使用NumPy计算deque中多维数组加权平均报错的解决问询

Hey there! Let's break down why your weighted average is throwing that error and how to fix it.

First, the core issue: when you pass a deque of (60,60,3) images to np.average, it gets converted into a 4D NumPy array with shape (5, 60, 60, 3) (since your deque maxlen is 5). Your motion_weights are a 1D array of length 5, and NumPy doesn't automatically know how to align those weights with the 4D array—even if you specify axis=0, you need to make sure the weights can broadcast across the other dimensions.

Here's the step-by-step fix:

Step 1: Convert the Deque to a NumPy Array #

First, turn your deque into a proper NumPy array so we can work with its dimensions clearly:

buffer_array = np.array(motion_buffer)

This gives you an array with shape (5, 60, 60, 3) (5 images, each 60x60 pixels with 3 color channels).

Step 2: Reshape Weights for Broadcasting #

Your weights need to match the axis you're averaging over (axis 0, the number of images), and also be compatible with the other dimensions. Reshape the 1D weights into a 4D array that can broadcast across the pixels and channels:

# Reshape (5,) to (5, 1, 1, 1) so each weight applies to all pixels/channels of one image
reshaped_weights = motion_weights.reshape(-1, 1, 1, 1)

Alternatively, you can use np.newaxis for the same effect:

reshaped_weights = motion_weights[:, np.newaxis, np.newaxis, np.newaxis]

Step 3: Compute the Weighted Average #

Now call np.average with axis=0 and the reshaped weights:

motion_avg = np.average(buffer_array, axis=0, weights=reshaped_weights)

The result will be a (60,60,3) array—exactly the shape of your original images, which is what you want!

Full Working Code #

Here's the complete example with simulated image data:

from collections import deque
import numpy as np

# Initialize buffer and weights
motion_buffer = deque(maxlen=5)
motion_weights = np.array([5./15, 4./15, 3./15, 2./15, 1./15])

# Add 5 sample (60,60,3) images to the buffer
for _ in range(5):
    motion_buffer.append(np.random.rand(60, 60, 3))

# Convert deque to NumPy array
buffer_array = np.array(motion_buffer)

# Reshape weights for broadcasting
reshaped_weights = motion_weights.reshape(-1, 1, 1, 1)

# Calculate weighted average
motion_avg = np.average(buffer_array, axis=0, weights=reshaped_weights)

# Verify the result shape
print(motion_avg.shape)  # Output: (60, 60, 3)

Bonus: Handle Partially Filled Buffers #

If your deque might have fewer than 5 elements (before it's fully populated), adjust the weights to match the current buffer length:

current_length = len(motion_buffer)
# Use only the last N weights (since deque keeps the most recent elements)
current_weights = motion_weights[-current_length:]
reshaped_weights = current_weights.reshape(-1, 1, 1, 1)
motion_avg = np.average(buffer_array, axis=0, weights=reshaped_weights)

That's it! The key was making sure the weights can broadcast across all the dimensions of your images so NumPy knows how to apply each weight to an entire image, not just a single element.

内容的提问来源于stack exchange，提问作者Hrushi