python - pick TxK numpy array from TxN numpy array using TxK column index array -
this indirect indexing problem.
it can solved list comprehension.
the question whether, or, how solve within numpy,
when data.shape (t,n) , c.shape (t,k)
and each element of c int between 0 , n-1 inclusive, is, each element of c intended refer column number data.
the goal obtain out where
out.shape = (t,k) and each i in 0..(t-1)
the row out[i] = [ data[i, c[i,0]] , ... , data[i, c[i,k-1]] ]
concrete example:
data = np.array([\ [ 0, 1, 2],\ [ 3, 4, 5],\ [ 6, 7, 8],\ [ 9, 10, 11],\ [12, 13, 14]]) c = np.array([ [0, 2],\ [1, 2],\ [0, 0],\ [1, 1],\ [2, 2]]) out should out = [[0, 2], [4, 5], [6, 6], [10, 10], [14, 14]] the first row of out [0,2] because columns chosen given c's row 0, 0 , 2, , data[0] @ columns 0 , 2 0 , 2.
the sec row of out [4,5] because columns chosen given c's row 1, 1 , 2, , data[1] @ columns 1 , 2 4 , 5.
numpy fancy indexing doesn't seem solve in obvious way because indexing info c (e.g. data[c], np.take(data,c,axis=1) ) produces 3 dimensional array.
a list comprehension can solve it:
out = [ [data[rowidx,i1],data[rowidx,i2]] (rowidx, (i1,i2)) in enumerate(c) ]
if k 2 suppose marginally ok. if k variable, not good.
the list comprehension has rewritten each value k, because unrolls columns picked out of data each row of c. violates dry.
is there solution based exclusively in numpy?
you can avoid loops np.choose:
in [1]: %cpaste pasting code; come in '--' lone on line stop or utilize ctrl-d. info = np.array([\ [ 0, 1, 2],\ [ 3, 4, 5],\ [ 6, 7, 8],\ [ 9, 10, 11],\ [12, 13, 14]]) c = np.array([ [0, 2],\ [1, 2],\ [0, 0],\ [1, 1],\ [2, 2]]) -- in [2]: np.choose(c, data.t[:,:,np.newaxis]) out[2]: array([[ 0, 2], [ 4, 5], [ 6, 6], [10, 10], [14, 14]]) python numpy indexing
