Values returned from Python function differ in caller and callee

Question

Running the Python program

import sys
from bisect import bisect
import numpy as np

def insert_sorted(vec, x):
    """ insert x in sorted numpy array vec. Return the insertion position
    and the new list """
    ipos = bisect(vec, x)
    print("nexiting insert_sorted, len(vec), ipos, x =", len(vec), ipos, x)
    return ipos, np.insert(vec, ipos, x)

def growing_pos(vec):
    """ return the successive positions where the elements of vec would be inserted """
    n = len(vec)
    if n < 1:
        return vec.astype(int)
    vec_sorted = np.array([vec[0]])
    ipos = np.zeros(shape=n, dtype=np.int8)
    for i in range(n):
        if i > 0:
            ipos[i], vec_sorted = insert_sorted(vec_sorted, vec[i])
            print("i, ipos[i], vec[i] =", i, ipos[i], vec[i])
            if ipos[i] < 0: # debug
                # print("vec_sorted =",vec_sorted)
                sys.exit("in growing_pos, should not have ipos[i] < 0")
    return ipos

np.random.seed(123)
vec = np.random.uniform(size=1000)
print(growing_pos(vec))

The program stops with

exiting insert_sorted, len(vec), ipos, x = 132 83 0.6006985678335899
i, ipos[i], vec[i] = 132 83 0.6006985678335899

exiting insert_sorted, len(vec), ipos, x = 133 120 0.8658644583032646
i, ipos[i], vec[i] = 133 120 0.8658644583032646

exiting insert_sorted, len(vec), ipos, x = 134 130 0.9835216092035556
i, ipos[i], vec[i] = 134 -126 0.9835216092035556
in growing_pos, should not have ipos[i] < 0

I don’t understand why at the end ipos in insert_sorted() does not equal ipos[i] in growing_pos(). Using bisect_left() or bisect_right() instead of bisect() gives the same result.

Asked By: Fortranner

||

Source

Answer 1

You are declaring your index array with 8-bit integers:

ipos = np.zeros(shape=n, dtype=np.int8)

8-bit integer can hold values from -128 to 127. You are trying to insert value 130, which overflows and becomes -126. Newer numpy versions won’t allow that and will raise an exception.

Answered By: Juha

Values returned from Python function differ in caller and callee

Question:

Answers: