Julia

Julia was designed for scientific computing, to be easy to write and fast to run: it compiles each function the first time it’s called. Its data science tools, such as DataFrames.jl, CSV.jl, and GLM.jl, are packages installed with Julia’s package manager, Pkg. None is installed here, and the tests run without a network, so only Base and the standard library are available. That library already covers a lot: the backslash operator solves least squares, and Statistics has mean, median, var, std, and cor. The from-scratch versions skip them so the arithmetic stays visible.

Quirks

• A dot applies an operation to every element: x .- 1 subtracts 1 from each value, and sigmoid.(z) calls sigmoid on each one. Plain x - 1 is a MethodError.
• Arrays are indexed from 1, as in R.
• / always gives a float, so 7 / 2 is 3.5; ÷ is integer division.
• exp of a square matrix is the matrix exponential, not exp of each element; that needs exp.(M).
• The standard library has no CSV parser, so implementations here split each line on commas.

git clone https://github.com/qnixon/dswithx.git
cd dswithx
python3 harness/run_tests.py --language julia