parallelism.py

import time
import os
from multiprocessing import Pool
import pandas as pd
import utility


def multi_start(fun, starting_points, n_cpu, timeout=2000, i_start=0,):
    """Conduct multi_start with parallel computing
       fun is the function to implement computation
       starting_points is the list of all the starting points
       n_cpu is the number of cpu cores
       timeout is the maximum time allocated to each core
    """
    # Work on the copied starting_points
    starting_points = starting_points.copy()

    sol_pool = []
    i_end = i_start + len(starting_points)

    t0 = time.time()
    for k in range(len(starting_points)):
        print(f'Current start_i is {i_start}')

        pool = Pool(n_cpu)
        results = [pool.apply_async(fun, (init,)) for init in starting_points]

        for i, res in enumerate(results):
            i_global = i_start + i
            try:
                res.get(timeout=timeout)
            except:
                pool.terminate()
                print(f'The cpu time at initial point {i_global} exceeded {timeout}.')

                # Delete the starting point i and result i
                starting_points.pop(i)
                results.pop(i)
                break

        # Collect all the available solution
        pop_item = []
        for i, res in enumerate(results):
            if res.ready():
                sol_pool.append(res.get())
                pop_item.append(i)

        # Update starting_points (Remove unachievable point and successful points)
        starting_points = [init for i, init in enumerate(starting_points) if i not in pop_item]
        results = [res for i, res in enumerate(starting_points) if i not in pop_item]
        i_start = i_end - len(starting_points)

        if len(results) == 0:
            break

    cpu_time = time.time() - t0

    # Sort solutions
    sol_pool = utility.sort_solution(sol_pool)
    sol_best = sol_pool[0]

    return sol_best, sol_pool, cpu_time


def export_sol(sol_pool, prop, directory=''):
    if directory is None:
        directory = ''

    sol_prop_pool = [s[prop].flatten() for s in sol_pool]

    file_name = f'sol_{prop}_pool.xlsx'
    save_path = os.path.join(directory, file_name)
    pd.DataFrame(sol_prop_pool).to_excel(save_path)