julia_basic

Learn how to use Julia language. Why? It is similar to Python, but fast.

I use JuliaBox to run the IJulia notebook. They offer a system of 6gb memory.

But writing Julia in Atom editor is a good idea. I find it very useful in some ways. First, it offers a very good system of autocomplete code for beginners. Second, the Workspace is quite good, especially for who are R-studio addicted. And so on.

1. Some first words

Julia has a simlar indexer as in R, but different from Python. While you index from 0 in Python, Julia use the first index as 1 as in R.

For example:

x = [1,2,3]

x[0]
LoadError: MethodError: no method matching *(::Array{Int64,1}, ::Array{Int64,1})

x[1]
1

It is a bit good for R-users. In my view, I don't think beginning from 0 brings no additional value at all.

2. Some interesting codes

To re-index the data, we use this code:

car[:idx] = 1:size(car,1)

This is useful when you want to do event study. Above, car is the data frame name of the car dataset.

In R, thanks to Hadleyverse, we can use group_by and summarise to do any split-apply-combine strategy. In Julia, they offer a very strange way to do that. Let see this example: to calculate mean of price for each rep78 groups

for subdf in groupby(car, :rep78)
  println(mean(subdf[:price]))
end

Another interesting way is using by and do together:

by(car, :rep78) do df
  DataFrame(m = mean(df[:price]), sd = std(df[:price]))
end

In Python, I prefer the lambda such as lambda x: x + 1. In Julia, they have: x -> x + 1.

3. Benchmark

Finally, how is the performance between Julia and Python (we should ignore R, ^^)? Actually, Julia is quite as same as Python, as opposed to the benchmark in their homepage. Actually, some tasks Python are very fast now (ver 3.). In addition, with Numba, Python can be better with JIT compiler.

See the benchmark here from NASA.