MATLAB矩阵重塑求助：12×14×6400转1200×14×64报错

Hey there! Let's work through this reshaping problem step by step—your error message usually pops up when you use invalid indices (like non-integers or negatives) or misalign how MATLAB stores and rearranges matrix elements.

First, let’s confirm the math checks out: your original matrix has 12*14*6400 = 1,075,200 elements, and your target 1200*14*64 equals the same total—so this reshaping is totally feasible. The key is preserving the second dimension (14) while splitting the third dimension (6400) into 1200 groups of 64, which maps directly to expanding the first dimension from 12 to 1200 (since 12*100 = 1200 and 6400/100 = 64).

The Correct Approach (Using permute + reshape) #

MATLAB uses column-major (column-first) order when reshaping, so we need to adjust the dimension order first to make sure our splitting happens exactly how you want. Here's how to do it:

1. Swap the second and third dimensions to bring the first and third dimensions adjacent—this lets us split the third dimension's data alongside the first dimension.
2. Reshape the adjusted matrix to expand the first dimension to 1200 and shrink the third to 64.
3. Swap back the second and third dimensions to restore the original second dimension position.

Here's the code:

% Assume your original matrix is named A
A = rand(12, 14, 6400); % Example test data

% Step 1: Rearrange dimensions to 12×6400×14
A_adjusted = permute(A, [1 3 2]);

% Step 2: Reshape to 1200×64×14 (12*100=1200, 6400/100=64)
A_reshaped = reshape(A_adjusted, 1200, 64, 14);

% Step 3: Rearrange back to 1200×14×64
A_final = permute(A_reshaped, [1 3 2]);

Or a more concise one-liner if you prefer:

A_final = permute(reshape(permute(A, [1 3 2]), 1200, 64, 14), [1 3 2]);

Why This Works #

This method ensures that each of your original 12 "windows" (first dimension) gets split into 100 smaller windows, with each small window containing 64 consecutive elements from the original third dimension. The permute commands fix the column-major ordering quirk so your second dimension stays intact throughout the process.

What Might Have Caused Your Error #

If you tried to reshape directly without adjusting dimensions (e.g., reshape(A, 1200,14,64)), MATLAB would rearrange elements in column-first order, which would mix data across your original first dimension windows. If you used invalid indexing (like non-integer step sizes or out-of-bounds indices), you’d get that "Subscript indices must be real positive integers" error.

内容的提问来源于stack exchange，提问作者J.Koh