Currently, index 0 of C_R/G/B and D_R/G/B are still np.empty values. When running predictions this causes inconsistent and numerically unstable predictions. This PR fixes that issue. Moreover, using numpy vectorisation gives easier to read code + a performance boost.
Currently, index 0 of C_R/G/B and D_R/G/B are still np.empty values. When running predictions this causes inconsistent and numerically unstable predictions. This PR fixes that issue. Moreover, using numpy vectorisation gives easier to read code + a performance boost.