/bioinfo_benchmarks

Language benchmarks that are important for Bioinformatics scripting

Primary LanguageShell

Some Dubious and Pragmatic Benchmarks

This repo contains code for evaluating languages to use for string heavy programs (bioinformatics).

The benchmarks are not trying to find the fastest way to solve the problem of count the number of fields in the file, they are just meant to demonstrate the following:

  1. Scriptability - How easy is it to read from stdin and split lines?
  2. IO - Is there something inherently slow about how the language deals with IO?
  3. String Ops - Are the builtins for doing string manipulations efficient, or do they just look pretty?
  4. Lightweight Classes - can we cheaply create a class to hold data?
  5. Array and String allocations - Does the compiler / interpreter optimize allocations?
  6. (Not yet done) C binding - How easy is it to use a c library?
  7. (Not yet done) C interop - Is there a large cost to interacting with C libraries?

Psuedocode

Input is a tab delimted file. The Python script may be easier to read :)

object Record:
    name: the first field in the line
    count: the number of fields that have 'bc' between chars 1:4

def create_record(list of fields):
    count = numer of fields that contain 'bc', case insensitive, betwen chars 1:4
    return Record object reflecting the passed in line

def main:
    read lines from stdin
        split line on '\t'
        send line to create_record function

    print the sum or the counts of each record

Results

Lang Time
Python3 0m8.201s
Perl 0m8.033s
D - dmd 0m10.229s
D - ldc 0m8.689s
Nim 0m5.901
  • See Implementation notes

Language version for results

Python 3.6 Perl 5.26 Nim 0.19 Dlang (dmd) v2.085.1 Dlang (ldc2) 1.15.0

Implementation Notes

I chose to use a class in Python because in a real life scenario, I would create a dataclass to take advantage of mypy / the type system. Classes are a little bit slower though. In Perl, I would likely not have used a class, since there is no real type benefit to doing so. I used Class::XSAccessor here just level the playing field. If dicts are uses in the Perl and Python versions, they both go to about 7.5s, with Perl having a slight edge.