I have three huge numpy.arrays, and need to execute some conditional statements involving all three numpy.arrays fast. All numpy.arrays are of the same dimension (NxD)(N>2 and D>1) and of same datatype. Normally I would do as shown below
for i in range(n): for j in range(d): if np.sign(nabla[i][j]) != np.sign(delta[i][j]): g[i][j] = g[i][j] + 0.2 if np.sign(nabla[i][j]) == np.sign(delta[i][j]): g[i][j] = g[i][j] * 0.8
if I only had to operate with one numpy.array I would do
g[g < val] = newval
But I am receiving an error by applying the same principles since delta and nabla are more than two-dimensional.
You should consider using Boolean Indexing instead, e.g.:
mask = np.sign(nabla) == np.sign(delta) g[mask] *= 0.8 g[~mask] += 0.2
g = np.where(np.sign(nabla) == np.sign(delta), g * 0.8, g + 0.2)