Converting XML Documents into Relational Data Models.
The xml2relational Package
Joachim Zuckarelli
What xml2relational does
xml2relational is designed to convert XML documents with nested object hierarchies into a set of R dataframes. These dataframes represent the different tables in a relational data model and are connected amongst each other by foreign keys. Essentially, xml2relational flattens an object-oriented data structure into a relational data structure.
Once the relational structure is created (and that is basically a list of dataframes representing the different tables) you can export both the data model (as SQL CREATE statements) and the data (either as SQL INSERT statements or as CSV files) to get the data easily into a relational database.
Getting started
Installing and loading xml2relational
You can install the xml2relational package from CRAN by executing the following code in your script or in the R console:
install.packages("xml2relational", dependencies = TRUE)
After having installed the package you need to load it (attach it to the search path) by calling library():
library(xml2relational)
Our example data set
To demonstrate how xml2relational works, we will use a small sample dataset that is shipped together with the xml2relational package: the customer dataset.
Here is how it looks like:
<xml>
<customer>
<customerno>C0023751</customerno>
<givenname>Sarah</givenname>
<surname>Durbin</surname>
<email>[email protected]</email>
<address>
<street>139 W Jackson Blvd</street>
<postalcode>60604</postalcode>
<city>
<name>Chicago</name>
<state>Illinois</state>
</city>
<country>
<name>United States of America</name>
<isocode>US</isocode>
</country>
</address>
<username>queenofqueens</username>
</customer>
<customer>
<customerno>C0017439</customerno>
<givenname>Mark</givenname>
<surname>Durbin</surname>
<email>[email protected]</email>
<address>
<street>139 W Jackson Blvd</street>
<postalcode>60604</postalcode>
<city>
<name>Chicago</name>
<state>Illinois</state>
</city>
<country>
<name>United States of America</name>
<isocode>US</isocode>
</country>
</address>
<username>durby82</username>
</customer>
<customer>
<customerno>C0248538</customerno>
<givenname>Max</givenname>
<surname>Brunner</surname>
<email>[email protected]</email>
<address>
<street>Rotkreuzplatz 5</street>
<postalcode>80634</postalcode>
<city>
<name>Munich</name>
<state>Bavaria</state>
</city>
<country>
<name>Germany</name>
<isocode>DE</isocode>
</country>
</address>
<username>brunnermax_69</username>
</customer>
<customer>
<customerno>C0271182</customerno>
<givenname>Urs</givenname>
<surname>Richli</surname>
<email>[email protected]</email>
<address>
<street>Seestrasse 43</street>
<postalcode>6052</postalcode>
<city>
<name>Hergiswil</name>
<state>Luzern</state>
</city>
<country>
<name>Switzerland</name>
<isocode>CH</isocode>
</country>
</address>
<username>ursrichli</username>
</customer>
<customer>
<customerno>C0019935</customerno>
<givenname>Clara-Sophie</givenname>
<surname>Dr. Hellmann</surname>
<email>[email protected]</email>
<address>
<street>Brienner Strasse 11</street>
<postalcode>80333</postalcode>
<city>
<name>Munich</name>
<state>Bavaria</state>
</city>
<country>
<name>Germany</name>
<isocode>DE</isocode>
</country>
</address>
<username>helli</username>
</customer>
<customer>
<customerno>C0019935</customerno>
<givenname>Thomas</givenname>
<surname>Chang</surname>
<email>[email protected]</email>
<address>
<street>539 Lombard St</street>
<postalcode>94133</postalcode>
<city>
<name>San Francisco</name>
<state>California</state>
</city>
<country>
<name>United States of America</name>
<isocode>US</isocode>
</country>
</address>
<username>tchango123</username>
</customer>
</xml>
In this dataset we have a nested object structure. Specifically, each customer has an address consisting of several elements. Among those elements is the city which is again an object of its own, with a city name and state. The same applies to the country which is included with its name and its ISO country code. When you look at the (completely made-up) customers here, you will notice that the customers Sarah Durbin and Mark Durbin (the first two customers) share the same address. Also, Max Brunner and Clara-Sophie Hellmann both live in Munich, Germany (although at different addresses). Thomas Chang of San Francisco lives in the USA, as do the Durbins.
When we now process the data and derive the relational data model, xml2relational will take care of these ‘duplicates’.
Processing the data
Deriving the relational data model from this XML data is fairly simple:
customer.data <- toRelational("customers.xml")
The toRelational() function flattens the hierarchical structure of the XML data and distributes the data to a set of dataframes representing the tables of our relational data model. It returns these dataframes as a list (customer.data). We can now inspect this list to see the tables that have been generated:
class(customer.data)
## [1] "list"
names(customer.data)
## [1] "xml" "customer" "address" "city" "country"
class(customer.data$customer)
## [1] "data.frame"
Let us have a closer look at the customer dataframe:
customer.data$customer
## ID_customer customerno givenname surname
## 1 263023 C0023751 Sarah Durbin
## 2 597336 C0017439 Mark Durbin
## 3 59960 C0248538 Max Brunner
## 4 159381 C0271182 Urs Richli
## 5 83969 C0019935 Clara-Sophie Dr. Hellmann
## 6 465004 C0019935 Thomas Chang
## email FKID_address username
## 1 [email protected] 674038 queenofqueens
## 2 [email protected] 674038 durby82
## 3 [email protected] 149765 brunnermax_69
## 4 [email protected] 718252 ursrichli
## 5 [email protected] 977313 helli
## 6 [email protected] 112551 tchango123
As you can see, each customer record has been assigned a primary key, ID_customer. The argument prefix.primary of the toRelational() function lets you change the prefix that is used to identify primary key fields. Its default value is "ID_". Similiarly, using the prefix.foreign argument you can change the prefix used for the names of foreign key fields from its default value "FKID_" to whatever you like. The name of the key fields always consists of the prefix and the name of the table.
In the customer table we have a foreign key that relates to the address. You may have noticed that, as expected, the data records of Sarah and Mark Durbin point to the same address record as they live in the same place.
Let us now look into the address table:
customer.data$address
## ID_address street postalcode FKID_city FKID_country
## 1 674038 139 W Jackson Blvd 60604 735977 495268
## 2 149765 Rotkreuzplatz 5 80634 2299 352009
## 3 718252 Seestrasse 43 6052 448761 817914
## 4 977313 Brienner Strasse 11 80333 2299 352009
## 5 112551 539 Lombard St 94133 70561 495268
Again, the address points to other tables, namely the city and the country table. As we would have expected, the two Munich addresses point to the same city and the same country, and the two US addresses point to the same record in the country table.
You see how easy it is to flatten a hierarchical, objected-oriented XML data structure to a relational data model using the toRelational() function.
Saving the results
In the next step, we want to export our results. That can mean two things:
- exporting the data model (i.e. the structure of the tables)
- exporting the data, the content of the tables.
For the first task, xml2relational provides the getCreateSQL() function. This function returns ready-to-excecute SQL CREATE statements. It supports three built-in SQL flavors, MySQL, TransactSQL and Oracle. You add additional SQL flavors, if you like. In this case, you would use sql.style argument to provide a special dataframe containing the required definitions for the new SQL dialect. Please consult the online help texts for more information on how this is done.
In order to generate proper SQL CREATE statements, getCreateSQL() guesses the data types of the table fields from the data. If you do not like the results, you can provide your own function to derive the data types as datatype.func argument. This function would need to accept exactly one argument, a vector with the field vales of the field for which a datatype needs to be guessed. It then must return the datatype as a one-element character vector.
If you are not going to change the behavior of getCreateSQL() using these options, generating the SQL CREATE statements is very straightforward:
create.sql <- getCreateSQL(customer.data, "MySQL")
cat(create.sql, sep="\n\n")
## CREATE TABLE xml (
## PRIMARY KEY (ID_xml)
## , ID_xml BIGINT
## , FOREIGN KEY (FKID_customer) REFERENCES customer(ID_customer)
## , FKID_customer BIGINT
## );
##
## CREATE TABLE customer (
## PRIMARY KEY (ID_customer)
## , ID_customer BIGINT
## , customerno VARCHAR(8) NOT NULL
## , givenname VARCHAR(12) NOT NULL
## , surname VARCHAR(12) NOT NULL
## , email VARCHAR(34) NOT NULL
## , FOREIGN KEY (FKID_address) REFERENCES address(ID_address)
## , FKID_address BIGINT
## , username VARCHAR(13) NOT NULL
## );
##
## CREATE TABLE address (
## PRIMARY KEY (ID_address)
## , ID_address BIGINT
## , street VARCHAR(19) NOT NULL
## , postalcode BIGINT NOT NULL
## , FOREIGN KEY (FKID_city) REFERENCES city(ID_city)
## , FKID_city BIGINT
## , FOREIGN KEY (FKID_country) REFERENCES country(ID_country)
## , FKID_country BIGINT
## );
##
## CREATE TABLE city (
## PRIMARY KEY (ID_city)
## , ID_city BIGINT
## , name VARCHAR(13) NOT NULL
## , state VARCHAR(10) NOT NULL
## );
##
## CREATE TABLE country (
## PRIMARY KEY (ID_country)
## , ID_country BIGINT
## , name VARCHAR(24) NOT NULL
## , isocode VARCHAR(2) NOT NULL
## );
xml2relational tries to guess the datatype from the actual data. When you are working with the MySQL, Transact SQL (T-SQL) and Oracle dialects/flavors of SQL, this should be alright. Nevertheless, using the datatype.func argument of getcreateSQL() you can also provide your own function to determine the data type. This function would need to take exactly one argument, a data vector from a data table, and return the appropriate SQL data type as a one-element character vector. Alternatively, you can also use the built-in mechanism for determining the data type and just supply additional information on the SQL flavor that you use. Please consult the online help with ?getCreateSQL to learn more on providing the necessary information.
By setting the logical one.statement argument to TRUE you can let getcreateSQL() return the CREATE statements in one character value instead of a vector with one element per CREATE statement. In this case you can use the line.break argument to define how the different CREATE statement are to be separated (apart from a semicolon that is added by default).
To export the data as such you have two options:
- you export ready-to-execute SQL
INSERTstatements usinggetInsertSQL()function - you save the data to CSV files using
savetofiles().
Producing SQL INSERT statements for the data in one of the tables is very easy with getInsertSQL():
insert.sql <- getInsertSQL(customer.data, table.name = "city")
cat(insert.sql, sep="\n")
## INSERT INTO city(ID_city, name, state) VALUES (735977, 'Chicago', 'Illinois');
## INSERT INTO city(ID_city, name, state) VALUES (2299, 'Munich', 'Bavaria');
## INSERT INTO city(ID_city, name, state) VALUES (448761, 'Hergiswil', 'Luzern');
## INSERT INTO city(ID_city, name, state) VALUES (70561, 'San Francisco', 'California');
You can also export all the tables of your relational model with savetofiles():
savetofiles(customer.data)
This will save as many CSV files to your current working directory as you have tables in your model (customer.data). Each file is named for the name of the dataframe connected to the respective table, so city.csv will store the data from the city table.
More optional arguments for most of the functions discussed here are available. Please check the online help for more details.
Contact the author
I appreciate your questions, issues and feature requests. Contact me on [email protected], visit the GitHub repository on https://github.com/jsugarelli/xml2relational for the package source and follow me on Twitter to stay up-to-date!