Building compiled programs

1. Brain-storm in a dynamic language for algorithm exploration and testing.
2. Deliver a performant final-version in a compiled language.

Using dynamic languages

Does a package/library do exactly
what you need?
             ├── Yes:   Great!
             ├── Dunno: You need to read C/++ code.
             └── No:    You need to code in C/++.

What if the dynamic language was fast enough?

• Benchmarks
• Solves the two-language problem
• Julia is mostly coded in Julia!

LOC versus speed

Syntax

2π√3/5α₀

2*np.pi*np.sqrt(3)/(5*alpha0)

2π*√3/5α₀

Custom infix operators

# rotate coordinate `c` by `θ` radians
julia> ↺(c, θ) = [cos(θ) -sin(θ); sin(θ) cos(θ)]*c
↺ (generic function with 1 method)

julia> [1,0] ↺ π/2
2-element Array{Float64,1}:
 0.0
 1.0

Embedded units

It took a beetle 36 seconds to walk 25 cm. How many days will it take it to walk 3 km?

using Unitful:uconvert, cm, km, s, d
v = 25cm/36s
t = 3km/v
uconvert(d, t)

5 days

Some of my favourite things #1

• QuantEcon: Provide lectures on models including code!
  • QuantEcon.jl, a package with most of the functionality that you need, e.g. CompEcon.jl.
• Parameters.jl: Use parameters more comfortably.
• Queryverse.jl: Like R's tidyverse? There is something similar in Julia!

Some of my favourite things #2

• Macros aka meta programming:
  • Want to know how fast your function is? Use @btime foo(), it runs your function many times averaging durations. Also information about allocation of memory.
  • How to make your code faster? @trace
  • Want to see the machine code? @code_native

Some of my favourite things #3

• github integration: your code and its documentation accessible to everyone.
  • You can view the code of packages you are using and read it (most likely written in Julia)!
• Makes it easy to contribute yourself. File issues with developers directly, bugs are quickly fixed.
• Discuss your problems with the growing Julia community on discourse.

Too new!

• Ecosystems (e.g. packages, IDEs, debugger) not as mature as in other environments.
• Some of the more specialized libraries are missing.
• Harder to Google for answers.

Still working out all the kinks

• Loading some packages is still a bit slow.
• Plotting works but hasn't settled yet.
• Data analysis not yet as convienent as in R.
• Transitioning into v1.0.0: Major changes.

General purpose

• Need statistics? R
• Need vectorized operations on matrices? MATLAB
• Julia (like Python) is more general-purpose → can be harder to find what you're looking for.
  • But if you can do it all in Julia - that's nice.