Error pops up for seemingly correct code in the first assignment of the first week of the 4th course

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

While implementing back prop for the convolution layer. there was a line:
da_prev_pad[vert_start:vert_end, horiz_start:horiz_end, :] += W[:, :, :, c] * dZ[i, h, w, c]
, inside the nested for loop running through i, h, w, c that worked well.

But it seems to me as though, another correct assignment operation with just “=” doesn’t work at all, as the LHS represents different slices of da_prev_pad for different iterations of the for loop, we are working on different segments, so, it shouldn’t really matter whether we use “+=” or just “=”.

Please clarify this doubt. I have similar doubts in the implementation of back prop for pooling.
Thank you.

2

Hello @Susmit_Sarkar,

+= and = are two completely different operators and have very different effects and therefore one is always not a equivalent replacement for another.

+= adds whatever in the RHS to the object in the LHS.

= assigns whatever object in the RHS to the variable name in the LHS.

in short, += accumulates, and = only assigns.

The assignment requires us to accumulate up all the gradient values over those 4 nested loops, and to do accumulation, we need +=. In contrast, replacing += with = does not accumulate.

Cheers,
Raymond