Python数组索引中nonzero函数的代码解析与优化方案问询

Let’s break down what your original code does, then look at a cleaner, more readable alternative.

What the Original Code Does #

First, let’s unpack that line step by step (assuming bestClustAss is a NumPy matrix, since it uses the .A attribute):

1. bestClustAss[:,0].A: Converts the first column of the NumPy matrix into a regular NumPy array.
2. bestClustAss[:,0].A == 1: Creates a boolean array where each element is True if the corresponding value in the first column equals 1, and False otherwise.
3. nonzero(...): Returns a tuple of indices where the boolean array is True. Since we’re working with a 1D array here, we take the first element of the tuple ([0]) to get a flat list of row indices.
4. bestClustAss[..., 0] = le...: Assigns the value (from le, probably a label encoder or similar variable) to the first column of bestClustAss at all the row indices we found.

In plain terms: It finds every row in bestClustAss where the first column is 1, then sets that first column value to le... for those rows.

Why It’s Hard to Read #

• It uses NumPy’s matrix type (the .A attribute is specific to matrices), which is now discouraged in favor of regular NumPy arrays.
• The nested function calls (nonzero(...) wrapped around the equality check) make the logic harder to follow at a glance.
• The indexing syntax is dense and not intuitive for someone unfamiliar with older NumPy matrix conventions.

More Readable Python/NumPy Alternative #

Using modern NumPy array practices and boolean indexing (which is far more explicit), here’s a cleaner version:

If bestClustAss is a NumPy Array (Recommended) #

# Create a boolean mask for rows where the first column equals 1
mask = bestClustAss[:, 0] == 1

# Assign the value to those rows' first column
bestClustAss[mask, 0] = le...

If You Must Keep It as a NumPy Matrix #

If you’re stuck working with the matrix type for compatibility:

# Convert matrix to array first (easier to work with)
clust_ass_array = bestClustAss.A

# Create mask and update values
mask = clust_ass_array[:, 0] == 1
clust_ass_array[mask, 0] = le...

# Optional: Convert back to matrix if needed
bestClustAss = np.matrix(clust_ass_array)

This version makes the intent crystal clear: first define which rows we want to target, then update those rows. No nested calls, no confusing .A attribute (unless you absolutely need it), and the logic flows linearly.

内容的提问来源于stack exchange，提问作者shatter