
    • row_slace to multiply by a vector along the rows. The grad implemented is structured.
  • Monoid (Element-wise operation with only one sparse input).

    They all have a structured grad.

    • structured_sigmoid
    • structured_exp
    • structured_log
    • structured_pow
    • structured_minimum
    • structured_maximum
    • structured_add
    • sin
    • arcsin
    • tan
    • arctan
    • sinh
    • arcsinh
    • tanh
    • arctanh
    • rad2deg
    • deg2rad
    • rint
    • ceil
    • floor
    • trunc
    • sgn
    • log1p
    • expm1
    • sqr
    • sqrt
  • Dot Product

    • dot.

      • One of the inputs must be sparse, the other sparse or dense.
      • The grad implemented is regular.
      • No C code for perform and no C code for grad.
      • Returns a dense for perform and a dense for grad.

    • structured_dot.

      • The first input is sparse, the second can be sparse or dense.
      • The grad implemented is structured.
      • C code for perform and grad.
      • It returns a sparse output if both inputs are sparse and dense one if one of the inputs is dense.
      • Returns a sparse grad for sparse inputs and dense grad for dense inputs.

    • true_dot.

      • The first input is sparse, the second can be sparse or dense.
      • The grad implemented is regular.
      • No C code for perform and no C code for grad.
      • Returns a Sparse.
      • The gradient returns a Sparse for sparse inputs and by default a dense for dense inputs. The parameter grad_preserves_dense can be set to False to return a sparse grad for dense inputs.

    • sampling_dot.

      • Both inputs must be dense.
      • The grad implemented is structured for p.
      • Sample of the dot and sample of the gradient.
      • C code for perform but not for grad.
      • Returns sparse for perform and grad.

    • usmm.

      • You shouldn’t insert this op yourself!
        • There is an optimization that transform a dot to Usmm when possible.

      • This op is the equivalent of gemm for sparse dot.

      • There is no grad implemented for this op.

      • One of the inputs must be sparse, the other sparse or dense.

      • Returns a dense from perform.

  • Slice Operations
    • sparse_variable[N, N], returns a tensor scalar. There is no grad implemented for this operation.
    • sparse_variable[M:N, O:P], returns a sparse matrix There is no grad implemented for this operation.
    • Sparse variables don’t support [M, N:O] and [M:N, O] as we don’t support sparse vectors and returning a sparse matrix would break the numpy interface. Use [M:M+1, N:O] and [M:N, O:O+1] instead.
    • diag. The grad implemented is regular.
  • Concatenation
    • hstack. The grad implemented is regular.
    • vstack. The grad implemented is regular.
  • Probability

    There is no grad implemented for these operations.

    • Poisson and poisson
    • Binomial and csc_fbinomial, csc_dbinomial csr_fbinomial, csr_dbinomial
    • Multinomial and multinomial
  • Internal Representation

    They all have a regular grad implemented.

    • ensure_sorted_indices.
    • remove0.
    • clean to resort indices and remove zeros
  • To help testing

sparse – Sparse Op

Classes for handling sparse matrices.

To read about different sparse formats, see http://www-users.cs.umn.edu/~saad/software/SPARSKIT/paper.ps

theano.sparse.basic.add(x, y)

Add two matrices, at least one of which is sparse.

This method will provide the right op according to the inputs.

Parameters:
  • x – A matrix variable.
  • y – A matrix variable.
Returns:

x + y
Return type:

matrix

Notes

At least one of x and y must be a sparse matrix.

The grad will be structured only when one of the variable will be a dense matrix.

theano.sparse.basic.as_sparse(x, name=None)

Wrapper around SparseVariable constructor to construct a Variable with a sparse matrix with the same dtype and format.

Parameters:x – A sparse matrix.
Returns:SparseVariable version of x.
Return type:object
theano.sparse.basic.as_sparse_or_tensor_variable(x, name=None)

Same as as_sparse_variable but if we can’t make a sparse variable, we try to make a tensor variable.

Parameters:x – A sparse matrix.
Returns:
Return type:SparseVariable or TensorVariable version of x
theano.sparse.basic.as_sparse_variable(x, name=None)

Wrapper around SparseVariable constructor to construct a Variable with a sparse matrix with the same dtype and format.

Parameters:x – A sparse matrix.
Returns:SparseVariable version of x.
Return type:object
theano.sparse.basic.cast(variable, dtype)

Cast sparse variable to the desired dtype.

Parameters:
  • variable – Sparse matrix.
  • dtype – The dtype wanted.
Returns:
Return type:

Same as x but having dtype as dtype.

Notes

The grad implemented is regular, i.e. not structured.

theano.sparse.basic.clean(x)

Remove explicit zeros from a sparse matrix, and re-sort indices.

CSR column indices are not necessarily sorted. Likewise for CSC row indices. Use clean when sorted indices are required (e.g. when passing data to other libraries) and to ensure there are no zeros in the data.

Parameters:x – A sparse matrix.
Returns:The same as x with indices sorted and zeros removed.
Return type:matrix

Notes

The grad implemented is regular, i.e. not structured.

theano.sparse.basic.col_scale(x, s)

Scale each columns of a sparse matrix by the corresponding element of a dense vector.

Parameters:
  • x – A sparse matrix.
  • s – A dense vector with length equal to the number of columns of x.
Returns:

  • A sparse matrix in the same format as x which each column had been
  • multiply by the corresponding element of s.

Notes

The grad implemented is structured.

theano.sparse.basic.csm_data(csm)

Return the data field of the sparse variable.

theano.sparse.basic.csm_indices(csm)

Return the indices field of the sparse variable.

theano.sparse.basic.csm_indptr(csm)

Return the indptr field of the sparse variable.

theano.sparse.basic.csm_shape(csm)

Return the shape field of the sparse variable.

theano.sparse.basic.dot(x, y)

Operation for efficiently calculating the dot product when one or all operands is sparse. Supported format are CSC and CSR. The output of the operation is dense.

Parameters:
  • x – Sparse or dense matrix variable.
  • y – Sparse or dense matrix variable.
Returns:
Return type:

The dot product x.`y` in a dense format.

Notes

The grad implemented is regular, i.e. not structured.

At least one of x or y must be a sparse matrix.

When the operation has the form dot(csr_matrix, dense) the gradient of this operation can be performed inplace by UsmmCscDense. This leads to significant speed-ups.

theano.sparse.basic.hstack(blocks, format=None, dtype=None)

Stack sparse matrices horizontally (column wise).

This wrap the method hstack from scipy.

Parameters:
  • blocks – List of sparse array of compatible shape.
  • format – String representing the output format. Default is csc.
  • dtype – Output dtype.
Returns:

The concatenation of the sparse array column wise.
Return type:

array

Notes

The number of line of the sparse matrix must agree.

The grad implemented is regular, i.e. not structured.

theano.sparse.basic.mul(x, y)

Multiply elementwise two matrices, at least one of which is sparse.

This method will provide the right op according to the inputs.

Parameters:
  • x – A matrix variable.
  • y – A matrix variable.
Returns:

x + y
Return type:

matrix

Notes

At least one of x and y must be a sparse matrix. The grad is regular, i.e. not structured.

theano.sparse.basic.row_scale(x, s)

Scale each row of a sparse matrix by the corresponding element of a dense vector.

Parameters:
  • x – A sparse matrix.
  • s – A dense vector with length equal to the number of rows of x.
Returns:

A sparse matrix in the same format as x whose each row has been multiplied by the corresponding element of s.
Return type:

matrix

Notes

The grad implemented is structured.

theano.sparse.basic.sp_ones_like(x)

Construct a sparse matrix of ones with the same sparsity pattern.

Parameters:x – Sparse matrix to take the sparsity pattern.
Returns:The same as x with data changed for ones.
Return type:matrix
theano.sparse.basic.sp_sum(x, axis=None, sparse_grad=False)

Calculate the sum of a sparse matrix along the specified axis.

It operates a reduction along the specified axis. When axis is None, it is applied along all axes.

Parameters:
  • x – Sparse matrix.
  • axis – Axis along which the sum is applied. Integer or None.
  • sparse_grad (bool) – True to have a structured grad.
Returns:

The sum of x in a dense format.
Return type:

object

Notes

The grad implementation is controlled with the sparse_grad parameter. True will provide a structured grad and False will provide a regular grad. For both choices, the grad returns a sparse matrix having the same format as x.

This op does not return a sparse matrix, but a dense tensor matrix.

theano.sparse.basic.sp_zeros_like(x)

Construct a sparse matrix of zeros.

Parameters:x – Sparse matrix to take the shape.
Returns:The same as x with zero entries for all element.
Return type:matrix
theano.sparse.basic.structured_dot(x, y)

Structured Dot is like dot, except that only the gradient wrt non-zero elements of the sparse matrix a are calculated and propagated.

The output is presumed to be a dense matrix, and is represented by a TensorType instance.

Parameters:
  • a – A sparse matrix.
  • b – A sparse or dense matrix.
Returns:

The dot product of a and b.
Return type:

matrix

Notes

The grad implemented is structured.

theano.sparse.basic.sub(x, y)

Subtract two matrices, at least one of which is sparse.

This method will provide the right op according to the inputs.

Parameters:
  • x – A matrix variable.
  • y – A matrix variable.
Returns:

x - y
Return type:

matrix

Notes

At least one of x and y must be a sparse matrix.

The grad will be structured only when one of the variable will be a dense matrix.

theano.sparse.basic.true_dot(x, y, grad_preserves_dense=True)

Operation for efficiently calculating the dot product when one or all operands are sparse. Supported formats are CSC and CSR. The output of the operation is sparse.

Parameters:
  • x – Sparse matrix.
  • y – Sparse matrix or 2d tensor variable.
  • grad_preserves_dense (bool) – If True (default), makes the grad of dense inputs dense. Otherwise the grad is always sparse.
Returns:

  • The dot product x.`y` in a sparse format.
  • Notex
  • —–
  • The grad implemented is regular, i.e. not structured.
theano.sparse.basic.verify_grad_sparse(op, pt, structured=False, *args, **kwargs)

Wrapper for theano.test.unittest_tools.py:verify_grad wich converts sparse variables back and forth.

Parameters:
  • op – Op to check.
  • pt – List of inputs to realize the tests.
  • structured – True to tests with a structured grad, False otherwise.
  • args – Other verify_grad parameters if any.
  • kwargs – Other verify_grad keywords if any.
Returns:
Return type:

None
theano.sparse.basic.vstack(blocks, format=None, dtype=None)

Stack sparse matrices vertically (row wise).

This wrap the method vstack from scipy.

Parameters:
  • blocks – List of sparse array of compatible shape.
  • format – String representing the output format. Default is csc.
  • dtype – Output dtype.
Returns:

The concatenation of the sparse array row wise.
Return type:

array

Notes

The number of column of the sparse matrix must agree.

The grad implemented is regular, i.e. not structured.

theano.sparse.tests.test_basic.sparse_random_inputs(format, shape, n=1, out_dtype=None, p=0.5, gap=None, explicit_zero=False, unsorted_indices=False)

Return a tuple containing everything needed to perform a test.

If out_dtype is None, theano.config.floatX is used.

Parameters:
  • format – Sparse format.
  • shape – Shape of data.
  • n – Number of variable.
  • out_dtype – dtype of output.
  • p – Sparsity proportion.
  • gap – Tuple for the range of the random sample. When length is 1, it is assumed to be the exclusive max, when gap = (a, b) it provide a sample from [a, b[. If None is used, it provide [0, 1] for float dtypes and [0, 50[ for integer dtypes.
  • explicit_zero – When True, we add explicit zero in the returned sparse matrix
  • unsorted_indices – when True, we make sure there is unsorted indices in the returned sparse matrix.
Returns:

(variable, data) where both variable and data are list.
Note:

explicit_zero and unsorted_indices was added in Theano 0.6rc4