newton.utils.leaky_max

newton.utils.leaky_max(a, b)

Compute a numerically stable, differentiable approximation of max(a, b).

This is equivalent to smooth_max(a, b, 1e-5).

Parameters:

• a (float) – The first value.

• b (float) – The second value.

Returns:

A smooth, “leaky” maximum of a and b.

Return type:

float