Nothing to Julia in Minutes
Nothing to Julia in Minutes
OS X Command Line Tools
The first step before doing anything else will be to make sure that we have our OS X basic command-lined dev tools. You probably won’t use these tools far into the future, but having these will allow us to bootstrap into the software we want later.
Install
Run the following command in Terminal.
1
computer-name:~ xcode-select --install
Homebrew
The starting point for managing our installs will be Homebrew. Homebrew is a great tool for finding, installing, and upgrading open source tools on OS X.
Install
The easiest way to install this is via the terminal. Just copy paste the following into the terminal.
class="highlight">1
computer-name:~ /usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
Subsequently, we’ll need to make sure that everything is updated.
class="highlight">1
computer-name:~ brew update; brew upgrade;
The call to update should make sure that our current list of packages managed by Brew is updated. Subsequently, upgrade will update all currently installed packages.
Julia
Julia…what can we say. A thing of beauty. Let get her running.
Install
For my part, I’ve had some historical problems building Julia from source on OS X. I think that most of these issues have been resolved. However, the build process can also be quite slow. So, instead, lets take a look at grabbing the pre-built binaries and linking to those as needed.
Get the Julia v.0.4.5 DMG here.
Double click the DMG and drag Julia-0.4.5. into your applications folder.
Running (Easy)
The easiest thing to do at this point is to double click the Julia-0.4.5. application. It should open up a REPL window saying something like the following at the top:
class="highlight">1
2
3
4
5
6
7
8
9
_ _ _(_)_ | A fresh approach to technical computing
(_) | (_) (_) | Documentation: http://docs.julialang.org
_ _ _| |_ __ _ | Type "?help" for help.
| | | | | | |/ _` | |
| | |_| | | | (_| | | Version 0.4.5 (2016-03-18 00:58 UTC)
_/ |\__'_|_|_|\__'_| | Official http://julialang.org/ release
|__/ | x86_64-apple-darwin13.4.0
julia>
Much like Python and Matlab, Julia ships with an interactive scripting environment which allows you run commands one at a time and investigate the results to your hearts content. Matlab-like options such as whos() also exist. There are some other advanced features of the REPL that we’ll cover later (such as running bash commands).
Jupyter
Jupyter is a visual environment for composing Julia, Python, and R scripts. One can write code in line with figures and TeX text, which makes it a really nice for sharing results back and forth with others as well as for keeping notes for yourself.
Install
1. python
We want to make sure that we have a version of Python that we can actually play with. Unfortunately, the one given to us by the OS X devtools does not like to place nice with others. Lets get Python 3 as well as some buildtools via Homebrew.
class="highlight">1
2
computer-name:~ brew install python3
(...build output...)
2. jupyter
From here we should be able to use pip3 to install jupyter.
class="highlight">1
computer-name:~ pip3 install jupyter
Usage
Now we should have jupyter installed on the system. To run Jupyter and create a notebook from terminal, you should now be able to simply run
class="highlight">1
computer-name:~ jupyter notebook
which will open up a browser window to the home directory for Jupyter.
Julia Kernel for Jupyter
Since Jupyter is general across a number of different possible languages, we need to configure Jupyter for using our current instally of Julia. To do this, we will open up Julia and install the IJulia.jl package. We can do this from Julia by running
class="highlight">1
2
julia> Pkg.add("IJulia")
julia> Pkg.build("IJulia")
Plotting Libraries
matplotlib
Certainly one of the most important libraries to install. We can install the Python package matplotlib and use it in Julia in the following manner. First, lets make sure that we have a copy external to Julia by installing matplotlib via pip3.
class="highlight">1
computer-name:~ pip3 install matplotlib
Next, lets install a couple of packages within Julia to call matplotlib from within a Julia script.
class="highlight">1
2
3
4
julia> Pkg.add("PyCall")
julia> Pkg.build("PyCall")
julia> Pkg.add("PyPlot")
julia> Pkg.build("PyPlot")
And voilà, we’re done! Julia has a nice interface for interacting with Python libraries, which we will see in the future.
Scikit-Learn
We will also be taking advantage of some features of Scikit-Learn, so lets install that as well. Again, we can install Scikit-Learn both exeternally on our system, and also allowing for internal references to the Scikit libraries from within Julia. First, lets install Scikit via pip3…
class="highlight">1
computer-name:~ pip3 install scikit-learn
So now we have a Scikit for use from our Python3 build. But also, lets install a copy for Julia. We will make use of the ScikitLearn.jl wrapper to ease calling Scikit funcitons from within Julia (though we could also do this directly via PyCall.jl). To ease some common build problems, we will use the Conda.jl wrapper of the python Conda package manager to create a Julia-local copy of Scikit-learn. Then we’ll install the wrapper package and be done.
class="highlight">1
2
3
4
5
julia> Pkg.add("Conda")
julia> Pkg.build("Conda")
julia> Conda.add("scikit-learn")
julia> Pkg.add("ScikitLearn")
julia> Pkg.build("ScikitLearn")
We can test to make sure that we have it working via
class="highlight">1
2
3
4
5
6
7
8
9
julia> using ScikitLearn
julia> @sk_import datasets: load_digits
julia> data = load_digits()
Dict{Any,Any} with 5 entries:
"images" => 1797x8x8 Array{Float64,3}:…
"target_names" => [0,1,2,3,4,5,6,7,8,9]
"data" => 1797x64 Array{Float64,2}:…
"DESCR" => "Optical Recognition of Handwritten Digits Data Set\n===========================…
"target" => [0,1,2,3,4,5,6,7,8,9 … 5,4,8,8,4,9,0,8,9,8]
Lets have some fun and use ASCII Plots to show us an image of some digits
class="highlight">1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
julia> Pkg.add("ASCIIPlots")
julia> using ASCIIPlots
julia> imagesc(squeeze(data["images"][1,:,:],1))
+ @## +
@#@## @#+
+ @#+ @##
+ @# # #
+ # # #
+ @# + @##
+ @#+ # @#
# @##
julia> imagesc(squeeze(data["images"][8,:,:],1))
# # @#@#@#+
# # + # @#
# @#+
+ # # @#@##
+ # @#@#+
@##
# @#+
@##
Seems to work!
Eric W. Tramel, Ph.D.
Senior Machine Learning Scientist, Owkin