In [1]:
# If you want the figures to appear in the notebook,
# and you want to interact with them, use
# %matplotlib notebook
# If you want the figures to appear in the notebook,
# and you don't want to interact with them, use
# %matplotlib inline
# If you want the figures to appear in separate windows, use
# %matplotlib qt5
# To switch from one to another, you have to select Kernel->Restart
%matplotlib inline
from modsim import *
This notebook develops a simple growth model, like the ones in Chapter 3, and uses it to demonstrate a parameter sweep.
The system we'll model is a rabbit farm. Suppose you start with an initial population of rabbits and let them breed. For simplicity, we'll assume that all rabbits are on the same breeding cycle, and we'll measure time in "seasons", where a season is the reproductive time of a rabbit.
If we provide all the food, space and other resources a rabbit might need, we expect the number of new rabbits each season to be proportional to the current population, controlled by a parameter, birth_rate, that represents the number of new rabbits per existing rabbit, per season. As a starting place, I'll assume birth_rate = 0.9.
Sadly, during each season, some proportion of the rabbits die. In a detailed model, we might keep track of each rabbit's age, because the chance of dying is probably highest for young and old rabbits, and lowest in between. But for simplicity, we'll model the death process with a single parameter, death_rate, that represent the number
of deaths per rabbit per season. As a starting place, I'll assume death_rate = 0.5.
Here's a system object that contains these parameters as well as:
p0,t0, andt_end, measured in seasons.
In [2]:
system = System(t0 = 0,
t_end = 10,
p0 = 10,
birth_rate = 0.9,
death_rate = 0.5)
system
Out[2]:
Here's a version of run_simulation, similar to the one in Chapter 3, with both births and deaths proportional to the current population.
In [3]:
def run_simulation(system):
"""Runs a proportional growth model.
Adds TimeSeries to `system` as `results`.
system: System object with t0, t_end, p0,
birth_rate and death_rate
"""
results = TimeSeries()
results[system.t0] = system.p0
for t in linrange(system.t0, system.t_end):
births = system.birth_rate * results[t]
deaths = system.death_rate * results[t]
results[t+1] = results[t] + births - deaths
system.results = results
Now we can run the simulation and display the results:
In [4]:
run_simulation(system)
system.results
Out[4]:
Notice that the simulation actually runs one season past t_end. That's an off-by-one error that I'll fix later, but for now we don't really care.
The following function plots the results.
In [5]:
def plot_results(system, title=None):
"""Plot the estimates and the model.
system: System object with `results`
"""
newfig()
plot(system.results, 'bo', label='rabbits')
decorate(xlabel='Season',
ylabel='Rabbit population',
title=title)
And here's how we call it.
In [6]:
plot_results(system, title='Proportional growth model')
Let's suppose our goal is to maximize the number of rabbits, so the metric we care about is the final population. We can extract it from the results like this:
In [7]:
def final_population(system):
t_end = system.results.index[-1]
return system.results[t_end]
And call it like this:
In [8]:
final_population(system)
Out[8]:
To explore the effect of the parameters on the results, we'll define make_system, which takes the system parameters as function parameters(!) and returns a System object:
In [9]:
def make_system(birth_rate=0.9, death_rate=0.5):
system = System(t0 = 0,
t_end = 10,
p0 = 10,
birth_rate = birth_rate,
death_rate = death_rate)
return system
Now we can make a System, run a simulation, and extract a metric:
In [10]:
system = make_system()
run_simulation(system)
final_population(system)
Out[10]:
To see the relationship between birth_rate and final population, we'll define sweep_birth_rate:
In [11]:
def sweep_birth_rate(birth_rates, death_rate=0.5):
for birth_rate in birth_rates:
system = make_system(birth_rate=birth_rate,
death_rate=death_rate)
run_simulation(system)
p_end = final_population(system)
plot(birth_rate, p_end, 'gs', label='rabbits')
decorate(xlabel='Births per rabbit per season',
ylabel='Final population')
The first parameter of sweep_birth_rate is supposed to be an array; we can use linspace to make one.
In [12]:
birth_rates = linspace(0, 1, 21)
birth_rates
Out[12]:
Now we can call sweep_birth_rate.
The resulting figure shows the final population for a range of values of birth_rate.
Confusingly, the results from a parameter sweep sometimes resemble a time series. It is very important to remember the difference. One way to avoid confusion: LABEL THE AXES.
In the following figure, the x-axis is birth_rate, NOT TIME.
In [13]:
birth_rates = linspace(0, 1, 21)
sweep_birth_rate(birth_rates)
The code to sweep the death rate is similar.
In [14]:
def sweep_death_rate(death_rates, birth_rate=0.9):
for death_rate in death_rates:
system = make_system(birth_rate=birth_rate,
death_rate=death_rate)
run_simulation(system)
p_end = final_population(system)
plot(death_rate, p_end, 'r^', label='rabbits')
decorate(xlabel='Deaths per rabbit per season',
ylabel='Final population')
And here are the results. Again, the x-axis is death_rate, NOT TIME.
In [15]:
death_rates = linspace(0.1, 1, 20)
sweep_death_rate(death_rates)
In the previous sweeps, we hold one parameter constant and sweep the other.
You can also sweep more than one variable at a time, and plot multiple lines on a single axis.
To keep the figure from getting too cluttered, I'll reduce the number of values in birth_rates:
In [16]:
birth_rates = linspace(0.4, 1, 4)
birth_rates
Out[16]:
By putting one for loop inside another, we can enumerate all pairs of values.
The results show 4 lines, one for each value of birth_rate.
(I did not plot the lines between the data points because of a limitation in plot.)
In [17]:
for birth_rate in birth_rates:
for death_rate in death_rates:
system = make_system(birth_rate=birth_rate,
death_rate=death_rate)
run_simulation(system)
p_end = final_population(system)
plot(death_rate, p_end, 'c^', label='rabbits')
decorate(xlabel='Deaths per rabbit per season',
ylabel='Final population')
If you suspect that the results depend on the difference between birth_rate and death_rate, you could run the same loop, plotting the "net birth rate" on the x axis.
If you are right, the results will fall on a single curve, which means that knowing the difference is sufficient to predict the outcome; you don't actually have to know the two parameters separately.
In [18]:
for birth_rate in birth_rates:
for death_rate in death_rates:
system = make_system(birth_rate=birth_rate,
death_rate=death_rate)
run_simulation(system)
p_end = final_population(system)
net_birth_rate = birth_rate - death_rate
plot(net_birth_rate, p_end, 'mv', label='rabbits')
decorate(xlabel='Net births per rabbit per season',
ylabel='Final population')
On the other hand, if you guess that the results depend on the ratio of the parameters, rather than the difference, you could check by plotting the ratio on the x axis.
If the results don't fall on a single curve, that suggests that the ratio alone is not sufficient to predict the outcome.
In [19]:
for birth_rate in birth_rates:
for death_rate in death_rates:
system = make_system(birth_rate=birth_rate,
death_rate=death_rate)
run_simulation(system)
p_end = final_population(system)
birth_ratio = birth_rate / death_rate
plot(birth_ratio, p_end, 'y>', label='rabbits')
decorate(xlabel='Ratio of births to deaths',
ylabel='Final population')
In [ ]: