Implementing Pool on a for loop with a lot of inputs

问题

I have been trying to improve my code (with numba and multiprocessing), but I cannot quite get it, because my function has a lot of arguments.

I have already simplified it with other functions (see below)...

As each agent (a class instance) is independent of each other for these actions, I would like to replace the for with Pool.

So I would get a large function pooling() that I would call and pass the list of agents

from multiprocessing import Pool

p = Pool(4)
p.map(pooling, list(agents))

But, where do I ADD all the arguments that the pooling function will need?

As it is:

def check_demographics(month, my_agents, families, firms, year, mortality_men, mortality_women, fertility, state_id):

    dummy = list(my_agents)
    d = str(state_id.iloc[0])

# Place where I would like to replace the LOOP. All below would be a function 

    for agent in dummy:

        if agent.get_region_id()[:2] == d:

            # Brithday
            if month % 12 == agent.month - 1:
                agent.update_age()

            # Mortality probability 
            if agent.get_gender() == 'Male':
                prob = mortality_men[mortality_men['age'] == agent.get_age()][year].iloc[0]

            # When gender is Female
            else:
                # Extract specific agent data to calculate mortality 'Female'
                prob = mortality_women[mortality_women['age'] == agent.get_age()][year].iloc[0]

            # Give birth decision 
                age = agent.get_age()
                if 14 < age < 50:
                    pregnant(agent, fertility, year, families, my_agents)

            # Mortality procedures 
            if fixed_seed.random() < prob:
                mortal(my_agents, my_graveyard, families, agent, firms)

It is the most time consuming function in my programme. And @jit is not helping much.

Thanks a bunch

回答1:

Yes, there is a lot of parameters! Consider using a class.

Well, since Pool.map support only one iterable argument, you need to group everything in one place. I suggest you to use the "Facade" pattern: an intermediate class used to store all required parameters and having a single method (I call it check) without parameter (it's a method).

class Facade(object):
    def __init__(self, agent, d, families, fertility, firms, month, mortality_men, mortality_women, my_agents,
                 my_graveyard, year):
        self.agent = agent
        self.d = d
        self.families = families
        self.fertility = fertility
        self.firms = firms
        self.month = month
        self.mortality_men = mortality_men
        self.mortality_women = mortality_women
        self.my_agents = my_agents
        self.my_graveyard = my_graveyard
        self.year = year

    def check(self):
        (agent, d, families, fertility, firms,
         month, mortality_men, mortality_women,
         my_agents, my_graveyard, year) = (
            self.agent, self.d, self.families, self.fertility, self.firms,
            self.month, self.mortality_men, self.mortality_women,
            self.my_agents, self.my_graveyard, self.year)
        if agent.get_region_id()[:2] == d:

            # Brithday
            if month % 12 == agent.month - 1:
                agent.update_age()

            # Mortality probability
            if agent.get_gender() == 'Male':
                prob = mortality_men[mortality_men['age'] == agent.get_age()][year].iloc[0]

            # When gender is Female
            else:
                # Extract specific agent data to calculate mortality 'Female'
                prob = mortality_women[mortality_women['age'] == agent.get_age()][year].iloc[0]

                # Give birth decision
                age = agent.get_age()
                if 14 < age < 50:
                    pregnant(agent, fertility, year, families, my_agents)

            # Mortality procedures
            if fixed_seed.random() < prob:
                mortal(my_agents, my_graveyard, families, agent, firms)

Remark: my refactoring is really ugly, but I wanted to keep variable names unchanged for clarity.

Then your loop can be something like that:

def check_demographics(month, my_agents, families, firms,
                       year, mortality_men, mortality_women,
                       fertility, state_id, my_graveyard):
    d = str(state_id.iloc[0])
    pool = Pool(4)
    facades = [Facade(agent, d, families, fertility, firms,
                      month, mortality_men, mortality_women,
                      my_agents, my_graveyard, year)
               for agent in my_agents]
    pool.map(Facade.check, facades)

You said that each agent is independent of each other but, after analysing the loop, I see that you need the complete list of agents (the my_agents parameters). It's obvious with the Facade class. So your agent list must not change and the internal state of each agent must be frozen during looping.

来源：https://stackoverflow.com/questions/38212265/implementing-pool-on-a-for-loop-with-a-lot-of-inputs