Set up large database in MySQL for analysis in R

Question

I have reached the limit of RAM in analyzing large datasets in R. I think my next step is to import these data into a MySQL database and use the RMySQL package. Largely because I don't know database lingo, I haven't been able to figure out how to get beyond installing MySQL with hours of Googling and RSeeking (I am running MySQL and MySQL Workbench on Mac OSX 10.6, but can also run Ubuntu 10.04).

Is there a good reference on how to get started with this usage? At this point I don't want to do any sort of relational databasing. I just want to import .csv files into a local MySQL database and do the subsetting in with RMySQL.

I appreciate any pointers (including "You're way off base!" as I'm new to R and newer to large datasets... this one's around 80 mb)

Answer 1

The documentation for RMySQL is pretty good - but it does assume that you know the basics of SQL. These are:

• creating a database
• creating a table
• getting data into the table
• getting data out of the table

Step 1 is easy: in the MySQL console, simply "create database DBNAME". Or from the command line, use mysqladmin, or there are often MySQL admin GUIs.

Step 2 is a little more difficult, since you have to specify the table fields and their type. This will depend on the contents of your CSV (or other delimited) file. A simple example would look something like:

use DBNAME;
create table mydata(
  id INT(11) NOT NULL AUTO_INCREMENT PRIMARY KEY,
  height FLOAT(3,2)
); 

Which says create a table with 2 fields: id, which will be the primary key (so has to be unique) and will autoincrement as new records are added; and height, which here is specified as a float (a numeric type), with 3 digits total and 2 after the decimal point (e.g. 100.27). It's important that you understand data types.

Step 3 - there are various ways to import data to a table. One of the easiest is to use the mysqlimport utility. In the example above, assuming that your data are in a file with the same name as the table (mydata), the first column a tab character and the second the height variable (with no header row), this would work:

mysqlimport -u DBUSERNAME -pDBPASSWORD DBNAME mydata

Step 4 - requires that you know how to run MySQL queries. Again, a simple example:

select * from mydata where height > 50;

Means "fetch all rows (id + height) from the table mydata where height is more than 50".

Once you have mastered those basics, you can move to more complex examples such as creating 2 or more tables and running queries that join data from each.

Then - you can turn to the RMySQL manual. In RMySQL, you set up the database connection, then use SQL query syntax to return rows from the table as a data frame. So it really is important that you get the SQL part - the RMySQL part is easy.

There are heaps of MySQL and SQL tutorials on the web, including the "official" tutorial at the MySQL website. Just Google search "mysql tutorial".

Personally, I don't consider 80 Mb to be a large dataset at all; I'm surprised that this is causing a RAM issue and I'm sure that native R functions can handle it quite easily. But it's good to learn new skill such as SQL, even if you don't need them for this problem.

Answer 2

The sqldf package might give you an easier way to do what you need: http://code.google.com/p/sqldf/. Especially if you are the only person using the database.

Edit: Here is why I think it would be useful in this case (from the website):

With sqldf the user is freed from having to do the following, all of which are automatically done:

• database setup
• writing the create table statement which defines each table
• importing and exporting to and from the database
• coercing of the returned columns to the appropriate class in common cases

See also here: http://stackoverflow.com/questions/1727772/quickly-reading-very-large-tables-as-dataframes-in-r/1820610#1820610

Answer 3

I have a pretty good suggestion. For 80MB use SQLite. SQLite is a super public domain, lightweight, super fast file-based database that works (almost) just like a SQL database. http://www.sqlite.org/index.html

You don't have to worry about running any kind of server or permissions, your database handle is just a file.

Also, it stores all data as a string, so you don't even have to worry about storing the data as types (since all you need to do is emulate a single text table anyway).

Someone else mentioned sqldf: http://code.google.com/p/sqldf/

which does interact with SQLite: http://code.google.com/p/sqldf/#9._How_do_I_examine_the_layout_that_SQLite_uses_for_a_table?_whi

So your SQL create statement would be like this

create table tablename (
  id INT(11) INTEGER PRIMARY KEY,
  first_column_name TEXT,
  second_column_name TEXT,
  third_column_name TEXT 
);

Otherwise, neilfws' explanation is a pretty good one.

P.S. I'm also a little surprised that your script is choking on 80mb. It's not possible in R to just seek through the file in chunks without opening it all up in memory?

Answer 4

I agree with what's been said so far. Though I guess getting started with MySQL (databases) in general is not a bad idea for the long if you are going to deal with data. I mean I checked your profile which says finance PhD student. I don't know if that means quant. finance, but it is likely that you will come across really large datasets in your career. I you can afford some time, I would recommend to learn something about databases. It just helps. The documentation of MySQL itself is pretty solid and you can a lot of additional (specific) help here at SO.

I run MySQL with MySQL workbench on Mac OS X Snow Leopard too. So here´s what helped me to get it done comparatively easy.

• I installed MAMP , which gives my an local Apache webserver with PHP, MySQL and the MySQL tool PHPmyadmin, which can be used as a nice webbased alternative for MySQL workbench (which is not always super stable on a Mac :) . You will have a little widget to start and stop servers and can access some basic configuration settings (such as ports through your browser) . It´s really one-click install here.

• Install the Rpackage RMySQL . I will put my connection string here, maybe that helps:

• Create your databases with MySQL workbench. INT and VARCHAR (for categorical variables that contain characters) should be the field types you basically need at the beginning.

• Try to find the import routine that works best for you. I don't know if you are a shell / terminal guy – if so you'll like what was suggested by neilfws. You could also use LOAD DATA INFILE which is I prefer since it's only one query as opposed to INSERT INTO (line by line)

If you specify the problems that you have more accurately, you'll get some more specific help – so feel free to ask ;)

I assume you have to work a lot with time series data – there is a project (TSMySQL) around that use R and relational databases (such as MySQL, but also available for other DBMS) to store time series data. Besides you can even connect R to FAME (which is popular among financers, but expensive). The last paragraph is certainly nothing basic, but I thought it might help you to consider if it´s worth the hustle to dive into it a little deeper.