调用k_mean.fit触发ValueError：序列设置数组元素错误求助

Hey there, let's break down why you're hitting this error and how to get your clustering working smoothly.

What's causing the error? #

The root issue here is that the arrays in your features_list don't all have the same shape. When you convert this list to a NumPy array with np.asarray(features_list), you end up with an object-type array (instead of a regular numerical array) where each element is a sub-array of varying dimensions.

Even when you try to reshape this object array, it can't be converted into a clean 2D array of shape (number_of_samples, number_of_features)—which is exactly what KMeans expects. The error message is telling you that KMeans can't process this messy, inconsistent sequence of data.

Looking at your printed features_flat output, you can see nested array([[[...]]]) structures, which confirms your features are still multi-dimensional and inconsistent.

How to fix it #

Follow these steps to resolve the issue:

1. Check and standardize feature shapes
   First, verify that every feature in your list has the same dimensions. Run this quick check:

   for idx, feat in enumerate(features_list):
       print(f"Feature {idx} shape: {feat.shape}")
   

   If you see different shapes here, you need to go back to your feature extraction step. Make sure you're extracting features the same way for every image—for example, using the same CNN layer output, or resizing features to a fixed dimension before adding them to features_list.

2. Flatten features correctly
   Once all your features have the same shape, flatten each one individually and then combine them into a 2D numerical array. This avoids the object-type array problem:

   # Flatten each feature and stack into a 2D array
   features_flat = np.array([feat.flatten() for feat in features_list])
   

3. Verify your final array
   Double-check that your flattened array is ready for KMeans:

   print(f"Flattened features shape: {features_flat.shape}")  # Should be (n_samples, n_features)
   print(f"Flattened features dtype: {features_flat.dtype}")  # Should be a numerical type like float64
   

Example working code #

Here's a complete example to illustrate the fix:

import numpy as np
from sklearn import cluster

# Simulate a list of consistent-shape features (e.g., 2x2x4 arrays)
features_list = []
for _ in range(50):
    # Generate a random 3D feature array with fixed shape
    feat = np.random.rand(2, 2, 4)
    features_list.append(feat)

# Flatten each feature properly
features_flat = np.array([f.flatten() for f in features_list])
print(f"Processed features shape: {features_flat.shape}")  # Output: (50, 16)

# Run KMeans without errors
k_means = cluster.KMeans(n_clusters=10, n_jobs=-1)
k_means.fit(features_flat)
print("Clustering finished successfully!")

Key takeaway #

KMeans requires a uniform 2D numerical array where each row is a single sample's feature vector. Your error comes from inconsistent feature shapes leading to an unprocessable object array. Standardize your feature dimensions first, then flatten correctly, and you'll be good to go!

内容的提问来源于stack exchange，提问作者Emily Chu