`pyeddl.tensor` — Tensor

pyeddl.tensor.DEV_CPU = 0: A constant representing the CPU device

pyeddl.tensor.DEV_GPU = 1000: A constant representing the GPU device

class pyeddl.tensor.Tensor(shape, dev=0)

Create an uninitialized tensor.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

=== Creation Methods ===

static fromarray(array, dev=0)

Create a tensor from a NumPy array.

The tensor will be initialized with values, shape, etc. from the array.

The data type is automatically converted to float32, with one exception: due to the way pybind11 overloads are picked, if the input array is 1D, it must be of type float32. See http://github.com/deephealthproject/pyeddl/issues/10.

Parameters

array – NumPy array
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static zeros(shape, dev=0)

Create a tensor of the specified shape and fill it with zeros.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static ones(shape, dev=0)

Create a tensor of the specified shape and fill it with ones.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static full(shape, value, dev=0)

Create a tensor of the specified shape and fill it with value.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU
value – value to fill the tensor with

Returns

Tensor

static arange(start, end, step, dev=0)

Create a 1D tensor with evenly spaced values within a given interval.

Parameters

start – start of interval
end – end of interval
step – spacing between values
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static range(start, end, step, dev=0)

Create a 1D tensor with evenly spaced values within a given interval.

Parameters

start – start of interval
end – end of interval
step – spacing between values
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static linspace(start, end, steps, dev=0)

Create a 1D tensor with evenly spaced values within a given interval.

Parameters

start – starting value
end – end value
steps – number of samples to generate
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static logspace(start, end, steps, base, dev=0)

Create a 1D tensor with evenly spaced values on a log scale.

Parameters

start – base ** start is the starting value
end – base ** end is the final value of the sequence
steps – number of samples to generate
base – the base of the log space
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static eye(size, offset=0, dev=0)

Create a size x size tensor with ones on the diagonal and zeros elsewhere.

Parameters

size – size of the (square) tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static randu(shape, dev=0)

Create a tensor with uniformly distributed random values.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

static randn(shape, dev=0)

Create a tensor with normally distributed random values.

Parameters

shape – shape of the tensor to create
dev – device to use: DEV_CPU or DEV_GPU

Returns

Tensor

=== Data Copying Methods ===

toCPU()

Clone the tensor to the CPU.

Returns: None

toGPU()

Clone the tensor to the GPU.

Returns: None

clone()

Return a clone of the tensor (same device).

Returns: Tensor

select(indices)

Perform NumPy-like slicing on the tensor.

Parameters: indices – list of strings representing the indices to be selected. These indices must follow a NumPy-like syntax. For instance: ["1:3", "2"].
Returns: Tensor

set_select(indices, A)

Sets tensor elements to the values of tensor A using the selected indices.

Parameters: indices – list of strings representing the indices to be set. These indices must follow a NumPy-like syntax. For instance: ["1:3", "2"].
Returns: None

static copy(A, B)

Copy data from tensor A to tensor B.

Parameters

A – a tensor
B – a tensor

Returns

None

=== Serialization Methods ===

static load(fname, format='')

Load a tensor from a file.

Parameters

fname – name of the file to load the tensor from
format – file format (e.g., “bin”, “jpg”)

Returns

Tensor

save(fname, format='')

Save the tensor to a file.

Parameters

fname – name of the file to save the tensor to
format – file format (e.g., “bin”, “jpg”)

Returns

None

=== Math Methods ===

abs_()

Compute the element-wise absolute value of the tensor.

Modifies the tensor.

Returns: None

abs()

Compute the element-wise absolute value of the tensor.

Returns a new tensor.

Returns: Tensor

acos_()

Compute the element-wise inverse cosine of the tensor.

Modifies the tensor.

Returns: None

acos()

Compute the element-wise inverse cosine of the tensor.

Returns a new tensor.

Returns: Tensor

add_(other)

Adds other to the tensor.

Modifies the tensor.

Parameters: other – a tensor or scalar
Returns: None

add(other)

Adds other to the tensor.

Returns a new tensor.

Parameters: other – a tensor or scalar
Returns: Tensor

asin_()

Compute the element-wise inverse sine of the tensor.

Modifies the tensor.

Returns: None

asin()

Compute the element-wise inverse sine of the tensor.

Returns a new tensor.

Returns: Tensor

atan_()

Compute the element-wise inverse tangent of the tensor.

Modifies the tensor.

Returns: None

atan()

Compute the element-wise inverse tangent of the tensor.

Returns a new tensor.

Returns: Tensor

ceil_()

Compute the element-wise ceiling (smallest integer i such that i >= x) of the tensor.

Modifies the tensor.

Returns: None

ceil()

Compute the element-wise ceiling (smallest integer i such that i >= x) of the tensor.

Returns a new tensor.

Returns: Tensor

clamp_(min, max)

Limit the tensor’s values between min and max.

Modifies the tensor.

Parameters

min – minimum value
max – maximum value

Returns

None

clamp(min, max)

Limit the tensor’s values between min and max.

Returns a new tensor.

Parameters

min – minimum value
max – maximum value

Returns

Tensor

clampmax_(max)

Limit the tensor’s values to a maximum value.

Modifies the tensor.

Parameters: max – maximum value
Returns: None

clampmax(max)

Limit the tensor’s values to a maximum value.

Returns a new tensor.

Parameters: max – maximum value
Returns: Tensor

clampmin_(min)

Limit the tensor’s values to a minimum value.

Modifies the tensor.

Parameters: min – minimum value
Returns: None

clampmin(min)

Limit the tensor’s values to a minimum value.

Returns a new tensor.

Parameters: min – minimum value
Returns: Tensor

cos_()

Compute the element-wise cosine of the tensor.

Modifies the tensor.

Returns: None

cos()

Compute the element-wise cosine of the tensor.

Returns a new tensor.

Returns: Tensor

cosh_()

Compute the element-wise hyperbolic cosine of the tensor.

Modifies the tensor.

Returns: None

cosh()

Compute the element-wise hyperbolic cosine of the tensor.

Returns a new tensor.

Returns: Tensor

diag_(k=0)

Select diagonal elements.

Modifies the tensor (elements other than those in the selected diagonal are set to zero).

Parameters: k – offset (0 for the main diagonal, positive for the nth diagonal above the main one, negative for the nth diagonal below the main one)
Returns: None

diag(k=0)

Select diagonal elements.

Returns a new tensor which is the same as this one, except that elements other than those in the selected diagonal are set to zero.

Parameters: k – offset (0 for the main diagonal, positive for the nth diagonal above the main one, negative for the nth diagonal below the main one)
Returns: Tensor

div_(other)

Divides the tensor by other.

Modifies the tensor.

Parameters: other – a tensor or scalar
Returns: None

div(other)

Divides the tensor by other.

Returns a new tensor.

Parameters: other – a tensor or scalar
Returns: Tensor

exp_()

Compute the element-wise exponential of the tensor.

Modifies the tensor.

Returns: None

exp()

Compute the element-wise exponential of the tensor.

Returns a new tensor.

Returns: Tensor

floor_()

Compute the element-wise floor (largest integer i such that i <= x) of the tensor.

Modifies the tensor.

Returns: None

floor()

Compute the element-wise floor (largest integer i such that i <= x) of the tensor.

Returns a new tensor.

Returns: Tensor

log_()

Compute the element-wise natural logarithm of the tensor.

Modifies the tensor.

Returns: None

log()

Compute the element-wise natural logarithm of the tensor.

Returns a new tensor.

Returns: Tensor

log2_()

Compute the element-wise base-2 logarithm of the tensor.

Modifies the tensor.

Returns: None

log2()

Compute the element-wise base-2 logarithm of the tensor.

Returns a new tensor.

Returns: Tensor

log10_()

Compute the element-wise base-10 logarithm of the tensor.

Modifies the tensor.

Returns: None

log10()

Compute the element-wise base-10 logarithm of the tensor.

Returns a new tensor.

Returns: Tensor

logn_(n)

Compute the element-wise base-n logarithm of the tensor.

Modifies the tensor.

Parameters: n – logarithm base
Returns: None

logn(n)

Compute the element-wise base-n logarithm of the tensor.

Returns a new tensor.

Parameters: n – logarithm base
Returns: Tensor

max()

Return the maximum value of the tensor.

Returns: scalar

min()

Return the minimum value of the tensor.

Returns: scalar

mod_(v)

Compute the element-wise reminder of the division of the tensor by v .

Modifies the tensor.

Parameters: v – a scalar
Returns: None

mod(v)

Compute the element-wise reminder of the division of the tensor by v .

Returns a new tensor.

Parameters: v – a scalar
Returns: Tensor

mult_(other)

Multiplies the tensor by other, element-wise.

Modifies the tensor.

Parameters: other – a tensor or scalar
Returns: None

mult(other)

Multiplies the tensor by other, element-wise.

Returns a new tensor.

Parameters: other – a tensor or scalar
Returns: Tensor

static mult2D(A, B)

Computes the matrix product of A and B.

Parameters

A – a tensor
B – a tensor

Returns

Tensor

neg_()

Negate all elements in the tensor.

Modifies the tensor.

Returns: None

neg()

Negate all elements in the tensor.

Returns a new tensor.

Returns: Tensor

normalize_(min, max)

Normalize tensor values to the [min, max] range.

v' = r * (v - A_min) + min; r = (max - min) / (A_max - A_min)

Modifies the tensor.

Returns: None

normalize(min, max)

Normalize tensor values to the [min, max] range.

v' = r * (v - A_min) + min; r = (max - min) / (A_max - A_min)

Returns a new tensor.

Returns: Tensor

reciprocal_()

Compute the element-wise reciprocal of the tensor.

Modifies the tensor.

Returns: None

reciprocal()

Compute the element-wise reciprocal of the tensor.

Returns a new tensor.

Returns: Tensor

round_()

Round tensor values to the nearest integer.

Modifies the tensor.

Returns: None

round()

Round tensor values to the nearest integer.

Returns a new tensor.

Returns: Tensor

rsqrt_()

Compute the element-wise reciprocal of the square root of the tensor.

Modifies the tensor.

Returns: None

rsqrt()

Compute the element-wise reciprocal of the square root of the input tensor.

Returns a new tensor.

Returns: Tensor

sigmoid_()

Compute the element-wise sigmoid of the tensor.

Modifies the tensor.

Parameters: A – a tensor
Returns: None

sigmoid()

Compute the element-wise sigmoid of the tensor.

Returns a new tensor.

Returns: Tensor

sign_()

Compute the element-wise sign (-1 if x < 0, 0 if x == 0, 1 if x > 0) of the tensor.

Modifies the tensor.

Returns: None

sign()

Compute the element-wise sign (-1 if x < 0, 0 if x == 0, 1 if x > 0) of the tensor.

Returns a new tensor.

Returns: Tensor

sin_()

Compute the element-wise sine of the tensor.

Modifies the tensor.

Returns: None

sin()

Compute the element-wise sine of the tensor.

Returns a new tensor.

Returns: Tensor

sinh_()

Compute the element-wise hyperbolic sine of the tensor.

Modifies the tensor.

Returns: None

sinh()

Compute the element-wise hyperbolic sine of the tensor.

Returns a new tensor.

Returns: Tensor

sqr_()

Compute the element-wise square of the tensor.

Modifies the tensor.

Returns: None

sqr()

Compute the element-wise square of the tensor.

Returns a new tensor.

Returns: Tensor

sqrt_()

Compute the element-wise square root of the tensor.

Modifies the tensor.

Returns: None

sqrt()

Compute the element-wise square root of the tensor.

Returns a new tensor.

Returns: Tensor

sub_(other)

Subtracts other from the tensor.

Modifies the tensor.

Parameters: other – a tensor or scalar
Returns: None

sub(other)

Subtracts other from the tensor.

Returns a new tensor.

Parameters: other – a tensor or scalar
Returns: Tensor

tan_()

Compute the element-wise tangent of the tensor.

Modifies the tensor.

Returns: None

tan()

Compute the element-wise tangent of the tensor.

Returns a new tensor.

Returns: Tensor

tanh_()

Compute the element-wise hyperbolic tangent of the tensor.

Modifies the tensor.

Returns: None

tanh()

Compute the element-wise hyperbolic tangent of the tensor.

Returns a new tensor.

Returns: Tensor

trace(k=0)

Sum diagonal elements.

Parameters: k – offset (0 for the main diagonal, positive for the nth diagonal above the main one, negative for the nth diagonal below the main one)
Returns: float

trunc_()

Truncate (discard the fractional part) the tensor, element-wise.

Modifies the tensor.

Returns: None

trunc()

Truncate (discard the fractional part) the tensor, element-wise.

Returns a new tensor.

Returns: Tensor

=== Indexing, Slicing, Joining, Mutating ===

static concat(A, axis=0, output=None)

Join a sequence of tensors along the specified axis.

Parameters

A – list of tensors with the same shape
axis – axis along which to join
output – output tensor

Returns

Tensor

static stack(A, axis=0, output=None)

Repeat tensor elements along the specified dimension.

Parameters

A – input tensor
axis – axis along which to repeat the values
output – output tensor

Returns

Tensor

static repeat(A, repeats, axis=0, output=None, derivative=False)

Repeat the elements of a tensor along the specified dimension.

Parameters

A – input tensor
repeats – number of repetitions
axis – axis along which to repeat values
output – output tensor
derivative – apply derivative for output = repeat(A)

Returns

Tensor

static tile(A, repeats)

Construct an array by repeating A for the given number of times.

Parameters

A – input tensor
repeats – number of repetitions of A

Returns

Tensor

static broadcast(A, B, output=None)

Return a new tensor A to be broadcasted into B

Parameters

A – input tensor
B – input tensor
output – output tensor

Returns

Tensor

=== Transformations ===

scale(new_shape, mode=WrappingMode.Constant, cval=0.0, coordinate_transformation_mode=TransformationMode.Asymmetric, keep_size=False)

Scale the tensor. The array is scaled using spline interpolation.

Parameters

new_shape – list with the target size
mode – a WrappingMode
cval – Value to fill past edges of input if mode is WrappingMode.Constant
coordinate_transformation_mode – a TransformationMode
keep_size – whether to keep the original size

Returns

a new tensor

=== Other Methods ===

fill_(v)

Fill the tensor with the specified value.

Parameters: v – a scalar value
Returns: None

reshape_(new_shape)

Change the tensor’s shape.

Parameters: new_shape – the new shape (list of integers)
Returns: None

getdata()

Get the tensor’s data as a NumPy array.

Returns: a NumPy array

print()

Print the tensor’s values.

Returns: None

info()

Print info on the tensor (shape, strides, …).

Returns: None

getShape()

Return the tensor’s shape.

Returns: the tensor’s shape (a list of integers)

static onehot(in: pyeddl._core.Tensor, vocs: int) → pyeddl._core.Tensor

static hardware_supported()

Get a list of hardware accelerators for which EDDL has been compiled.

Returns: list of supported accelerators

static is_hardware_supported(hardware)

Check if a specific hardware is supported.

Returns: True if the hardware is supported

pyeddl.tensor — Tensor

`pyeddl.tensor` — Tensor