I have three numpy arrays:
X
: a 3073 x 49000 matrix
W
: a 10 x 3073 matrix
y
: a 49000 x 1 vector
y
contains values between 0 and 9, each value represents a row in W
.
I would like to add the first column of X
to the row in W
given by the first element in y
. I.e. if the first element in y
is 3, add the first column of X
to the fourth row of W
. And then add the second column of X
to the row in W
given by the second element in y
and so on, until all columns of X
has been aded to the row in W
specified by y
, which means a total of 49000 added rows.
W[y] += X.T
does not work for me, because this will not add more than one vector to a row in W
.
Please note: I'm only looking for vectorized solutions. I.e. no for-loops.
EDIT: To clarify I'll add an example with small matrix sizes adapted from Salvador Dali's example below.
In [1]: import numpy as np
In [2]: a, b, c = 3, 4, 5
In [3]: np.random.seed(0)
In [4]: X = np.random.randint(10, size=(b,c))
In [5]: W = np.random.randint(10, size=(a,b))
In [6]: y = np.random.randint(a, size=(c,1))
In [7]: X
Out[7]:
array([[5, 0, 3, 3, 7],
[9, 3, 5, 2, 4],
[7, 6, 8, 8, 1],
[6, 7, 7, 8, 1]])
In [8]: W
Out[8]:
array([[5, 9, 8, 9],
[4, 3, 0, 3],
[5, 0, 2, 3]])
In [9]: y
Out[9]:
array([[0],
[1],
[1],
[2],
[0]])
In [10]: W[y.ravel()] + X.T
Out[10]:
array([[10, 18, 15, 15],
[ 4, 6, 6, 10],
[ 7, 8, 8, 10],
[ 8, 2, 10, 11],
[12, 13, 9, 10]])
In [11]: W[y.ravel()] = W[y.ravel()] + X.T
In [12]: W
Out[12]:
array([[12, 13, 9, 10],
[ 7, 8, 8, 10],
[ 8, 2, 10, 11]])
The problem is to get BOTH column 0 and column 4 in X added to row 0 in W, as well as both column 1 and 2 in X added to row 1 in W.
The desired outcome is thus:
W = [[17, 22, 16, 16],
[ 7, 11, 14, 17],
[ 8, 2, 10, 11]]
See Question&Answers more detail:
os