std_op

→TensorImplPtr

int std_op(const int & a, const int & axes, bool keepdims)

source edit

Python wrappers

pylucid.ops.stdfunction

Compute the (biased) standard deviation along axes.

Defined as $\mathrm{std}(x) = \sqrt{\mathrm{var}(x, \text{axes}, \text{keepdims})}$ , where the underlying variance uses the biased estimator $\frac{1}{N}\sum (x - \bar{x})^2$ . No new backward node is needed: the gradient flows naturally through SqrtBackward and VarBackward via the chain rule.

Math

\mathrm{std}(x) = \sqrt{\frac{1}{N}\sum_{i \in \text{axes}} (x_i - \bar{x})^2}.

See Also

var_op — variance the standard deviation is built on.

Parameters

aTensorImplPtr

Input tensor of float dtype.

axesstd::vector<int>

Axes to reduce. Negative indices wrap.

keepdimsbool

If true, retains reduced dims as size-1.

Returns

TensorImplPtr

Output tensor of the reduced shape.