In python with numpy what is the fastest way to turn an array like
array([0,2,3,1,0,0,1])
into another array
array([1,1,2,2,2,3,6])
where the first array gives the frequency of each index (i.e. index 0 has a frequency of 0, index 1 has a frequency of 2, index 2 has a frequency of 3, and so on) and the second repeats each index as many times as specified in the first array.
Background: I use this to 'enflate' (I can't find any better word for it) a k by k Matrix M (sparse or not) given a length k frequency vector f:
f = np.array([0,2,3,1,0,0,1])
f_ = np.array([1,1,2,2,2,3,6])
M_ = M[f_[:,None],f_]
Use np.repeat on the range array covering the length of the input array with the array itself for the count of repetitions -
np.repeat(np.arange(len(a)), a)
Sample run -
In [109]: a
Out[109]: array([0, 2, 3, 1, 0, 0, 1])
In [110]: np.repeat(np.arange(len(a)), a)
Out[110]: array([1, 1, 2, 2, 2, 3, 6])