I have a numpy csr matrix and I want to get it's mean, but it contains a lot of zeros, because I eliminated all values that are on the main diagonal and below it taking only the upper triangle values, and now my csr matrix when converted to array looks like that:
0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0. 0. 0.
0. 0. 0. 0. 0.63646664 0.34827262
0.24316454 0.1362165 0.63646664 0.15762204 0.31692202 0.12114576
0.35917146
As far as I understand this zeros are important to be there in order for the csr matrix to work and display things like this:
(0,5) 0.5790418
(3,10) 0.578210
(5,20) 0.912370
(67,5) 0.1093109
I saw that csr matrix has it's own mean function , but does this mean function takes into account all the zeros, therefore dividing on the number of elements in the array including the zeros? Because I need the mean on only the non zero values. My matrix contains the similarities between multiple vectors and is more like a list of matrices something like that:
[[ 0. 0.63646664 0.48492084 0.42134077 0.14366401 0.10909745
0.06172853 0.08116201 0.19100626 0.14517247 0.23814955 0.1899649
0.20181049 0.25663533 0.21003358 0.10436352 0.2038447 1.
0.63646664 0.34827262 0.24316454 0.1362165 0.63646664 0.15762204
0.31692202 0.12114576 0.35917146]
[ 0. 0. 0.58644824 0.4977052 0.15953415 0.46110612
0.42580993 0.3236768 0.48874263 0.44671607 0.59153001 0.57868948
0.27357541 0.51645488 0.43317846 0.50985032 0.37317457 0.63646664
1. 0.51529235 0.56963948 0.51218525 1. 0.38345582
0.55396192 0.32287605 0.46700191]
[ 0. 0. 0. 0.6089113 0.53873289 0.3367261
0.29264493 0.13232082 0.43288206 0.80079927 0.37842518 0.33658945
0.61990095 0.54372307 0.49982101 0.23555037 0.39283379 0.48492084
0.58644824 0.64524906 0.31279271 0.39476181 0.58644824 0.39028705
0.43856802 0.32296735 0.5541861 ]]
So how can I take the mean on only the non-zero values?
My other question is how can I remove all values that are equal to something, as I pointed out above I probably have to turn the certain value to a zero? But how do I do that ? For example I want to get rid of all values that are equal to 1.0 or bigger? Here is the code I have till this point to make my matrix:
vectorized_words = parse.csr_matrix(vectorize_words(nostopwords,glove_dict))
#calculating the distance/similarity between each vector in the matrix
cos_similiarity = cosine_similarity(vectorized_words, dense_output=False)
# since there are duplicates like (5,0) and (0,5) which we should remove, I use scipy's triu function
coo_cossim = cos_similiarity.tocoo()
vector_similarities = sparse.triu(coo_cossim, k = 1).tocsr()
Yes, csr_matrix.mean()
does include all of the zeros when calculating the mean. As a simple example:
from scipy.sparse import csr_matrix
m = csr_matrix(([1,1], ([2,3],[3,3])), shape=(5,5))
m.toarray()
# returns:
array([[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0],
[0, 0, 0, 1, 0],
[0, 0, 0, 1, 0],
[0, 0, 0, 0, 0]], dtype=int32)
# test the mean method
m.mean(), m.mean(axis=0), m.mean(axis=1)
# returns:
0.080000000000000002,
matrix([[ 0. , 0. , 0. , 0.4, 0. ]]),
matrix([[ 0. ],
[ 0. ],
[ 0.2],
[ 0.2],
[ 0. ]])
If you need to perform a calculation that does not include zeros, you will have to build the result with other methods. It is not terribly hard to do though:
nonzero_mean = m.sum() / m.count_nonzero()
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.