Author Archives: randyzwitch - Articles

Tabular Data I/O in Julia

By: randyzwitch - Articles

Re-posted from: http://randyzwitch.com/julia-import-data/

Importing tabular data into Julia can be done in (at least) three ways: reading a delimited file into an array, reading a delimited file into a DataFrame and accessing databases using ODBC.

Reading a file into an array using readdlm

The most basic way to read data into Julia is through the use of the readdlm function, which will create an array:

readdlm(source, delim::Char, T::Type; options...)

If you are reading in a fairly normal delimited file, you can get away with just using the first two arguments, source and delim:It’s important to note that by only specifying the first two arguments, you leave it up to Julia to determine the type of array to return. In the code example above, an array of type ‘Any’ is returned, as the .csv file I read in was not of homogenous type such as Int64 or ASCIIString. If you know for certain which type of array you want, you specify the data type using the type argument:

It’s probably the case that unless you are looking to do linear algebra or other specific mathy type work, you’ll likely find that reading your data into a DataFrame will be more comfortable to work with (especially if you are coming from an R, Python/pandas or even spreadsheet tradition).

To write an array out to a file, you can use the writedlm function (defaults to comma-separated):

writedlm(filename, array, delim::Char)

Reading a file into a DataFrame using readtable

As I covered in my prior blog post about Julia, you can also read in delimited files into Julia using the DataFrames package, which returns a DataFrame instead of an array. Besides just being able to read in delimited files, the DataFrames package also supports reading in gzippped files on the fly:From what I understand, in the future you will be able to read files directly from Amazon S3 into a DataFrame (this is already supported in the AWS package), but for now, the DataFrames package works only on local files. Writing a DataFrame to file can be done with the writetable function: writetable(filename::String, df::DataFrame) By default, the writetable function will use the delimiter specified by the filename extension and default to printing the column names as a header.

Accessing Databases using ODBC

The third major way of importing tabular data into Julia is through the use of ODBC access to various databases such as MySQL and PostgreSQL.

Using a DSN

The Julia ODBC package provides functionality to connect to a database using a Data Source Name (DSN). Assuming you store all the credentials in your DSN (server name, username, password, etc.), connecting to a database is as easy as:

Of course, if you don’t want to store your password in your DSN (especially in the case where there are multiple users for a computer), you can pass the “usr” and “pwd” arguments to the ODBC.connect function:

ODBC.connect(dsn; usr="", pwd="")

Using a connection string

Alternatively, you can build your own connection strings within a Julia session using the advancedconnect function:Regardless of which way you connect, you can query data using the query function. If you want your output as a DataFrame, you can assign the result of the function to an object. If you want to save the results to a file, you specify the “file” argument:

Summary

Overall, importing data into Julia is no easier/more difficult than any other language. The biggest thing I’ve noticed thus far is that Julia is a bit less efficient than Python/pandas or R in terms of the amount of RAM needed to store data. In my experience, this is really only an issue once you are working with 1GB+ files (of course, depending on the resources available to you on your machine).

A Beginner’s Look at Julia

By: randyzwitch - Articles

Re-posted from: http://randyzwitch.com/julia-language-beginners/

Over the past month or so, I’ve been playing with a new scientific programming language called ‘Julia‘, which aims to be a high-level language with performance approaching that of C. With that goal in mind, Julia could be a replacement for the ‘multi-language’ problem of needing to move between R, Python, MATLAB, C, Fortran, Scala, etc. within a single scientific programming project.  Here are some observations that might be helpful for others looking to get started with Julia.

 

Get used to ‘Git’ and ‘make’

While there are pre-built binaries for Julia, due to the rapid pace of development, it’s best to build Julia from source. To be able to keep up with the literally dozen code changes per day, you can clone the Julia GitHub repository to your local machine. If you use one of the GitHub GUI’s, this is as easy as hitting the ‘Sync Branch’ button to receive all of the newest code updates.

To install Julia, you need to compile the code. The instructions for each supported operating system are listed on the Julia GitHub page. For Mac users, use Terminal to navigate to the directory where you cloned Julia, then run the following command, where ‘n’ refers to the number of concurrent processes you want the compiler to use:

make -j n

I use 8 concurrent processes on a 2013 MacBook Pro and it works pretty well. Certainly much faster than a single process. Note that the first time you run the ‘make’ command, the build process will take much longer than successive builds, as Julia downloads all the required libraries needed. After the first build, you can just run the ‘make’ command with a single process, as the code updates don’t take very long to build.

Package management is also done via GitHub. To add Julia packages to your install, you use the Pkg.add() function, with the package name in double-quotes.

Julia code feels very familiar

Text file import

Although the Julia documentation makes numerous references to MATLAB in terms of code similarity, Julia feels very familiar to me as an R and Python user. Take reading a .csv file into a dataframe and finding the dimensions of the resulting object:
In each language, the basic syntax is to call a ‘read’ function, specify the .csv filename, then the defaults of the function read in a basic file. I also could’ve specified other keyword arguments, but for purposes of this example I kept it simple.

Looping

Looping in Julia is similar to other languages. Python requires proper spacing for each level of a loop, with a colon for each evaluated expression. And although you generally don’t use many loops in R, to do so requires using parenthesis and brackets.

If you’re coming from a Python background, you can see that there’s not a ton of difference between Python looping into a dictionary vs. Julia. The biggest differences are the use of the ‘end’ control-flow word and that Julia doesn’t currently have the convenience “Counter” object type. R doesn’t natively have a dictionary type, but you can add a similar concept using the hash package.

Vectorization

While not required to achieve high performance, Julia also provides the functional programming construct of vectorization and list comprehensions. In R, you use the ‘apply’ family of functions instead of loops in order to apply a function to multiple elements in a list. In Python, there are the ‘map’ and ‘reduce’ functions, but there is also the concept of list comprehensions. In Julia, both of the aforementioned functionalities are possible.

In each case, the syntax is just about the same to apply a function across a list/array of numbers.

A small, but intense community

One thing that’s important to note about Julia at this stage is that it’s very early. If you’re going to be messing around with Julia, there’s going to be a lot of alone-time experimenting and reading the Julia documentation. There are also several other resources including a Julia-Users Google group, Julia for R programmers, individual discussions on GitHub in the ‘Issues’ section of each Julia package, and a few tutorials floating around (here and here).

Beyond just the written examples though, I’ve found that the budding Julia community is very helpful and willing in terms of answering questions. I’ve been bugging the hell out of John Myles White and he hasn’t complained (yet!), and even when code issues are raised through the users group or on GitHub, ultimately everyone has been very respectful and eager to help. So don’t be intimidated by the fact that Julia has a very MIT and Ph.D-ness to it…jump right in and migrate some of your favorite code over from other languages.

While I haven’t moved to using Julia for my everyday workload, I am getting facility to the point where I’m starting to consider using Julia for selected projects. Once the language matures a bit more, JuliaStudio starts to approach RStudio in terms of functionality, and I get more familiar with the language in general, I can see Julia taking over for at least one if not all of my scientific programming languages.