OK so if you haven’t read through the first balanced simulation post, I highly recommend you do, this one is going to be aimed at how we manage to expand the population size and still have a ‘run-able’ solution. That is, something that can possibly finish in our lifetimes :D.
To start with I finally did a Google search for large scale simulation software designs and algorithms that are bound to exist and found this white paper. Now, I won’t make you go through it, but I will break it down a little here before we dive into how I think we should move forward with our simulations.
The Breakdown
Alright so this paper doesn’t actually appear to be a ‘white paper’ in the proper sense. It is displayed on a white background though so… All kidding aside, I find its breakdown on different ways of handling the problem we face very useful. It all starts with agents and ‘macros’ or objects that represent the behavior of many agents. This isn’t exactly the problem we are having but at the very least defining a means by which to represent a population greater than say 1 million people is definitely useful to our end goal.
The article outlines 4 ways of handling simulations like these, each of which are capable of being viewed at both the micro(agent level) and macro(population level). They are as follows:
Zoom
In this architecture a macro object is composed of each of the micro agents upon which the micro agents are destroyed. When the micro level is chosen to be viewed we have a macro to micro function that creates a bunch of micro agents that describe the macro state after which the macro object is destroyed.
The good thing about this method is you can save memory by moving to a macro object as the population gets too large. Bad part is you lose all fidelity in the agents states because they are destroyed.
Puppeteer
This architecture differs from the Zoom method in a few ways
- The agents are never destroyed
- Only their update method is called (for example age might increase) no decisions are made by the agent (for example whether they ‘choose’ to get pregnant)
- The macro object is created and granted control over the agents decisions
- Hence the name Puppeteer
Memory could easily become an issue here as both the macro representation, and all of the agents state representation are present at any given time.
View
This architecture is directed at identifying emergent properties of a set of agents that might be behaving similarly, so in other words grouping by behavior. The grouping macro object(the ‘view’) never effects the decisions of the agents and only represents a group of them by their similar behaviors.
No memory savings here either — but definitely worth ear marking for future use if we expand this to a 2D or 3D simulation.
Cohabitation
Here is the most complex of all four. This one has macro and micro objects that bidirectionally act on each other — this one is used to represent complex simulations such as those in biological systems or social networks. Its hard to explain how two things could work to define each other when they are in themselves a part of the other, but if you think about it, that’s very representative of our experience of the world.
Where to next?
OK so we have some models of how we might handle many small agents, and macro-tize them to larger scales allowing us to handle large scale simulations. But, which do we choose, and where do we begin? It’s a good question, and I think it starts with finally exploring database usage. Kind of out of left field right? Yeah I agree, but I think we need to know if we get better performance by utilizing queries in a DB. Lets find out!
I agree, we shouldn’t rebuild everything to include database usage, lets just build a toy problem with a single object that can be represented by an object relational mapping(ORM), like SQLAlchemy, and see which one goes slower — object lists, or databases.
…the results are laughable…
OK, first off, as I started implementing this I knew the winner was going to be clear, but even then I just had to know.
To explain, each cycle(x-axis) adds 1000 people to our population and measures the time to update the entire population in seconds(y-axis). If you are curious here is their respective code blerbs.
# -*- coding: utf-8 -*-
"""
Created on Tue Jan 18 17:13:13 2022
This script is intended to time lists for a simple simulation use case.
@author: Travis Adsitt
"""
import matplotlib.pyplot as plt
import time
# Person to handle the most simple update method
class Person:
def __init__(self, age, alive):
self.age = age
self.alive = alive
def update(self):
if self.alive:
self.age += 1
if __name__ == "__main__":
# Variables to keep track of things
total_time = 0
populations = []
cycle_times = []
total_cycles = [cycle for cycle in range(0,100)]
curr_population = []
for cycle in total_cycles:
# Start timeing the cycle
start_time = time.time()
# Add 1000 people to our population
curr_population.extend([Person(0, True) for i in range(0,1000)])
# Select all the alive ones
for i in curr_population:
i.update()
# Calculate our measurements
cycle_time = time.time() - start_time
total_time += cycle_time
# Not the way you would want to do it if people ever died, but
# works for our testing here.
total_pop = len(curr_population)
populations.append(total_pop)
cycle_times.append(cycle_time)
# Print something so we know it's not dead
print(f"Cycle Time = {cycle_time}, Population = {total_pop}")
print(f"Total Time = {total_time}")
# plt.plot(total_cycles, populations, label="Population")
plt.plot(total_cycles, cycle_times, label="Cycle Time")
plt.legend()
# -*- coding: utf-8 -*-
"""
Created on Tue Jan 18 17:13:13 2022
This script is intended to time sql ORM for a simple simulation use case.
@author: Travis Adsitt
"""
from sqlalchemy import create_engine, Boolean, Column, Integer, String, ForeignKey, select, func
from sqlalchemy.orm import declarative_base, sessionmaker
import matplotlib.pyplot as plt
import time
import logging
# declarative base class
Base = declarative_base()
# Person to handle the most simple update method
class Person(Base):
__tablename__ = 'person'
id = Column(Integer, primary_key=True)
age = Column(Integer)
alive = Column(Boolean)
def update(self):
if self.alive:
self.age += 1
if __name__ == "__main__":
# Create the db engine
engine = create_engine("sqlite+pysqlite:///:memory:", echo=False, future=True)
Base.metadata.create_all(engine)
# Create a session
session = sessionmaker(engine)()
# Disable all logging on sqlalchemy
logging.basicConfig()
logging.getLogger('sqlalchemy').setLevel(logging.ERROR)
# Variables to keep track of things
total_time = 0
populations = []
cycle_times = []
total_cycles = [cycle for cycle in range(0,100)]
for cycle in total_cycles:
# Start timeing the cycle
start_time = time.time()
# Add 1000 people to our population
session.bulk_save_objects(
[
Person(id=None, age=0, alive=True) for i in range(0,1000)
]
)
# Select all the alive ones
for i in session.execute(select(Person).filter_by(alive=True)).scalars():
i.update()
# Commit our changes
session.commit()
# Calculate our measurements
cycle_time = time.time() - start_time
total_time += cycle_time
# Not the way you would want to do it if people ever died, but
# works for our testing here.
total_pop = session.execute(session.query(func.count(Person.alive))).scalar_one()
populations.append(total_pop)
cycle_times.append(cycle_time)
# Print something so we know it's not dead
print(f"Cycle Time = {cycle_time}, Population = {total_pop}")
print(f"Total Time = {total_time}")
# plt.plot(total_cycles, populations, label="Population")
plt.plot(total_cycles, cycle_times, label="Cycle Time")
plt.legend()
I guess the bright side is, now I remember how to implement basic ORM with SQLAlchemy, which will be useful for data storage :). I think there is a way that you might be able to speed this up — but it is inescapable that SQL look ups have just too much overhead to even come close to competing. Maybe at a certain scale the query selects could be a bit more efficient, and paired with multi-thread/process you could get through the list quicker.
Anyways, that’s it for tonight — hope to make more progress in the region of macro/micro(agent) creation tomorrow, was fun exploring this tonight.
END DAY 1
OK so it has been a few days since I worked on this, I am moving to a new company in my day job so it’s been a bit difficult to context switch to personal projects such as this. Buuut I think I have a little time to work this up further, let me just quickly remind myself where we left off. Ah that’s right, Database management doesn’t make sense for this — and we need to refocus our attention to the multi-agent system model we want to implement.
I think that it would be best to use the ‘View’ model for a couple of reasons:
- Basically everything we are doing here can be broken into probabilities
- There are ‘easy’ methods we could write to translate between Macro and Micro forms
Alright then, lets start looking at each of the life-time of our agents and converting them to probabilities that can be managed by simple population integers.
Approaching it from top down changes the design greatly, first of all we can look at each persons lifetime and split it into three time spans basically:
- Youth
- This is when the person is unable to have children
- Adulthood
- This is the span of time (192-420) where a person can have children
- Elderly
- This is when the person is too old to have children but is still alive(420-876)
Splitting it into these sections is useful because as far as I can tell we should be able to use lists of different events to calculate when and how many people are pregnant or having children. Same goes for alive and dead, it will be easier to explain possibly after I have written the code (also, it should prove if this is plausible). Toooo the code!
END DAY 2
OK, OK, I know, I didn’t post the code yesterday — things came up. Lets get it together, today I post the code for a ‘macro’ view of our population 🙂
import random
from threading import Thread
from enum import Enum
import matplotlib.pyplot as plt
# Simulation Variables
START_POPULATION = 1000
RUN_TIME = 10000
MALE_LIFE_SPAN = 876
FEMALE_LIFE_SPAN = 876
MALE_ADULT_AGE = 192
MALE_END_REPRODUCTION_AGE = MALE_LIFE_SPAN
FEMALE_ADULT_AGE = 192
FEMALE_END_REPRODUCTION_AGE = 420
GESTATION_LENGTH = 9
NUM_SAMPLES = 1000
RENDERED_SAMPLES = 100
# Host variables
MAX_THREADS = 15
class Gender(Enum):
Male=0
Female=1
def chunked_list_generator(l, chunks=1):
"""
Helper function to chunk a list into 'chunks' pieces, using the Knuth
algorithm already present in the random.shuffle method of python
Parameters
----------
l : list
The list to chunk up.
chunks : int, optional
The number of chunks to split the list into. The default is 1.
Yields
------
y_list : list
Each chunk as they are produced.
"""
# Get our list length and the number of items to place in each
len_list = len(l)
number_per_chunk = int(len_list / chunks)
# First iterator to go chunk by chunk
for curr_start in range(0, len_list, number_per_chunk):
# y_list = []
chunk_end = curr_start + number_per_chunk
yield (curr_start, chunk_end)
def get_new_population(num):
"""
Easy way to yield a random set of people
Parameters
----------
num : int
The number of people to create.
Yields
-------
lists : tuple(int,int)
The male and female population
"""
new_males = 0
new_females = 0
for p in range(0, num):
if bool(random.getrandbits(1)):
new_males += 1
else:
new_females += 1
return (new_males, new_females)
def get_viable_repo_population(pop_list, start, end):
if(len(pop_list) <= start):
return 0
gross_viable_pop = pop_list[-start]
if(len(pop_list) <= end):
return gross_viable_pop
else:
return gross_viable_pop - pop_list[-end]
def get_num_newly_pregnant(male_pop_list, female_pop_list, pregnant_pop_list):
"""
Get the number of newly pregnant females
Parameters
----------
male_pop : int
The number of 'old enough to have children' males.
female_pop : int
The number of 'old enough to have children' females.
pregnant_pop : int
The currently pregnant population.
Returns
-------
int
The number to 'impregnate'.
"""
# Take the current population of females, and subtract it from the total
# number of pregnant females
viable_females = get_viable_repo_population(
female_pop_list,
FEMALE_ADULT_AGE,
FEMALE_END_REPRODUCTION_AGE
) - sum(pregnant_pop_list[-GESTATION_LENGTH:])
viable_males = get_viable_repo_population(
male_pop_list,
MALE_ADULT_AGE,
MALE_END_REPRODUCTION_AGE
)
total_pop = viable_females + viable_males
new_preg_pop = 0
if viable_females < 1 or viable_males < 1:
return 0
people = []
while(viable_females > 0 and total_pop > 0):
is_female_prob = (viable_females / total_pop) * 100
rand_int = random.randint(0, 100)
# Get the gender of our new random person
if(rand_int < is_female_prob):
# Female
people.append(Gender.Female.value)
viable_females -= 1
else:
# Male
people.append(Gender.Male.value)
viable_males -= 1
# If we have two people check if there is a new pregnancy
if(len(people) == 2):
# 0 + 1 == pregnancy
if sum(people) == 1:
new_preg_pop += 1
people = []
total_pop = viable_females + viable_males
return new_preg_pop
def get_num_dead(male_pop_list, female_pop_list):
dead_males = 0
dead_females = 0
if(len(male_pop_list) > MALE_LIFE_SPAN and male_pop_list[-1] > 0):
dead_males = male_pop_list[-MALE_LIFE_SPAN]
if(len(female_pop_list) > FEMALE_LIFE_SPAN and female_pop_list[-1] > 0):
dead_females = female_pop_list[-FEMALE_LIFE_SPAN]
return (dead_males, dead_females)
def get_num_born(pregnant_pop_list):
if(len(pregnant_pop_list) < GESTATION_LENGTH):
return (0, 0)
birthing = pregnant_pop_list[-GESTATION_LENGTH]
males_birthed = 0
females_birthed = 0
for p in range(0, birthing):
male_female_rand = random.getrandbits(1)
if bool(male_female_rand):
males_birthed += 1
else:
females_birthed += 1
return (males_birthed, females_birthed)
def cull_list(l, max_len):
return l[-max_len:]
if __name__ == "__main__":
# These lists should never be longer than 876
male_population = []
female_population = []
time_list = [0]
# This list could be as short as 9
pregnant_population = [0]
# This list can be any length
dead_population = [0]
male, female = get_new_population(START_POPULATION)
male_population.append(male)
female_population.append(female)
for month in range(1, RUN_TIME):
new_males, new_females = get_num_born(pregnant_population)
dead_males, dead_females = get_num_dead(male_population, female_population)
newly_pregnant = get_num_newly_pregnant(male_population, female_population, pregnant_population)
curr_male_pop = (male_population[-1] + new_males) - dead_males
curr_female_pop = (female_population[-1] + new_females) - dead_females
male_population.append(curr_male_pop)
female_population.append(curr_female_pop)
pregnant_population.append(newly_pregnant)
time_list.append(month)
male_population = cull_list(male_population, NUM_SAMPLES)
female_population = cull_list(female_population, NUM_SAMPLES)
pregnant_population = cull_list(pregnant_population, NUM_SAMPLES)
time_list = cull_list(time_list, NUM_SAMPLES)
plt.plot(time_list[-RENDERED_SAMPLES:], male_population[-RENDERED_SAMPLES:], label="Male Population")
plt.plot(time_list[-RENDERED_SAMPLES:], female_population[-RENDERED_SAMPLES:], label="Female Population")
plt.plot(time_list[-RENDERED_SAMPLES:], pregnant_population[-RENDERED_SAMPLES:], label="Pregnant Population")
plt.legend()
plt.draw()
plt.pause(0.05)
OK, so this is just a snapshot of the code before I complete the multi-threaded version… This thing flies and can handle far larger populations (it was ticking slowly with more than 25 million people WITHOUT multi-threading).
So, multi-threading still hasn’t saved us, I still cant get a long enough run for us to see how the whole thing balances out eventually — more thinking to be done. Either way, it was a very fun thought exercise to run through. Below is the final code for this week, along with a moving graph (you’ll get to see my fancy new ‘render only 100 months’ graphing technique). Thank you for reading this week!
import random
from threading import Thread
from enum import Enum
import matplotlib.pyplot as plt
from queue import Queue
# Simulation Variables
START_POPULATION = 1000
RUN_TIME = 10000
MALE_LIFE_SPAN = 876
FEMALE_LIFE_SPAN = 876
MALE_ADULT_AGE = 192
MALE_END_REPRODUCTION_AGE = MALE_LIFE_SPAN
FEMALE_ADULT_AGE = 192
FEMALE_END_REPRODUCTION_AGE = 420
GESTATION_LENGTH = 9
NUM_SAMPLES = 1000
RENDERED_SAMPLES = 100
# Host variables
MAX_THREADS = 15
POPULATION_PER_THREAD = 10000
class Gender(Enum):
Male=0
Female=1
def chunked_list_generator(l, chunks=1):
"""
Helper function to chunk a list into 'chunks' pieces, using the Knuth
algorithm already present in the random.shuffle method of python
Parameters
----------
l : list
The list to chunk up.
chunks : int, optional
The number of chunks to split the list into. The default is 1.
Yields
------
y_list : list
Each chunk as they are produced.
"""
# Get our list length and the number of items to place in each
len_list = len(l)
number_per_chunk = int(len_list / chunks)
# First iterator to go chunk by chunk
for curr_start in range(0, len_list, number_per_chunk):
# y_list = []
chunk_end = curr_start + number_per_chunk
yield (curr_start, chunk_end)
def get_new_population(num):
"""
Easy way to yield a random set of people
Parameters
----------
num : int
The number of people to create.
Yields
-------
lists : tuple(int,int)
The male and female population
"""
new_males = 0
new_females = 0
for p in range(0, num):
if bool(random.getrandbits(1)):
new_males += 1
else:
new_females += 1
return (new_males, new_females)
def get_viable_repo_population(pop_list, start, end):
"""
A method to get the number of repoductively viable population given a start
age and an end age. This is mostly used for the male and female populations for
reproductions.
Parameters
----------
pop_list : list(int)
A population list of integers.
start : int
Start age.
end : int
End age.
Returns
-------
int
Viable population.
"""
if(len(pop_list) <= start):
return 0
gross_viable_pop = pop_list[-start]
if(len(pop_list) <= end):
return gross_viable_pop
else:
return gross_viable_pop - pop_list[-end]
def threaded_pregnancy(male_pop, female_pop, pregnant_pop, queue=None):
"""
A method to handle one threads worth of work given the male population
female population and pregnant population
Parameters
----------
male_pop : int
Number of males in the population.
female_pop : int
Number of females in the population.
pregnant_pop : int
Number of pregnancies currently.
queue : Queue, optional
A queue to place our results, if none will just return results assuming
single threaded. The default is None.
Returns
-------
new_preg_pop : int
Number new pregnancies.
"""
viable_females = female_pop
viable_males = male_pop
total_pop = male_pop + viable_females
new_preg_pop = 0
people = []
while(viable_females > 0 and total_pop > 0):
is_female_prob = (viable_females / total_pop) * 100
rand_int = random.randint(0, 100)
# Get the gender of our new random person
if(rand_int < is_female_prob):
# Female
people.append(Gender.Female.value)
viable_females -= 1
else:
# Male
people.append(Gender.Male.value)
viable_males -= 1
# If we have two people check if there is a new pregnancy
if(len(people) == 2):
# 0 + 1 == pregnancy
if sum(people) == 1:
new_preg_pop += 1
people = []
total_pop = viable_females + viable_males
if not queue:
return new_preg_pop
else:
queue.put(new_preg_pop)
def get_viable_population_thread_count(population_count):
"""
A helper function to get the thread count given a population size
Parameters
----------
population_count : int
The population size.
Returns
-------
int
Total number of threads that should be used.
"""
calc_threads = int(population_count / POPULATION_PER_THREAD) + 1
return calc_threads if calc_threads < MAX_THREADS else MAX_THREADS
def get_num_newly_pregnant(male_pop_list, female_pop_list, pregnant_pop_list):
"""
Get the number of newly pregnant females
Parameters
----------
male_pop : int
The number of 'old enough to have children' males.
female_pop : int
The number of 'old enough to have children' females.
pregnant_pop : int
The currently pregnant population.
Returns
-------
int
The number to 'impregnate'.
"""
# Take the current population of females, and subtract it from the total
# number of pregnant females
prev_preg_count = sum(pregnant_pop_list[-GESTATION_LENGTH:])
viable_females = get_viable_repo_population(
female_pop_list,
FEMALE_ADULT_AGE,
FEMALE_END_REPRODUCTION_AGE
) - prev_preg_count
viable_males = get_viable_repo_population(
male_pop_list,
MALE_ADULT_AGE,
MALE_END_REPRODUCTION_AGE
)
total_pop = viable_females + viable_males
new_preg_pop = 0
if viable_females < 1 or viable_males < 1:
return 0
threads = get_viable_population_thread_count(total_pop)
# print(threads)
if threads == 1:
new_preg_pop = threaded_pregnancy(
viable_males,
viable_females,
pregnant_pop_list
)
else:
single_threaded_females = int(viable_females % threads)
single_threaded_males = int(viable_males % threads)
single_threaded_preg = int(prev_preg_count % threads)
single_threaded_preg = threaded_pregnancy(
single_threaded_males,
single_threaded_females,
single_threaded_preg
)
males_per_thread = int(viable_males / threads)
females_per_thread = int((viable_females - single_threaded_preg) / threads)
preg_per_thread = int((single_threaded_preg + prev_preg_count) / threads)
thread_objs = []
thread_queue = Queue()
for t in range(0, threads):
thread_objs.append(Thread(
target=threaded_pregnancy,
args=(males_per_thread, females_per_thread, preg_per_thread, thread_queue, )
))
thread_objs[-1].start()
for t in thread_objs:
t.join()
new_preg = sum([thread_queue.get_nowait() for i in range(0, thread_queue.qsize())])
new_preg_pop = single_threaded_preg + new_preg
return new_preg_pop
def get_num_dead(male_pop_list, female_pop_list):
"""
A method to get the total dead this month
Parameters
----------
male_pop_list : list(int)
Population list for men.
female_pop_list : list(int)
Population list for females.
Returns
-------
dead_males : int
Number of dead males.
dead_females : int
Number of dead females.
"""
dead_males = 0
dead_females = 0
if(len(male_pop_list) > MALE_LIFE_SPAN and male_pop_list[-1] > 0):
dead_males = male_pop_list[-MALE_LIFE_SPAN]
if(len(female_pop_list) > FEMALE_LIFE_SPAN and female_pop_list[-1] > 0):
dead_females = female_pop_list[-FEMALE_LIFE_SPAN]
return (dead_males, dead_females)
def threaded_birth(num_to_birth, male_queue=None, female_queue=None):
"""
A method for one threads work for birth.
Parameters
----------
num_to_birth : int
Number of new people.
male_queue : Queue, optional
A queue to dump our number of new boys. The default is None.
female_queue : Queue, optional
A queue to dump our number of new girls. The default is None.
Returns
-------
males_birthed : int
Number of males birthed.
females_birthed : int
Number of females birthed.
"""
males_birthed = 0
females_birthed = 0
for p in range(0, num_to_birth):
male_female_rand = random.getrandbits(1)
if bool(male_female_rand):
males_birthed += 1
else:
females_birthed += 1
if not male_queue:
return (males_birthed, females_birthed)
else:
male_queue.put(males_birthed)
female_queue.put(females_birthed)
def get_num_born(pregnant_pop_list):
"""
A method attempt to auto-multi-thread the birthing process. Multi-thread
currently commented out due to strange behavior at high numbers.
Parameters
----------
pregnant_pop_list : list(int)
A population list of pregnancies.
Returns
-------
int
Number of males born.
int
Number of females born.
"""
if(len(pregnant_pop_list) < GESTATION_LENGTH):
return (0, 0)
birthing = pregnant_pop_list[-GESTATION_LENGTH]
# if birthing < POPULATION_PER_THREAD:
return threaded_birth(birthing)
# else:
# threads = get_viable_population_thread_count(birthing)
# thread_objs = []
#
# male_queue = Queue()
# female_queue = Queue()
#
# single_thread_num = int(birthing % threads)
# males, females = threaded_birth(single_thread_num)
#
# thread_num = int(birthing / threads)
# for t in range(0, threads - 1):
# thread_objs.append(Thread(
# target=threaded_birth, args=(
# thread_num,
# male_queue,
# female_queue
# )
# ))
# thread_objs[-1].start()
#
# for t in thread_objs:
# t.join()
#
# new_males = sum([male_queue.get_nowait() for i in range(0, male_queue.qsize())])
# new_females = sum([female_queue.get_nowait() for i in range(0, female_queue.qsize())])
# return (males + new_males, females + new_females)
def cull_list(l, max_len):
"""
A method to shorten a list from the front
Parameters
----------
l : TYPE
DESCRIPTION.
max_len : TYPE
DESCRIPTION.
Returns
-------
TYPE
DESCRIPTION.
"""
return l[-max_len:]
if __name__ == "__main__":
# These lists should never be longer than 876
male_population = []
female_population = []
time_list = [0]
# This list could be as short as 9
pregnant_population = [0]
# This list can be any length
dead_population = [0]
male, female = get_new_population(START_POPULATION)
male_population.append(male)
female_population.append(female)
for month in range(1, RUN_TIME):
# Get number born and dead
new_males, new_females = get_num_born(pregnant_population)
dead_males, dead_females = get_num_dead(male_population, female_population)
# Get newly pregnant and current male and female populations
newly_pregnant = get_num_newly_pregnant(male_population, female_population, pregnant_population)
curr_male_pop = (male_population[-1] + new_males) - dead_males
curr_female_pop = (female_population[-1] + new_females) - dead_females
# Append new data to appropriate lists
male_population.append(curr_male_pop)
female_population.append(curr_female_pop)
pregnant_population.append(newly_pregnant)
dead_population.append(dead_males + dead_females)
time_list.append(month)
# Cull our lists
male_population = cull_list(male_population, NUM_SAMPLES)
female_population = cull_list(female_population, NUM_SAMPLES)
pregnant_population = cull_list(pregnant_population, NUM_SAMPLES)
dead_population = cull_list(dead_population, NUM_SAMPLES)
time_list = cull_list(time_list, NUM_SAMPLES)
# Plot things
plt.plot(time_list[-RENDERED_SAMPLES:], male_population[-RENDERED_SAMPLES:], label="Male Population")
plt.plot(time_list[-RENDERED_SAMPLES:], female_population[-RENDERED_SAMPLES:], label="Female Population")
plt.plot(time_list[-RENDERED_SAMPLES:], pregnant_population[-RENDERED_SAMPLES:], label="Pregnant Population")
# plt.plot(time_list[-RENDERED_SAMPLES:], dead_population[-RENDERED_SAMPLES:], label="Dead Population")
plt.legend()
plt.draw()
plt.pause(0.05)