Simple demo code for fast array replacement (with performance comparison)

replace_demo.py

#!/usr/bin/env python
from __future__ import print_function
import argparse
import numpy as np
import time

# a random sequence
arr = np.random.choice(['a', 'b', 'c', 'd'], 5000, replace=True)
positive_values = ['a', 'b']

def main(repeat_times=10):
    result_naive = np.array(naive_impl(arr, positive_values))
    result_np = np_impl(arr, positive_values)
    print("produce same results: {}".format(np.alltrue(result_np == result_naive)))

    exec_durations = []
    for _ in range(repeat_times):
        tic = time.time()
        _ = naive_impl(arr, positive_values)
        toc = time.time()
        exec_durations.append(toc-tic)
    print("mean execution time for naive_impl: {} sec".format(np.mean(exec_durations)))

    exec_durations = []
    for _ in range(repeat_times):
        tic = time.time()
        _ = np_impl(arr, positive_values)
        toc = time.time()
        exec_durations.append(toc-tic)
    print("mean execution time for np_impl: {} sec".format(np.mean(exec_durations)))
    return 0

def naive_impl(arr, pos_values):
    ret_arr = []
    for value in arr:
        if value in pos_values:
            ret_arr.append('positive')
        else:
            ret_arr.append('negative')
    return ret_arr

def np_impl(arr, pos_values):
    return np.where(np.isin(arr, pos_values), ['positive'], ['negative'])


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("-n", "--repeat-times", dest="repeat_times",
                        help="repeat times for profiling",
                        type=int, default=10)
    args = vars(parser.parse_args())
    main(**args)

If we make arr in the line 8 a list, there will be no significant performance boost in both implementation.
Here we can see that numpy array is a good substitution to list if you are handling homogeneous array.