Tensor objects

A TensorVariable represents a multidimensional container of items of the same type and shape. The number of dimensions and items in a tensor is defined by its shape, a tuple of N non-negative integers or None values that specify the sizes of each dimension (when known). The type of items in the array is determined by its dtype attribute.

Example

A 2-dimensional array of size 2x2 composed of integer elements:

>>> x = at.as_tensor([[2, 3], [3, 4]], "int8")
>>> x.type.ndim
2
>>> x.type.shape
(2, 2)
>>> x.type.dtype
'int8'

A matrix of unknown dimension size:

>>> import aesara.tensor as at
>>> x = at.matrix('x')
>>> x.type.ndim
2
>>> x.type.shape
(None, None)
>>> x.type.dtype
'float32'

Note that Aesara returns None when the shape of the tensor is unknown.

Constructing tensor variables

New tensors can be constructed using the routines defined in Tensor creation, or at a low-level by calling the TensorType type:

>>> import aesara as at
>>> mytype = at.TensorType("int32", shape=(None, 3))
>>> x = mytype()

TensorType(dtype[, shape, name, broadcastable])

Symbolic Type representing numpy.ndarrays.

Tensor attributes

TensorVariable.shape
TensorVariable.size
TensorVariable.name
TensorVariable.astype(dtype)
TensorVariable.ndim	The rank of this tensor.
TensorVariable.dtype	The dtype of this tensor.

Tensor methods

A TensorVariable has many methods which operate on the array and typically return a new TensorVariable.

Shape manipulation

TensorVariable.reshape(shape[, ndim])	Return a reshaped view/copy of this variable.
TensorVariable.dimshuffle(*pattern)	Reorder the dimensions of this variable, optionally inserting broadcasted dimensions.
TensorVariable.transpose(*axes)	Transpose this array.
TensorVariable.T
TensorVariable.swapaxes(axis1, axis2)	See aesara.tensor.basic.swapaxes.
TensorVariable.flatten([ndim])	Returns a view of this variable with ndim dimensions, whose shape for the first ndim-1 dimensions will be the same as self, and shape in the remaining dimension will be expanded to fit in all the data from self.
TensorVariable.ravel()	See flatten.
TensorVariable.squeeze()	Remove broadcastable dimensions from the shape of an array.

Item selection and manipulation

TensorVariable.take(indices[, axis, mode])
TensorVariable.repeat(repeats[, axis])	See aesara.tensor.basic.repeat.
TensorVariable.choose(choices[, mode])	Construct an array from an index array and a set of arrays to choose from.
TensorVariable.sort([axis, kind, order])	See aesara.tensor.sort.sort.
TensorVariable.argsort([axis, kind, order])	See aesara.tensor.sort.argsort.
TensorVariable.compress(a[, axis])	Return selected slices only.
TensorVariable.searchsorted(v[, side, sorter])
TensorVariable.nonzero([return_matrix])	See aesara.tensor.basic.nonzero.
TensorVariable.nonzero_values()	See aesara.tensor.basic.nonzero_values.
TensorVariable.diagonal([offset, axis1, axis2])
TensorVariable.get_scalar_constant_value()

Calculation

TensorVariable.sum([axis, dtype, keepdims, ...])	See aesara.tensor.math.sum.
TensorVariable.prod([axis, dtype, keepdims, ...])	See aesara.tensor.math.prod.
TensorVariable.cumsum([axis])
TensorVariable.cumprod([axis])
TensorVariable.norm(L[, axis, keepdims])
TensorVariable.mean([axis, dtype, keepdims, ...])	See aesara.tensor.math.mean.
TensorVariable.var([axis, ddof, keepdims, ...])	See aesara.tensor.math.var.
TensorVariable.std([axis, ddof, keepdims, ...])	See aesara.tensor.math.std.
TensorVariable.min([axis, keepdims])	See aesara.tensor.math.min.
TensorVariable.argmin([axis, keepdims])	See aesara.tensor.math.argmin.
TensorVariable.max([axis, keepdims])	See aesara.tensor.math.max.
TensorVariable.argmax([axis, keepdims])	See aesara.tensor.math.argmax.
TensorVariable.any([axis, keepdims])
TensorVariable.clip(a_min, a_max)	See aesara.tensor.math.clip.
TensorVariable.conjugate()	See aesara.tensor.math.conj.
TensorVariable.ptp([axis])	See aesara.tensor.math.ptp.
TensorVariable.trunc()
TensorVariable.round([mode])	See aesara.tensor.math.round.
TensorVariable.trace()
TensorVariable.arccos()
TensorVariable.arccosh()
TensorVariable.arcsin()
TensorVariable.arcsinh()
TensorVariable.arctan()
TensorVariable.arctanh()
TensorVariable.ceil()
TensorVariable.cos()
TensorVariable.cosh()
TensorVariable.deg2rad()
TensorVariable.exp()
TensorVariable.exp2()
TensorVariable.expm1()
TensorVariable.floor()
TensorVariable.log()
TensorVariable.log10()
TensorVariable.log1p()
TensorVariable.log2()
TensorVariable.rad2deg()
TensorVariable.sin()
TensorVariable.sinh()
TensorVariable.sqrt()
TensorVariable.tan()
TensorVariable.tanh()

Tensor methods

Comparison

Danger

The Python operators == and != do not work as a comparison operator in the usual sense in Aesara. Use eq() and neq() respectively instead.

TensorVariable.__lt__(other)	Return self<value.
TensorVariable.__le__(other)	Return self<=value.
TensorVariable.__gt__(other)	Return self>value.
TensorVariable.__ge__(other)	Return self>=value.

Unary operations

TensorVariable.__abs__()
TensorVariable.__neg__()
TensorVariable.__invert__()

Arithmetic

TensorVariable.__add__(other)
TensorVariable.__radd__(other)
TensorVariable.__sub__(other)
TensorVariable.__mul__(other)
TensorVariable.__truediv__(other)
TensorVariable.__floordiv__(other)
TensorVariable.__rtruediv__(other)
TensorVariable.__rfloordiv__(other)
TensorVariable.__mod__(other)
TensorVariable.__rmod__(other)
TensorVariable.__divmod__(other)
TensorVariable.__rdivmod__(other)
TensorVariable.__pow__(other)
TensorVariable.__rpow__(other)
TensorVariable.__and__(other)
TensorVariable.__or__(other)
TensorVariable.__ror__(other)
TensorVariable.__xor__(other)
TensorVariable.__rxor__(other)
TensorVariable.__rand__(other)
TensorVariable.__rsub__(other)
TensorVariable.__rmul__(other)
TensorVariable.__div__(other)
TensorVariable.__rdiv__(other)
TensorVariable.__ceil__()
TensorVariable.__floor__()
TensorVariable.__trunc__()

Matrix multiplication:

TensorVariable.__dot__(right)
TensorVariable.__rdot__(left)
TensorVariable.__matmul__(right)
TensorVariable.__rmatmul__(left)

Numerical types

A string indicating the numerical type of the ndarray for which a Variable of this Type represents.

The dtype attribute of a TensorType instance can be any of the following strings.

dtype	domain	bits
'int8'	signed integer	8
'int16'	signed integer	16
'int32'	signed integer	32
'int64'	signed integer	64
'uint8'	unsigned integer	8
'uint16'	unsigned integer	16
'uint32'	unsigned integer	32
'uint64'	unsigned integer	64
'float32'	floating point	32
'float64'	floating point	64
'complex64'	complex	64 (two float32)
'complex128'	complex	128 (two float64)