What is a sparse matrix in theano?

Recipe Objective - What is a sparse matrix in theano?

The sparse module is similar to the tensor module in terms of functionality. Because sparse matrices do not hold data in a contiguous array, the difference is hidden under the surface.

Theano has two compressed sparse formats: csc and csr, which are based on columns and rows, respectively. They have the same data, indices, indptr, and form properties.
Attributes:
1. The data attribute is a one-dimensional ndarray that contains all of the sparse matrix's non-zero members.
2. The data position in the sparse matrix is stored using the indices and indptr properties.
3. The shape attribute is the same as the dense matrix's shape attribute.

Access YOLO OCR Character Recognition Project with Source Code

For more related projects -

/projects/data-science-projects/deep-learning-projects
/projects/data-science-projects/tensorflow-projects

Compressed Sparse Column (csc) Matrix

In the Compressed Sparse Column (csc) format, indices refer to indexes within the matrix's column vectors, whereas indptr specifies the position of the column in the data and indices attributes.

# Importing libraries
import numpy as np
import scipy.sparse as sci_sp

# Creating 'data' array
data = np.asarray([10, 20, 30])

# Creating 'indices' array
indices = np.asarray([0, 1, 2])

# Creating 'indptr' array
indptr = np.asarray([0, 3, 2, 1])

# Creating csc matrix
matrix = sci_sp.csc_matrix((data, indices, indptr), shape=(3, 3))
matrix.toarray()


Output -
array([[10,  0,  0],
       [20,  0,  0],
       [30,  0,  0]])

Compressed Sparse Row (csr) Matrix

In the Compressed Sparse Row (csr) format, indices refer to indexes within the matrix's row vectors, whereas indptr specifies the position of the row in the data and indices attributes.

# Importing libraries
import numpy as np
import scipy.sparse as sci_sp

# Creating 'data' array
data = np.asarray([10, 20, 30])

# Creating 'indices' array
indices = np.asarray([0, 1, 2])

# Creating 'indptr' array
indptr = np.asarray([0, 3, 2, 1])

# Creating csr matrix
matrix = sci_sp.csr_matrix((data, indices, indptr), shape=(3, 3))
matrix.toarray()


Output -
array([[10, 20, 30],
       [ 0,  0,  0],
       [ 0,  0,  0]])

In this way, we can create a sparse matrix in theano.