larq.math¶
Math operations that are specific to extremely quantized networks.
heaviside¶
larq.math.heaviside(x)
Heaviside step function with output values 0 and 1.
\[ q(x) = \begin{cases} +1 & x > 0 \\ \hphantom{+}0 & x \leq 0 \end{cases} \]
Arguments
x: Input Tensor
Returns
A Tensor with same type as x.
sign¶
larq.math.sign(x)
A sign function that will never be zero \[ f(x) = \begin{cases} -1 & x < 0 \\ \hphantom{-}1 & x \geq 0 \end{cases} \]
This function is similar to tf.math.sign but will return a binary value and will never be zero.
Arguments
x: Input Tensor
Returns
A Tensor with same type as x.