Python矩阵对齐问题：移除transpose后维度不匹配报错求解

Let's break down what's happening here and fix that shape mismatch error.

Your Code Context #

Here's the relevant code snippet you shared:

import numpy as np
from numpy.linalg import norm

# ... (assumed setup for arr and new_arr)
for i in range(len(new_arr)): 
    new_arr[i] = new_arr[i].transpose()
n = len(arr)
u = [0 * n for i in range(n)]
e = [0 * n for i in range(n)]
print(e)
print("Q")
Q = []
for i in range(len(arr)):
    u[i] = new_arr[i] - ((np.dot(new_arr[i],e[i-1])*e[i-1]))
    e[i] = u[i] / norm(u[i])#Cari gapake norm
    e[i] = e[i].transpose()  # Removing this line causes the error
    print(u[i])
    print(e[i])

When you remove e[i] = e[i].transpose(), you get the error:

shapes (1,3) and (1,3) not aligned: 3 (dim 1) != 1 (dim 0)

Why This Happens #

The core issue is how NumPy's np.dot() works with 2D arrays:

• For matrix multiplication (which dot performs on 2D arrays), the number of columns in the first array must equal the number of rows in the second array.
• When you skip the transpose, e[i] ends up as a row vector with shape (1,3). When you compute np.dot(new_arr[i], e[i-1]), if new_arr[i] is also a (1,3) row vector, you're trying to multiply (1,3) · (1,3)—which fails because 3 columns (from the first array) don't match 1 row (from the second).

In your original code, the transpose converts e[i] to a column vector ((3,1)), which makes new_arr[i] (a (1,3) row vector) and e[i-1] compatible for matrix multiplication: (1,3) · (3,1) produces a scalar, which works with your subtraction logic.

Fixes You Can Use #

You have a few options depending on whether you want to keep using transpose or not:

1. Keep the Transpose (Simplest Fix)

The original transpose is doing exactly what you need—converting the row vector to a column vector to make the dot product valid. If this works for your overall logic, there's no need to remove it.

2. Use Element-Wise Inner Product Instead of Matrix Multiplication

If you want to avoid transpose, you can use np.inner() (which computes the inner product regardless of row/column orientation for 2D arrays) or flatten the arrays to 1D first for np.dot():

Option A: Use np.inner()

# Replace the u[i] calculation line with this:
u[i] = new_arr[i] - (np.inner(new_arr[i], e[i-1]) * e[i-1])

Option B: Flatten arrays to 1D

# Flatten both arrays to 1D before the dot product
dot_product = np.dot(new_arr[i].flatten(), e[i-1].flatten())
u[i] = new_arr[i] - (dot_product * e[i-1])

3. Initialize e as Column Vectors From the Start

Instead of transposing later, you can reshape e[i] to a column vector when you compute it:

# Replace the e[i] lines with this:
e[i] = (u[i] / norm(u[i])).reshape(-1, 1)  # Reshapes to (3,1) column vector

All these fixes ensure the dot product operation uses compatible shapes, so you won't get the alignment error anymore.

内容的提问来源于stack exchange，提问作者Hensel