c++ - How to iterate an array in Python pointer-likely and efficiently? -

i quite new in python, iterate array calculate next element based on previous elements. can think of in c++ manner how achieve in python using numpy or pandas? know similar method using shifting way seems not efficient.

the following simple fibonacci example:

int arr[10]; arr[0] = 1; arr[1] = 1;  int* pt = &arr[2];         <--get iterator moving pointer int count = 8; while (count > 0) {     *pt = pt[-1] + pt[-2]; <--access previous k element based on current index     count--;     pt++;                  <--point next element } (int = 0; < 10; i++)     cout << arr[i] << endl; 

list:

l = [1, 1] while len(l) != 10:     l.append(l[-1] + l[-2]) <--but have append element every time print l 

the direct numpy immitation of c++ is

arr=np.ones((10,)) in range(2,arr.shape[0]):      arr[i]=arr[i-1]+arr[i-2] 

producing:

array([  1.,   1.,   2.,   3.,   5.,   8.,  13.,  21.,  34.,  55.]) 

this isn't efficient way of doing in numpy, it's discussion start.

most fast numpy operations use compiled c code, , results buffered, tricky perform fast sequential operations. best think of numpy operations acting in parallel, on terms of array @ once (without preference order). exceptions ufunc cumsum , cumprod - cumulative processes, , unbuffered ufunc called at. i'd have play around see if there's way of calculating series 1 of these tools.

another option implement calculation in c or c++ , link it. cython convenient tool doing this.

---

http://wiki.scipy.org/cookbook/ctypes#head-6a582bd7b101bca0df6e5631a06e2d1d76e35a95 example of using ctypes , c code numpy calculate fibonachi.

http://numpy-discussion.10968.n7.nabble.com/vectorizing-recursive-sequences-td35532.html describes cleaver way of calculating series. depends on ufunc taking out array. author admits depends on implementation details, calculation not buffered.

arr=np.ones((10,)) np.add(arr[:-2], arr[1:-1], out=arr[2:]) 

the elements of 1st 2 arguments added, element element, , stored in out array.

arr[2:] = arr[:-2]+arr[1:-1] 

does not work because of buffering

---

http://docs.cython.org/src/tutorial/cython_tutorial.html#fibonacci-fun

is cython example of fibonacci. should fast, prints results, opposed accumulating them in array. still shouldn't hard cython/c version stores results in cython array or memoryview.

here's cython script, can saved pyx, compiled , imported. refined usefulness , speed, it's enough test concept:

import numpy np narr = np.ones((10,), dtype=np.dtype("i")) cdef int [:] narr_view = narr  cpdef void fib(int[:] arr):     = arr.shape[0]     in range(2,i):         arr[i] = arr[i-2]+arr[i-1] fib(narr_view) print 'fib arr:', narr