torchgear.math

torchgear.math.abs2(x: Tensor) → Tensor

Computes the squared magnitude of a tensor.

Parameters:

x (Tensor) – Input tensor (real or complex).

Returns:

Squared magnitude for complex input, or square for real input.

Return type:

Tensor

torchgear.math.expi(x: Tensor) → Tensor

Computes the complex exponential of a real-valued tensor:

\[y=\text{e}^{\text{i} x}\]

Parameters:

x – A real-valued tensor.

Returns:

Complex exponential of x.

Return type:

Tensor

torchgear.math.polynomial(x, coefficients: Sequence)

Computes the value of a polynomial:

\[f(x)=a_0+a_1x+a_2x^2+\ldots+a_nx^n\]

Parameters:

• x – Value of \(x\).

• coefficients (Sequence) – Coefficients \(a_0,\ldots,a_n\). None in coefficients is interpreted as zero but the last coefficient (i.e. the coefficient of the highest-order-term) cannot be None.

Returns:

Value of \(f(x)\).

torchgear.math.ssqrt(x: Tensor) → tuple[Tensor, Tensor]

Secure version of torch.sqrt(). Negative values are treated as zero to avoid NaNs in result.

Parameters:

x (Tensor) – Input tensor.

Returns:

A 2-tuple of tensor. The first is torch.sqrt(x.relu()) and the second is a bool tensor representing non-negative values in x.