NumPy代码中X[y == label].sum(axis=0)语法理解咨询

Hey there! Let's break down this line piece by piece so you can see exactly how it works with your sample data.

Step 1: Boolean Indexing with y == label #

In NumPy, when you use a boolean array to index a 2D array (like your X), it acts as a row filter: it selects only the rows where the boolean value is True.

For example, when label = 1, y == label gives you the boolean array [False, True, False, True]. Applying this to X (X[y == label]) will pull out the rows of X that correspond to True positions—so rows 1 and 3:

# X[y == 1] results in:
array([[1, 0, 1, 1],
       [1, 0, 1, 0]])

Step 2: Summing Along Columns with .sum(axis=0) #

The .sum(axis=0) method calculates the sum vertically, across columns (instead of horizontally across rows, which would be axis=1).

Using the subset above, summing each column gives:

• Column 0: 1 + 1 = 2
• Column 1: 0 + 0 = 0
• Column 2: 1 + 1 = 2
• Column 3: 1 + 0 = 1

So the result is array([2, 0, 2, 1]), which gets stored as counts[1].

Step 3: Putting It All Together for Both Labels #

When label = 0, y == label gives [True, False, True, False]. Filtering X gives rows 0 and 2:

# X[y == 0] results in:
array([[0, 1, 0, 1],
       [0, 0, 0, 1]])

Summing these columns gives array([0, 1, 0, 2]), stored as counts[0].

Final Result #

After the loop runs, your counts dictionary will look like this:

{0: array([0, 1, 0, 2]), 1: array([2, 0, 2, 1])}

Each value is the total number of times each feature (column) appears as non-zero in samples belonging to that label.

内容的提问来源于stack exchange，提问作者Samarth Upadhyaya