将 function 应用于 numpy 矩阵列表

Question

I am working on python. So, I got a main matrix lets call it "X". Then I got a list of other 5000 matrices which is called Z. I want to apply between X and each matrix within Z the sklearn.metrics.pairwise.cosine_similarity function and take the average within matrix. I do not want to do a for loop as it takes a ton of time. Is there a way, without multiprocessing to do this without the for loop?

from sklearn.metrics.pairwise import cosine_similarity
Z = [numpy.random.rand(7,12),numpy.random.rand(7,12),numpy.random.rand(7,12)] #in my example X contains 5000 matrices, here only 3
X = numpy.random.rand(7,12)

#the loop I want to get rid of
cosine_similarities = []
for matrix in Z:
    cosine_similarities.append(cosine_similarity(X, matrix).mean())

Answer 1

Here is a way to avoid loops. If all matrices in Z are of same shape, you can simply stack them and find cosine similarity and then split them:

Y = np.vstack(Z)
cosine_similarities  = sklearn.metrics.pairwise.cosine_similarity(X, Y).reshape(len(Z),X.shape[0],-1)

it will return an array that each row is the cosine similarity of X and elements of Z . And if you want to take average of that:

cosine_similarities = cosine_similarities.mean((1,2))