The reshape function requires the total number of elements in the input array to be preserved in the output array. Hence, the product of the specified dimensions must equal the total number of elements in the input.
In the provided code:
X = rand(4348,23); %Sample data with same size
[num_samples, num_features] = size(X);
num_padded_rows = ceil(num_samples/144)*144;
num_padded_cols = ceil(num_features/72)*72;
X_padded = [X, zeros(num_samples, num_padded_cols - num_features)];
numel(X_padded)
%X_reshaped = reshape(X_padded, [num_padded_rows/144, 8, 3, 3, ceil(num_padded_cols/72)]);
prod([num_padded_rows/144, 8, 3, 3, ceil(num_padded_cols/72)])
The total number of elements are unequal, leading to an error.
One way to resolve this is to pass "[]" as the first dimension to the reshape function, maintaining the remaining dimensions as [8,3,3,2] for the CNN layer. This lets the function calculate the necessary size automatically, ensuring the element count remains unchanged.
X_reshaped = reshape(X_padded, [], 8, 3, 3, 2);
size(X_reshaped)
More details on the reshape function can be found in the following documentation:
Hope this helps!
