python - Numba JIT changing results if values are printed -

i started working numba today, because have nested for-loop can take quite while regular python code.

i have macports version of python-2.7 llvm-3.6 , pip version of numba (everything up-to-date)

here code i'm using:

import pandas pd numba import jit numpy import nan, full  @jit def movingaverage(adj_close, maxma):     ma = full([len(adj_close), maxma], nan, dtype=float64)     ind = range( 1, len(adj_close)+1 )     d in ind:         m = max( 0, d-maxma-1)         adj = adj_close[d-1:m:-1] if (m or d==maxma+1) else adj_close[d-1::-1]         cs = adj.cumsum()         in range( len(adj) ):             ma[d-1][i] = ( cs[i] / (i+1) )         print ma     return ma 

i'm calculating rolling mean input adj_close maxma days.

adj_close array of values, 1 value per day

i started creating ma, holder values going calculated. , work out vaules each day individually (note first day can have average involving 1 day, second, 2 , on maxma)

if input adj_close = array(range(5), dtype=float64) , maxma = 3 right answer follows:

array([[  0.,  nan,  nan],        [  1.,   0.5,  nan],        [  2.,   1.5,   1.],        [  3.,   2.5,   2.],        [  4.,   3.5,   3.]]) 

however, if take out print ma line, before return of function, returns part of answer:

array([[ nan,  nan,  nan],        [ nan,  nan,  nan],        [ nan,  nan,  nan],        [  3.,   2.5,   2.],        [  4.,   3.5,   3.]]) 

why happening? why @jit needs print between loops answer right? can rid of print statement (that increases runtime)?

---

edit: i'm accepting @joshadel suggestion , opened issue @ numba's github. i'm, therefore, accepting @mseifert answer workaround solved problem me.

i think numba strange here because of mixture of python , nopython mode. if use python 3.5 returns identical , without print.

for python 2.7 think problem because for-loop either compiled in nopython mode (without print) or in python mode (with print). converted python when exits loop. that's guessing. tried with:

import pandas pd numba import jit numpy import nan, full import numpy np  @jit def movingaverage(adj_close, maxma):     ma = full([len(adj_close), maxma], nan, dtype=np.float64)     ind = range( 1, len(adj_close)+1 )     d in ind:         m = max( 0, d-maxma-1)         adj = adj_close[d-1:m:-1] if (m or d==maxma+1) else adj_close[d-1::-1]         cs = adj.cumsum()         in range( len(adj) ):             ma[d-1][i] = ( cs[i] / (i+1) )         if d == ind[-1]:             return ma # notice return after last loop before loop terminates.     #return ma 

and return:

array([[  0.,  nan,  nan],        [  1.,   0.5,  nan],        [  2.,   1.5,   1.],        [  3.,   2.5,   2.],        [  4.,   3.5,   3.]]) 

this not effient way because of recalculation of len(adj_close)+1. stored somewhere.