Planning for the Duke SSRI Julia Short Course, March 22, 2017.

Topics to be covered:

Overview: Why Julia? (John)

• Goals:
  • as easy to use as Python, R, or Matlab
  • within a factor of 2 of FORTRAN or C
• General-purpose language (web server, systems, etc.), but focused on technical computing
• Under what conditions should I consider switching to Julia?

Julia basics (John/Tyler)

1. Interactive usage (John)

   • running the REPL; running Jupyter and IJulia
   • getting help
     • ? mode

2. Basic data operations (Tyler)

   • matrices and vectors
   • linear algebra
   • comprehensions
   • it's okay to use loops
   • but it's also okay to vectorize (but be careful!)
   • defining functions (so we can do ... below)

3. Other data types (John)

   • tuple
     • tuples are collections of function arguments
     • splatting and slurping (...)
   • dict
     • introduction (for Matlab people)
     • type safety (for Python people)
     • cultural: keys are often symbols
   • set (time permitting)

Plotting things (John/Tyler)

• PyPlot (John)
• Winston (Tyler) [Winston is basically like Matlab's plotting syntax ... I'd like to intersperse some plotting with the DataFrames discussion, since this is a natural application. Maybe you can show PyPlot for more pure-math plotting applications, and I can show Winston / Gadfly for data applications?] — Sounds good. I will probably use this as an excuse to show off Python interoperability
• Gadfly (Tyler): like ggplot for Julia

Theory: how Julia makes your code fast (John)

1. Compilers and compilation
   • stages of compilation
2. Types and the type system
   • why types are important
   • grappling with type difficulties
   • the type hierarchy
     • super and subtypes
   • getting help: which, @which, methodswith, typeof
   • types vs objects
3. Multiple dispatch
   • multiple dispatch vs methods
   • type annotations: when you do and don't need them
4. Enhancing your code's performance
   • @time
   • garbage collection; memory management
   • Debugging
5. Macros & meta-programming
   • [This might be worth a brief mention, but nothing more]
   • if code is just another data structure, you can write code that writes code!

Other Topics (Tyler, with input from John on gotchas)

1. Language similarites / Best coding practices / avoiding "gotchas"
   • Put everything in functions so as to avoid Julia's default global scoping
   • adding type info to list comprehensions
   • Basic syntax comparisons with sister languages (Matlab, Python, R, etc.)
   • Other "gotchas" might already be discussed above with the compilation details section
   • see here for a more complete list
2. Importing datasets from other languages
   • Matlab/R/Stata/SPSS/SAS datasets
   • CSV files
3. DataFrames
   • How to use
   • How to convert to matrices and back
   • NA type
   • GLM package for basic data analysis
4. Optimization
   • JuliaOpt
   • JuMP
5. Distributions
   • distributions as objects
   • evaluating distributions
   • drawing from distributions
   • automated log transformations for MLE applications
   • maybe worth noting that StatsFuns.jl logpdfs are automatically differentiable by ForwardDiff.jl PR

Presentation order of jupyter notebook files

1. intro_slides.ipynb
2. Julia basics.ipynb
3. PlottingWithWinston.ipynb
4. PlottingWithGadfly.ipynb
5. SimilarityOtherLanguages.ipynb
6. DataImport.ipynb
7. DataFrames.ipynb
8. Regressions.ipynb
9. Nonlinear optimization with JuMP.ipynb
10. Distributions.ipynb