CODEWITHSUNDEEP
pythonnumpyscikit-learncopy
Marcos Galletero Romero
Marcos Galletero Romero

Reputation: 164

X = np.copy(X) ... Is this really useless?

I am currently looking at the sklearn source code in github and I have encountered a line that bugs me. It's this one.

I understand the difference between copying by value and copying by reference (I guess the copy function from numpy is by value and x = y is by reference).

But clearly I don't understand the purpose of copying to itself the same variable as in

Can anyone throw some light upon this?

Thank you very much

Upvotes: 1

Views: 319

Answers (2)

J. Bakker
J. Bakker

Reputation: 356

How would you suspect that the function will look like if you didn't assigned it back to the variable it already defined.

def softmax(X, copy=True):
    """
    Calculate the softmax function.
    The softmax function is calculated by
    np.exp(X) / np.sum(np.exp(X), axis=1)
    This will cause overflow when large values are exponentiated.
    Hence the largest value in each row is subtracted from each data
    point to prevent this.
    Parameters
    ----------
    X : array-like, shape (M, N)
        Argument to the logistic function
    copy : bool, optional
        Copy X or not.
    Returns
    -------
    out : array, shape (M, N)
        Softmax function evaluated at every point in x
    """
    def _softmax(X):
        max_prob = np.max(X, axis=1).reshape((-1, 1))
        X -= max_prob
        np.exp(X, X)
        sum_prob = np.sum(X, axis=1).reshape((-1, 1))
        X /= sum_prob
        return X

    return _softmax(np.copy(X) if copy else X)

This makes the code less readable. Overwriting a local variable when preprocessing the variable is normally accepted as it makes readable code. Adding a second variable will make understanding the code harder. In my example I tried to create code that was as easy to understand without any duplication. But you cannot read this code serial anymore. You first need to read the last statement _softmax(np.copy(X) if copy else X) to understand how it is being used.

In performance this also introduces an additional call what you don't want in these low level functions.

Upvotes: 0

pixelou
pixelou

Reputation: 804

That softmax function offers to run inplace in order to save a duplication op, therefore overwriting the input X with the result of the softmax.

The copy is here to avoid changing the input in case it is needed elswhere.

X = np.copy(X) replaces the original reference with a reference to the copy which is safe to modify.

Upvotes: 1

Related Questions