linalg:var

(linalg:var array &key axis keepdims ddof)

Returns the variance -- the mean of the squared deviations from the mean -- of every element (no :axis) or along one axis, following the same :axis / :keepdims rules as linalg:sum. The divisor is n - ddof: the default :ddof 0 is numpy's np.var and torch's unbiased=False, while :ddof 1 gives the sample variance (Bessel's correction). The square root is linalg:std.