arrayfire.data module

Functions to create and manipulate arrays.

arrayfire.data.constant(val, d0, d1=None, d2=None, d3=None, dtype=<arrayfire.library.Dtype object>)

Create a multi dimensional array whose elements contain the same value.

Parameters:

val : scalar.

Value of each element of the constant array.

d0 : int.

Length of first dimension.

d1 : optional: int. default: None.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose elements are of value val. - If d1 is None, out is 1D of size (d0,). - If d1 is not None and d2 is None, out is 2D of size (d0, d1). - If d1 and d2 are not None and d3 is None, out is 3D of size (d0, d1, d2). - If d1, d2, d3 are all not None, out is 4D of size (d0, d1, d2, d3).

arrayfire.data.diag(a, num=0, extract=True)

Create a diagonal matrix or Extract the diagonal from a matrix.

Parameters:

a : af.Array.

1 dimensional or 2 dimensional arrayfire array.

num : optional: int. default: 0.

The index of the diagonal. - num == 0 signifies the diagonal. - num > 0 signifies super diagonals. - num < 0 signifies sub diagonals.

extract : optional: bool. default: True.

• If True , diagonal is extracted. a has to be 2D.
• If False, diagonal matrix is created. a has to be 1D.

Returns:

out : af.Array

• if extract is True, out contains the num’th diagonal from a.
• if extract is False, out contains a as the num’th diagonal.

arrayfire.data.flat(a)

Flatten the input array.

Parameters:

a : af.Array.

Multi dimensional array.

Returns:

out : af.Array

• 1 dimensional array containing all the elements from a.

arrayfire.data.flip(a, dim=0)

Flip an array along a dimension.

Parameters:

a : af.Array.

Multi dimensional array.

dim : optional: int. default: 0.

The dimension along which the flip is performed.

Returns:

out : af.Array

The output after flipping a along dim.

Examples

>>> import arrayfire as af
>>> a = af.randu(3, 3)
>>> af.display(a)
[3 3 1 1]
    0.7269     0.3569     0.3341
    0.7104     0.1437     0.0899
    0.5201     0.4563     0.5363

>>> af.display(b)
[3 3 1 1]
    0.5201     0.4563     0.5363
    0.7104     0.1437     0.0899
    0.7269     0.3569     0.3341

>>> af.display(c)
[3 3 1 1]
    0.3341     0.3569     0.7269
    0.0899     0.1437     0.7104
    0.5363     0.4563     0.5201

arrayfire.data.get_seed()

Get the seed for the random number generator.

Returns:

seed: int.

Seed for the random number generator

arrayfire.data.identity(d0, d1, d2=None, d3=None, dtype=<arrayfire.library.Dtype object>)

Create an identity matrix or batch of identity matrices.

Parameters:

d0 : int.

Length of first dimension.

d1 : int.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose first two dimensions form a identity matrix. - If d2 is None, out is 2D of size (d0, d1). - If d2 is not None and d3 is None, out is 3D of size (d0, d1, d2). - If d2, d3 are not None, out is 4D of size (d0, d1, d2, d3).

arrayfire.data.iota(d0, d1=None, d2=None, d3=None, dim=-1, tile_dims=None, dtype=<arrayfire.library.Dtype object>)

Create a multi dimensional array using the number of elements in the array as the range.

Parameters:

val : scalar.

Value of each element of the constant array.

d0 : int.

Length of first dimension.

d1 : optional: int. default: None.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

tile_dims : optional: tuple of ints. default: None.

The number of times the data is tiled.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose elements are along dim fall between [0 - self.elements() - 1].

Examples

>>> import arrayfire as af
>>> import arrayfire as af
>>> a = af.iota(3,3) # tile_dim is not specified, data is not tiled
>>> af.display(a) # the elements range from [0 - 8] (9 elements)
[3 3 1 1]
    0.0000     3.0000     6.0000
    1.0000     4.0000     7.0000
    2.0000     5.0000     8.0000

>>> b = af.iota(3,3,tile_dims(1,2)) # Asking to tile along second dimension.
>>> af.display(b)
[3 6 1 1]
    0.0000     3.0000     6.0000     0.0000     3.0000     6.0000
    1.0000     4.0000     7.0000     1.0000     4.0000     7.0000
    2.0000     5.0000     8.0000     2.0000     5.0000     8.0000

arrayfire.data.join(dim, first, second, third=None, fourth=None)

Join two or more arrayfire arrays along a specified dimension.

Parameters:

dim: int.

Dimension along which the join occurs.

first : af.Array.

Multi dimensional arrayfire array.

second : af.Array.

Multi dimensional arrayfire array.

third : optional: af.Array. default: None.

Multi dimensional arrayfire array.

fourth : optional: af.Array. default: None.

Multi dimensional arrayfire array.

Returns:

out : af.Array

An array containing the input arrays joined along the specified dimension.

Examples

>>> import arrayfire as af
>>> a = af.randu(2, 3)
>>> b = af.randu(2, 3)
>>> c = af.join(0, a, b)
>>> d = af.join(1, a, b)
>>> af.display(a)
[2 3 1 1]
    0.9508     0.2591     0.7928
    0.5367     0.8359     0.8719

>>> af.display(b)
[2 3 1 1]
    0.3266     0.6009     0.2442
    0.6275     0.0495     0.6591

>>> af.display(c)
[4 3 1 1]
    0.9508     0.2591     0.7928
    0.5367     0.8359     0.8719
    0.3266     0.6009     0.2442
    0.6275     0.0495     0.6591

>>> af.display(d)
[2 6 1 1]
    0.9508     0.2591     0.7928     0.3266     0.6009     0.2442
    0.5367     0.8359     0.8719     0.6275     0.0495     0.6591

arrayfire.data.lower(a, is_unit_diag=False)

Extract the lower triangular matrix from the input.

Parameters:

a : af.Array.

Multi dimensional array.

is_unit_diag: optional: bool. default: False.

Flag specifying if the diagonal elements are 1.

Returns:

out : af.Array

An array containing the lower triangular elements from a.

arrayfire.data.moddims(a, d0, d1=1, d2=1, d3=1)

Modify the shape of the array without changing the data layout.

Parameters:

a : af.Array.

Multi dimensional array.

d0: int.

The first dimension of output.

d1: optional: int. default: 1.

The second dimension of output.

d2: optional: int. default: 1.

The third dimension of output.

d3: optional: int. default: 1.

The fourth dimension of output.

Returns:

out : af.Array

• An containing the same data as a with the specified shape.
• The number of elements in a must match d0 x d1 x d2 x d3.

arrayfire.data.randn(d0, d1=None, d2=None, d3=None, dtype=<arrayfire.library.Dtype object>)

Create a multi dimensional array containing values from a normal distribution.

Parameters:

d0 : int.

Length of first dimension.

d1 : optional: int. default: None.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose elements are sampled from a normal distribution with mean 0 and sigma of 1. - If d1 is None, out is 1D of size (d0,). - If d1 is not None and d2 is None, out is 2D of size (d0, d1). - If d1 and d2 are not None and d3 is None, out is 3D of size (d0, d1, d2). - If d1, d2, d3 are all not None, out is 4D of size (d0, d1, d2, d3).

arrayfire.data.randu(d0, d1=None, d2=None, d3=None, dtype=<arrayfire.library.Dtype object>)

Create a multi dimensional array containing values from a uniform distribution.

Parameters:

d0 : int.

Length of first dimension.

d1 : optional: int. default: None.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose elements are sampled uniformly between [0, 1]. - If d1 is None, out is 1D of size (d0,). - If d1 is not None and d2 is None, out is 2D of size (d0, d1). - If d1 and d2 are not None and d3 is None, out is 3D of size (d0, d1, d2). - If d1, d2, d3 are all not None, out is 4D of size (d0, d1, d2, d3).

arrayfire.data.range(d0, d1=None, d2=None, d3=None, dim=0, dtype=<arrayfire.library.Dtype object>)

Create a multi dimensional array using length of a dimension as range.

Parameters:

val : scalar.

Value of each element of the constant array.

d0 : int.

Length of first dimension.

d1 : optional: int. default: None.

Length of second dimension.

d2 : optional: int. default: None.

Length of third dimension.

d3 : optional: int. default: None.

Length of fourth dimension.

dim : optional: int. default: 0.

The dimension along which the range is calculated.

dtype : optional: af.Dtype. default: af.Dtype.f32.

Data type of the array.

Returns:

out : af.Array

Multi dimensional array whose elements are along dim fall between [0 - self.dims[dim]-1] - If d1 is None, out is 1D of size (d0,). - If d1 is not None and d2 is None, out is 2D of size (d0, d1). - If d1 and d2 are not None and d3 is None, out is 3D of size (d0, d1, d2). - If d1, d2, d3 are all not None, out is 4D of size (d0, d1, d2, d3).

Examples

>>> import arrayfire as af
>>> a = af.range(3, 2) # dim is not specified, range is along first dimension.
>>> af.display(a) # The data ranges from [0 - 2] (3 elements along first dimension)
[3 2 1 1]
    0.0000     0.0000
    1.0000     1.0000
    2.0000     2.0000

>>> a = af.range(3, 2, dim=1) # dim is 1, range is along second dimension.
>>> af.display(a) # The data ranges from [0 - 1] (2 elements along second dimension)
[3 2 1 1]
    0.0000     1.0000
    0.0000     1.0000
    0.0000     1.0000

arrayfire.data.reorder(a, d0=1, d1=0, d2=2, d3=3)

Reorder the dimensions of the input.

Parameters:

a : af.Array.

Multi dimensional array.

d0: optional: int. default: 1.

The location of the first dimension in the output.

d1: optional: int. default: 0.

The location of the second dimension in the output.

d2: optional: int. default: 2.

The location of the third dimension in the output.

d3: optional: int. default: 3.

The location of the fourth dimension in the output.

Returns:

out : af.Array

• An array containing the input aftern reordering its dimensions.

Examples

>>> import arrayfire as af
>>> a = af.randu(5, 5, 3)
>>> af.display(a)
[5 5 3 1]
    0.4107     0.0081     0.6600     0.1046     0.8395
    0.8224     0.3775     0.0764     0.8827     0.1933
    0.9518     0.3027     0.0901     0.1647     0.7270
    0.1794     0.6456     0.5933     0.8060     0.0322
    0.4198     0.5591     0.1098     0.5938     0.0012

0.8703 0.9250 0.4387 0.6530 0.4224 0.5259 0.3063 0.3784 0.5476 0.5293 0.1443 0.9313 0.4002 0.8577 0.0212 0.3253 0.8684 0.4390 0.8370 0.1103 0.5081 0.6592 0.4718 0.0618 0.4420

0.8355 0.6767 0.1033 0.9426 0.9276 0.4878 0.6742 0.2119 0.4817 0.8662 0.2055 0.4523 0.5955 0.9097 0.3578 0.1794 0.1236 0.3745 0.6821 0.6263 0.5606 0.7924 0.9165 0.6056 0.9747

>>> b = af.reorder(a, 2, 0, 1)
>>> af.display(b)
[3 5 5 1]
    0.4107     0.8224     0.9518     0.1794     0.4198
    0.8703     0.5259     0.1443     0.3253     0.5081
    0.8355     0.4878     0.2055     0.1794     0.5606

0.0081 0.3775 0.3027 0.6456 0.5591 0.9250 0.3063 0.9313 0.8684 0.6592 0.6767 0.6742 0.4523 0.1236 0.7924

0.6600 0.0764 0.0901 0.5933 0.1098 0.4387 0.3784 0.4002 0.4390 0.4718 0.1033 0.2119 0.5955 0.3745 0.9165

0.1046 0.8827 0.1647 0.8060 0.5938 0.6530 0.5476 0.8577 0.8370 0.0618 0.9426 0.4817 0.9097 0.6821 0.6056

0.8395 0.1933 0.7270 0.0322 0.0012 0.4224 0.5293 0.0212 0.1103 0.4420 0.9276 0.8662 0.3578 0.6263 0.9747

arrayfire.data.replace(lhs, cond, rhs)

Select elements from one of two arrays based on condition.

Parameters:

lhs : af.Array or scalar

numerical array whose elements are replaced with rhs when conditional element is False

cond : af.Array

Conditional array

rhs : af.Array or scalar

numerical array whose elements are picked when conditional element is False

Examples

>>> import arrayfire as af
>>> a = af.randu(3,3)
>>> af.display(a)
[3 3 1 1]
    0.4107     0.1794     0.3775
    0.8224     0.4198     0.3027
    0.9518     0.0081     0.6456

>>> cond = (a >= 0.25) & (a <= 0.75)
>>> af.display(cond)
[3 3 1 1]
         1          0          1
         0          1          1
         0          0          1

>>> af.replace(a, cond, 0.3333)
>>> af.display(a)
[3 3 1 1]
    0.3333     0.1794     0.3333
    0.8224     0.3333     0.3333
    0.9518     0.0081     0.3333

arrayfire.data.select(cond, lhs, rhs)

Select elements from one of two arrays based on condition.

Parameters:

cond : af.Array

Conditional array

lhs : af.Array or scalar

numerical array whose elements are picked when conditional element is True

rhs : af.Array or scalar

numerical array whose elements are picked when conditional element is False

Returns:

out: af.Array

An array containing elements from lhs when cond is True and rhs when False.

Examples

>>> import arrayfire as af
>>> a = af.randu(3,3)
>>> b = af.randu(3,3)
>>> cond = a > b
>>> res = af.select(cond, a, b)

>>> af.display(a)
[3 3 1 1]
    0.4107     0.1794     0.3775
    0.8224     0.4198     0.3027
    0.9518     0.0081     0.6456

>>> af.display(b)
[3 3 1 1]
    0.7269     0.3569     0.3341
    0.7104     0.1437     0.0899
    0.5201     0.4563     0.5363

>>> af.display(res)
[3 3 1 1]
    0.7269     0.3569     0.3775
    0.8224     0.4198     0.3027
    0.9518     0.4563     0.6456

arrayfire.data.set_seed(seed=0)

Set the seed for the random number generator.

Parameters:

seed: int.

Seed for the random number generator

arrayfire.data.shift(a, d0, d1=0, d2=0, d3=0)

Shift the input along each dimension.

Parameters:

a : af.Array.

Multi dimensional array.

d0: int.

The amount of shift along first dimension.

d1: optional: int. default: 0.

The amount of shift along second dimension.

d2: optional: int. default: 0.

The amount of shift along third dimension.

d3: optional: int. default: 0.

The amount of shift along fourth dimension.

Returns:

out : af.Array

• An array the same shape as a after shifting it by the specified amounts.

Examples

>>> import arrayfire as af
>>> a = af.randu(3, 3)
>>> b = af.shift(a, 2)
>>> c = af.shift(a, 1, -1)
>>> af.display(a)
[3 3 1 1]
    0.7269     0.3569     0.3341
    0.7104     0.1437     0.0899
    0.5201     0.4563     0.5363

>>> af.display(b)
[3 3 1 1]
    0.7104     0.1437     0.0899
    0.5201     0.4563     0.5363
    0.7269     0.3569     0.3341

>>> af.display(c)
[3 3 1 1]
    0.4563     0.5363     0.5201
    0.3569     0.3341     0.7269
    0.1437     0.0899     0.7104

arrayfire.data.tile(a, d0, d1=1, d2=1, d3=1)

Tile an array along specified dimensions.

Parameters:

a : af.Array.

Multi dimensional array.

d0: int.

The number of times a has to be tiled along first dimension.

d1: optional: int. default: 1.

The number of times a has to be tiled along second dimension.

d2: optional: int. default: 1.

The number of times a has to be tiled along third dimension.

d3: optional: int. default: 1.

The number of times a has to be tiled along fourth dimension.

Returns:

out : af.Array

An array containing the input after tiling the the specified number of times.

Examples

>>> import arrayfire as af
>>> a = af.randu(2, 3)
>>> b = af.tile(a, 2)
>>> c = af.tile(a, 1, 2)
>>> d = af.tile(a, 2, 2)
>>> af.display(a)
[2 3 1 1]
    0.9508     0.2591     0.7928
    0.5367     0.8359     0.8719

>>> af.display(b)
[4 3 1 1]
    0.4107     0.9518     0.4198
    0.8224     0.1794     0.0081
    0.4107     0.9518     0.4198
    0.8224     0.1794     0.0081

>>> af.display(c)
[2 6 1 1]
    0.4107     0.9518     0.4198     0.4107     0.9518     0.4198
    0.8224     0.1794     0.0081     0.8224     0.1794     0.0081

>>> af.display(d)
[4 6 1 1]
    0.4107     0.9518     0.4198     0.4107     0.9518     0.4198
    0.8224     0.1794     0.0081     0.8224     0.1794     0.0081
    0.4107     0.9518     0.4198     0.4107     0.9518     0.4198
    0.8224     0.1794     0.0081     0.8224     0.1794     0.0081

arrayfire.data.upper(a, is_unit_diag=False)

Extract the upper triangular matrix from the input.

Parameters:

a : af.Array.

Multi dimensional array.

is_unit_diag: optional: bool. default: False.

Flag specifying if the diagonal elements are 1.

Returns:

out : af.Array

An array containing the upper triangular elements from a.