Benchmark::Trend
Measure performance trends of Ruby code based on the input size distribution.
Benchmark::Trend will help you estimate the computational complexity of Ruby code by running it on inputs increasing in size, measuring their execution times, and then fitting these observations into a model that best predicts how a given Ruby code will scale as a function of growing workload.
Why?
Tests provide safety net that ensures your code works correctly. What you don't know is how fast your code is! How does it scale with different input sizes? Your code may have computational complexity that doesn't scale with large workloads. It would be good to know before your application goes into production, wouldn't it?
Benchmark::Trend will allow you to uncover performance bugs or confirm that a Ruby code performance scales as expected.
Installation
Add this line to your application's Gemfile:
gem 'benchmark-trend'
And then execute:
$ bundle
Or install it yourself as:
$ gem install benchmark-trend
Contents
1. Usage
Let's assume we would like to find out behaviour of a Fibonacci algorithm:
def fibonacci(n)
n < 2 ? n : fibonacci(n - 1) + fibonacci(n - 2)
end
To measure the actual complexity of above function, we will use infer_trend
method and pass it as a first argument an array of integer sizes and a block to execute the method:
numbers = Benchmark::Trend.range(1, 28, ratio: 2)
trend, trends = Benchmark::Trend.infer_trend(numbers) do |n, i|
fibonacci(n)
end
The return type will provide a best trend name:
print trend
# => exponential
and a Hash of all the trend data:
print trends
# =>
# {:exponential=>
# {:trend=>"1.38 * 0.00^x",
# :slope=>1.382889711685203,
# :intercept=>3.822775903539121e-06,
# :residual=>0.9052392775178072},
# :power=>
# {:trend=>"0.00 * x^2.11",
# :slope=>2.4911044372815657e-06,
# :intercept=>2.1138475434240918,
# :residual=>0.5623418036957115},
# :linear=>
# {:trend=>"0.00 + -0.01*x",
# :slope=>0.0028434594496586007,
# :intercept=>-0.01370769842204958,
# :residual=>0.7290365425188893},
# :logarithmic=>
# {:trend=>"0.02 + -0.02*ln(x)",
# :slope=>0.01738674709454521,
# :intercept=>-0.015489004560847924,
# :residual=>0.3982368125757882}}
You can see information for the best trend by passing name into trends hash:
print trends[trend]
# =>
# {:trend=>"1.38 * 0.00^x",
# :slope=>1.382889711685203,
# :intercept=>3.822775903539121e-06,
# :residual=>0.9052392775178072},
2. API
2.1 range
To generate a range of values for testing code fitness use the range
method. It will generate a geometric sequence of numbers, where intermediate values are powers of range multiplier, by default 8:
Benchmark::Trend.range(8, 8 << 10)
# => [8, 64, 512, 4096, 8192]
You can change the default sequence power by using :ratio
keyword:
Benchmark::Trend.range(8, 8 << 10, ratio: 2)
# => [8, 16, 32, 64, 128, 256, 512, 1024, 2048, 4096, 8192]
2.2 infer_trend
To calculate an asymptotic behaviour of Ruby code by inferring its computational complexity use infer_trend
. This method takes as an argument an array of inputs which can be generated using range. The code to measure needs to be provided inside a block. Two parameters are always yielded to a block, first, the actual data input and second the current index matching the input.
For example, let's assume you would like to find out asymptotic behaviour of a Fibonacci algorithm:
def fibonacci(n)
n < 2 ? n : fibonacci(n - 1) + fibonacci(n - 2)
end
You could start by generating a range of inputs in powers of 2:
numbers = Benchmark::Trend.range(1, 32, ratio: 2)
# => [1, 2, 4, 8, 16, 32]
Then measure the performance of the Fibonacci algorithm for each of the data points and fit the observations into a model to predict behaviour as a function of input size:
trend, trends = Benchmark::Trend.infer_trend(numbers) do |n, i|
fibonacci(n)
end
The return includes the best fit name:
print trend
# => exponential
And a Hash of all measurements:
print trends
# =>
# {:exponential=>
# {:trend=>"1.38 * 0.00^x",
# :slope=>1.382889711685203,
# :intercept=>3.822775903539121e-06,
# :residual=>0.9052392775178072},
# :power=>
# {:trend=>"0.00 * x^2.11",
# :slope=>2.4911044372815657e-06,
# :intercept=>2.1138475434240918,
# :residual=>0.5623418036957115},
# :linear=>
# {:trend=>"0.00 + -0.01*x",
# :slope=>0.0028434594496586007,
# :intercept=>-0.01370769842204958,
# :residual=>0.7290365425188893},
# :logarithmic=>
# {:trend=>"0.02 + -0.02*ln(x)",
# :slope=>0.01738674709454521,
# :intercept=>-0.015489004560847924,
# :residual=>0.3982368125757882}}
In order to retrieve trend data for the best fit do:
print trends[trend]
# =>
# {:trend=>"1.38 * 0.00^x",
# :slope=>1.382889711685203,
# :intercept=>3.822775903539121e-06,
# :residual=>0.9052392775178072}
2.2.1 repeat
To increase stability of you tests consider repeating all time execution measurements using :repeat
keyword.
Start by generating a range of inputs for your algorithm:
numbers = Benchmark::Trend.range(1, 32, ratio: 2)
# => [1, 2, 4, 8, 16, 32]
and then run your algorithm for each input repeating measurements 100
times:
Benchmark::Trend.infer_trend(numbers, repeat: 100) { |n, i| ... }
2.3 fit
Use fit
method if you wish to fit arbitrary data into a model with a slope and intercept parameters that minimize the error.
For example, given a set of data points that exhibit linear behaviour:
xs = [1, 2, 3, 4, 5]
ys = xs.map { |x| 3.0 * x + 1.0 }
Fit the data into a model:
slope, intercept, error = Benchmark::Trend.fit(xs, ys)
And printing the values we get confirmation of the linear behaviour of the data points:
print slope
# => 3.0
print intercept
# => 1.0
print error
# => 1.0
2.4 fit_at
If you are interested how a model scales for a given input use fit_at
. This method expects that there is a fit model generated using infer_trend.
For example, measuring Fibonacci recursive algorithm:
numbers = Benchmark::Trend.range(1, 28, ratio: 2)
trend, trends = Benchmark::Trend.infer_trend(numbers) do |n, i|
fibonacci(n)
end
We get the following results:
trends[trend]
# =>
# {:trend=>"1.38 * 0.00^x",
# :slope=>1.382889711685203,
# :intercept=>3.822775903539121e-06,
# :residual=>0.9052392775178072}
And checking model at input of 50
:
Benchamrk::Trend.fit_at(trend, n: 50, slope: trends[trend][:slope], intercept: trends[trend][:intercept])
# => 41.8558455915123
We can see that Fibonacci with just a number 50 will take around 42 seconds to get the result!
How about Fibonacci with 100 as an input?
Benchamrk::Trend.fit_at(trend, n: 100, slope: trends[trend][:slope], intercept: trends[trend][:intercept])
# => 458282633.9777338
This means Fibonacci recursive algorithm will take about 1.45 year to complete!
3. Examples
3.1 Ruby array max
Suppose you wish to find an asymptotic behaviour of Ruby built Array max
method.
You could start with generating a range of inputs:
array_sizes = Benchmark::Trend.range(1, 100_000)
# => [1, 8, 64, 512, 4096, 32768, 100000]
Next, based on the generated ranges create arrays containing randomly generated integers:
number_arrays = array_sizes.map { |n| Array.new(n) { rand(n) } }
Then feed this information to infer a trend:
trend, trends = Benchmark::Trend.infer_trend(array_sizes) do |n, i|
number_arrays[i].max
end
Unsurprisingly, we discover that Ruby's max
call scales linearily with the input size:
print trend
# => linear
We can also see from the residual value that this is a near perfect fit:
print trends[trend]
# =>
# {:trend=>"0.00 + 0.00*x",
# :slope=>5.873536409841244e-09,
# :intercept=>3.028647045635842e-05,
# :residual=>0.9986764704492359}
Development
After checking out the repo, run bin/setup
to install dependencies. Then, run rake spec
to run the tests. You can also run bin/console
for an interactive prompt that will allow you to experiment.
To install this gem onto your local machine, run bundle exec rake install
. To release a new version, update the version number in version.rb
, and then run bundle exec rake release
, which will create a git tag for the version, push git commits and tags, and push the .gem
file to rubygems.org.
Contributing
Bug reports and pull requests are welcome on GitHub at https://github.com/[USERNAME]/benchmark-trend. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the Contributor Covenant code of conduct.
License
The gem is available as open source under the terms of the MIT License.
Code of Conduct
Everyone interacting in the Benchmark::Trend project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.
Copyright
Copyright (c) 2018 Piotr Murach. See LICENSE for further details.