pytorch_backend

class pyhf.tensor.pytorch_backend.pytorch_backend(**kwargs)

Bases: object

PyTorch backend for pyhf

__init__(**kwargs)

Attributes

name

precision

dtypemap

default_do_grad

array_subtype

The array content type for pytorch

array_type

The array type for pytorch

Methods

_setup()

Run any global setups for the pytorch lib.

abs(tensor)

astensor(tensor_in, dtype='float')

Convert to a PyTorch Tensor.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> tensor = pyhf.tensorlib.astensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
>>> tensor
tensor([[1., 2., 3.],
        [4., 5., 6.]])
>>> type(tensor)
<class 'torch.Tensor'>

Parameters:

tensor_in (Number or Tensor) – Tensor object

Returns:

A multi-dimensional matrix containing elements of a single data type.

Return type:

torch.Tensor

boolean_mask(tensor, mask)

clip(tensor_in, min_value, max_value)

Clips (limits) the tensor values to be within a specified min and max.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([-2, -1, 0, 1, 2])
>>> pyhf.tensorlib.clip(a, -1, 1)
tensor([-1., -1.,  0.,  1.,  1.])

Parameters:

• tensor_in (tensor) – The input tensor object

• min_value (scalar or tensor or None) – The minimum value to be clipped to

• max_value (scalar or tensor or None) – The maximum value to be clipped to

Returns:

A clipped tensor

Return type:

PyTorch tensor

concatenate(sequence, axis=0)

Join a sequence of arrays along an existing axis.

Parameters:

• sequence – sequence of tensors

• axis – dimension along which to concatenate

Returns:

the concatenated tensor

Return type:

output

conditional(predicate, true_callable, false_callable)

Runs a callable conditional on the boolean value of the evaluation of a predicate

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> tensorlib = pyhf.tensorlib
>>> a = tensorlib.astensor([4])
>>> b = tensorlib.astensor([5])
>>> tensorlib.conditional((a < b)[0], lambda: a + b, lambda: a - b)
tensor([9.])

Parameters:

• predicate (scalar) – The logical condition that determines which callable to evaluate

• true_callable (callable) – The callable that is evaluated when the predicate evaluates to true

• false_callable (callable) – The callable that is evaluated when the predicate evaluates to false

Returns:

The output of the callable that was evaluated

Return type:

PyTorch Tensor

divide(tensor_in_1, tensor_in_2)

einsum(subscripts, *operands)

This function provides a way of computing multilinear expressions (i.e. sums of products) using the Einstein summation convention.

Parameters:

• subscripts – str, specifies the subscripts for summation

• operands – list of array_like, these are the tensors for the operation

Returns:

the calculation based on the Einstein summation convention

Return type:

tensor

erf(tensor_in)

The error function of complex argument.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([-2., -1., 0., 1., 2.])
>>> pyhf.tensorlib.erf(a)
tensor([-0.9953, -0.8427,  0.0000,  0.8427,  0.9953])

Parameters:

tensor_in (tensor) – The input tensor object

Returns:

The values of the error function at the given points.

Return type:

PyTorch Tensor

erfinv(tensor_in)

The inverse of the error function of complex argument.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([-2., -1., 0., 1., 2.])
>>> pyhf.tensorlib.erfinv(pyhf.tensorlib.erf(a))
tensor([-2.0000, -1.0000,  0.0000,  1.0000,  2.0000])

Parameters:

tensor_in (tensor) – The input tensor object

Returns:

The values of the inverse of the error function at the given points.

Return type:

PyTorch Tensor

exp(tensor_in)

gather(tensor, indices)

isfinite(tensor)

log(tensor_in)

normal(x, mu, sigma)

The probability density function of the Normal distribution evaluated at x given parameters of mean of mu and standard deviation of sigma.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> pyhf.tensorlib.normal(0.5, 0., 1.)
tensor(0.3521)
>>> values = pyhf.tensorlib.astensor([0.5, 2.0])
>>> means = pyhf.tensorlib.astensor([0., 2.3])
>>> sigmas = pyhf.tensorlib.astensor([1., 0.8])
>>> pyhf.tensorlib.normal(values, means, sigmas)
tensor([0.3521, 0.4648])

Parameters:

• x (tensor or float) – The value at which to evaluate the Normal distribution p.d.f.

• mu (tensor or float) – The mean of the Normal distribution

• sigma (tensor or float) – The standard deviation of the Normal distribution

Returns:

Value of Normal(x|mu, sigma)

Return type:

PyTorch FloatTensor

normal_cdf(x, mu=0.0, sigma=1.0)

The cumulative distribution function for the Normal distribution

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> pyhf.tensorlib.normal_cdf(0.8)
tensor(0.7881)
>>> values = pyhf.tensorlib.astensor([0.8, 2.0])
>>> pyhf.tensorlib.normal_cdf(values)
tensor([0.7881, 0.9772])

Parameters:

• x (tensor or float) – The observed value of the random variable to evaluate the CDF for

• mu (tensor or float) – The mean of the Normal distribution

• sigma (tensor or float) – The standard deviation of the Normal distribution

Returns:

The CDF

Return type:

PyTorch FloatTensor

normal_dist(mu, sigma)

The Normal distribution with mean mu and standard deviation sigma.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> means = pyhf.tensorlib.astensor([5, 8])
>>> stds = pyhf.tensorlib.astensor([1, 0.5])
>>> values = pyhf.tensorlib.astensor([4, 9])
>>> normals = pyhf.tensorlib.normal_dist(means, stds)
>>> normals.log_prob(values)
tensor([-1.4189, -2.2258])

Parameters:

• mu (tensor or float) – The mean of the Normal distribution

• sigma (tensor or float) – The standard deviation of the Normal distribution

Returns:

The Normal distribution class

Return type:

PyTorch Normal distribution

normal_logpdf(x, mu, sigma)

ones(shape, dtype='float')

outer(tensor_in_1, tensor_in_2)

Outer product of the input tensors.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([1.0, 2.0, 3.0])
>>> b = pyhf.tensorlib.astensor([1.0, 2.0, 3.0, 4.0])
>>> pyhf.tensorlib.outer(a, b)
tensor([[ 1.,  2.,  3.,  4.],
        [ 2.,  4.,  6.,  8.],
        [ 3.,  6.,  9., 12.]])

Parameters:

• tensor_in_1 (tensor) – 1-D input tensor.

• tensor_in_2 (tensor) – 1-D input tensor.

Returns:

The outer product.

Return type:

PyTorch tensor

percentile(tensor_in, q, axis=None, interpolation='linear')

Compute the \(q\)-th percentile of the tensor along the specified axis.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([[10, 7, 4], [3, 2, 1]])
>>> pyhf.tensorlib.percentile(a, 50)
tensor(3.5000)
>>> pyhf.tensorlib.percentile(a, 50, axis=1)
tensor([7., 2.])

Parameters:

• tensor_in (tensor) – The tensor containing the data

• q (float or tensor) – The \(q\)-th percentile to compute

• axis (number or tensor) – The dimensions along which to compute

• interpolation (str) –

  The interpolation method to use when the desired percentile lies between two data points i < j:

  • 'linear': i + (j - i) * fraction, where fraction is the fractional part of the index surrounded by i and j.

  • 'lower': Not yet implemented in PyTorch.

  • 'higher': Not yet implemented in PyTorch.

  • 'midpoint': Not yet implemented in PyTorch.

  • 'nearest': Not yet implemented in PyTorch.

Returns:

The value of the \(q\)-th percentile of the tensor along the specified axis.

Return type:

PyTorch tensor

Added in version 0.7.0.

poisson(n, lam)

The continuous approximation, using \(n! = \Gamma\left(n+1\right)\), to the probability mass function of the Poisson distribution evaluated at n given the parameter lam.

Note

Though the p.m.f of the Poisson distribution is not defined for \(\lambda = 0\), the limit as \(\lambda \to 0\) is still defined, which gives a degenerate p.m.f. of

\[\begin{split}\lim_{\lambda \to 0} \,\mathrm{Pois}(n | \lambda) = \left\{\begin{array}{ll} 1, & n = 0,\\ 0, & n > 0 \end{array}\right.\end{split}\]

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> pyhf.tensorlib.poisson(5., 6.)
tensor(0.1606)
>>> values = pyhf.tensorlib.astensor([5., 9.])
>>> rates = pyhf.tensorlib.astensor([6., 8.])
>>> pyhf.tensorlib.poisson(values, rates)
tensor([0.1606, 0.1241])

Parameters:

• n (tensor or float) – The value at which to evaluate the approximation to the Poisson distribution p.m.f. (the observed number of events)

• lam (tensor or float) – The mean of the Poisson distribution p.m.f. (the expected number of events)

Returns:

Value of the continuous approximation to Poisson(n|lam)

Return type:

PyTorch FloatTensor

poisson_dist(rate)

The Poisson distribution with rate parameter rate.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> rates = pyhf.tensorlib.astensor([5, 8])
>>> values = pyhf.tensorlib.astensor([4, 9])
>>> poissons = pyhf.tensorlib.poisson_dist(rates)
>>> poissons.log_prob(values)
tensor([-1.7403, -2.0869])

Parameters:

rate (tensor or float) – The mean of the Poisson distribution (the expected number of events)

Returns:

The Poisson distribution class

Return type:

PyTorch Poisson distribution

poisson_logpdf(n, lam)

power(tensor_in_1, tensor_in_2)

product(tensor_in, axis=None)

ravel(tensor)

Return a flattened view of the tensor, not a copy.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> tensor = pyhf.tensorlib.astensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
>>> pyhf.tensorlib.ravel(tensor)
tensor([1., 2., 3., 4., 5., 6.])

Parameters:

tensor (Tensor) – Tensor object

Returns:

A flattened array.

Return type:

torch.Tensor

reshape(tensor, newshape)

shape(tensor)

simple_broadcast(*args)

Broadcast a sequence of 1 dimensional arrays.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> pyhf.tensorlib.simple_broadcast(
...   pyhf.tensorlib.astensor([1]),
...   pyhf.tensorlib.astensor([2, 3, 4]),
...   pyhf.tensorlib.astensor([5, 6, 7]))
[tensor([1., 1., 1.]), tensor([2., 3., 4.]), tensor([5., 6., 7.])]

Parameters:

args (Array of Tensors) – Sequence of arrays

Returns:

The sequence broadcast together.

Return type:

list of Tensors

sqrt(tensor_in)

stack(sequence, axis=0)

sum(tensor_in, axis=None)

tile(tensor_in, repeats)

Repeat tensor data along a specific dimension

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> a = pyhf.tensorlib.astensor([[1.0], [2.0]])
>>> pyhf.tensorlib.tile(a, (1, 2))
tensor([[1., 1.],
        [2., 2.]])

Parameters:

• tensor_in (tensor) – The tensor to be repeated

• repeats (tensor) – The tuple of multipliers for each dimension

Returns:

The tensor with repeated axes

Return type:

PyTorch tensor

to_numpy(tensor_in)

Convert the PyTorch tensor to a numpy.ndarray.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> tensor = pyhf.tensorlib.astensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
>>> tensor
tensor([[1., 2., 3.],
        [4., 5., 6.]])
>>> numpy_ndarray = pyhf.tensorlib.to_numpy(tensor)
>>> numpy_ndarray
array([[1., 2., 3.],
       [4., 5., 6.]])
>>> type(numpy_ndarray)
<class 'numpy.ndarray'>

Parameters:

tensor_in (tensor) – The input tensor object.

Returns:

The tensor converted to a NumPy ndarray.

Return type:

numpy.ndarray

tolist(tensor_in)

transpose(tensor_in)

Transpose the tensor.

Example

>>> import pyhf
>>> pyhf.set_backend("pytorch")
>>> tensor = pyhf.tensorlib.astensor([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
>>> tensor
tensor([[1., 2., 3.],
        [4., 5., 6.]])
>>> pyhf.tensorlib.transpose(tensor)
tensor([[1., 4.],
        [2., 5.],
        [3., 6.]])

Parameters:

tensor_in (tensor) – The input tensor object.

Returns:

The transpose of the input tensor.

Return type:

PyTorch FloatTensor

Added in version 0.7.0.

where(mask, tensor_in_1, tensor_in_2)

zeros(shape, dtype='float')