Divide by zero encountered in python 2 but works on python 3

Question

I'm trying to implement the adaBoost algorithm in python.

The code below runs and works as expected on python 3, but fails on python 2. On python 2 the line hyp_w_arr[itr] = 0.5 * log((1-err)/err)
returns "divide by zero encountered in long_scalars", while on python 3 everything works.

from numpy import *
import numpy.random
from sklearn.datasets import fetch_mldata
import sklearn.preprocessing

def ADAboost_learn(T,training_data ,training_labels, WL):

    for itr in range(T):
        print(itr)

        hyp_arr[itr][0],hyp_arr[itr][1] = WL(DLW)
        hyp_result = [1 if (DLW[index][0][hyp_arr[itr][0]] <= hyp_arr[itr][1])
                        else -1
                     for index in range(len(DLW))]

        err = sum([DLW[index][2] 
               if(hyp_result[index] != DLW[index][1]) else 0
               for index in range(len(DLW))])

        hyp_w_arr[itr] = 0.5 * log((1-err)/err)

are there python/numpy restrictions in 2.0 that dropped in python 3 regarding the division resolution?

Answer 1

Yes, in python 2 the division / is a intger division but in python 3 it is float. // is integer division in both 2 and 3.

You can solve your problem by adding

from __future__ import division