linalg:greater

(linalg:greater a b)

Returns the elementwise a > b comparison as a 0.0/1.0 mask (numpy's >, which produces a boolean array); either operand may be a scalar, and arrays broadcast by numpy's rules exactly as linalg:add describes. Multiply by the mask where numpy would boolean-index -- (linalg:greater x 0) is the relu-gradient mask, and a thresholded linalg:rand gives a dropout mask. The siblings are linalg:greater-equal, linalg:less, linalg:less-equal and linalg:equal.