Standard implementation of vectorize_sequences

Question

In François Chollet's Deep Learning with Python, appears this function:

def vectorize_sequences(sequences, dimension=10000):
    results = np.zeros((len(sequences), dimension))
    for i, sequence in enumerate(sequences):
        results[i, sequence] = 1.
    return results

I understand what this function does. This function is asked about in this quesion and in this question as well, also mentioned here, here, here, here, here & here. Despite being so wide-spread, this vectorization is, according to Chollet's book is done "manually for maximum clarity." I am interested whether there is a standard, not "manual" way of doing it.

Is there a standard Keras / Tensorflow / Scikit-learn / Pandas / Numpy implementation of a function which behaves very similarly to the function above?

Simpleton

Answer 1

Solution with MultiLabelBinarizer

Assuming sequences is an array of integers with maximum possible value upto dimension-1, we can use MultiLabelBinarizer from sklearn.preprocessing to replicate the behaviour of the function vectorize_sequences

from sklearn.preprocessing import MultiLabelBinarizer

mlb = MultiLabelBinarizer(classes=range(dimension))
mlb.fit_transform(sequences)

Solution with Numpy broadcasting

Assuming sequences is an array of integers with maximum possible value upto dimension-1

(np.array(sequences)[:, :, None] == range(dimension)).any(1).view('i1')

---

Worked out example

>>> sequences
[[4, 1, 0], 
 [4, 0, 3],
 [3, 4, 2]]

>>> dimension = 10
>>> mlb = MultiLabelBinarizer(classes=range(dimension))
>>> mlb.fit_transform(sequences)

array([[1, 1, 0, 0, 1, 0, 0, 0, 0, 0],
       [1, 0, 0, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 1, 1, 1, 0, 0, 0, 0, 0]])


>>> (np.array(sequences)[:, :, None] == range(dimension)).any(1).view('i1')

array([[0, 1, 1, 1, 0, 0, 0, 0, 0, 0],
       [1, 0, 1, 0, 1, 0, 0, 0, 0, 0],
       [1, 1, 0, 0, 1, 0, 0, 0, 0, 0]])

Encephalitis