- Item 18: Use Views to Simplify What Cannot Be Changed
- Item 19: Use ETL to Turn Nonrelational Data into Information
- Item 20: Create Summary Tables and Maintain Them
- Item 21: Use UNION Statements to "Unpivot" Non-normalized Data
This chapter is from the book
This chapter is from the book
This chapter is from the book
Item 19: Use ETL to Turn Nonrelational Data into Information
Extract, Transform, Load (ETL) is a set of procedures or tools you can use to Extract data from an external source, Transform it to conform to relational design rules or to conform to other requirements, then Load it into your database for further use or analysis. Nearly all database systems provide various utilities to aid in this process. These utilities are, quite simply, a means to convert raw data into information.
To get an idea of what these utilities can do, let’s take a look at some of the tools in Microsoft Access—one of the first Windows-era database systems to provide built-in ways to load and transform data into something useful. Assume you work as the marketing manager for a company that produces breakfast cereals. You need not only to analyze competitive sales from another manufacturer but also to break down this analysis by individual brands.
You can certainly glean total sales information from publicly available documents, but you really want to try to break down competitive sales by individual brand. To do this, you might strike up an agreement with a major grocery store chain to get them to provide their sales information by brand in return for a small discount on your products. The grocery chain promises to send you a spreadsheet containing sales data from all its stores broken down by competitive brand for the previous year. The data you receive might look like Table 3.1.
Table 3.1 Sample competitive sales data
Product |
Jan |
Feb |
Mar |
|||
Alpha-Bits |
|
57775.87 |
|
40649.37 |
|
. . . |
Golden Crisp |
|
33985.68 |
|
17469.97 |
|
. . . |
Good Morenings |
|
40375.07 |
|
36010.81 |
|
. . . |
Grape-Nuts |
|
55859.51 |
|
38189.64 |
|
. . . |
Great Grains |
|
37198.23 |
|
41444.41 |
|
. . . |
Honey Bunches of Oats |
|
63283.28 |
|
35261.46 |
|
. . . |
. . . additional rows . . . |
||||||
It is clear that some blank columns that you do not need were added for readability. You also need to transform the data to end up with one row per product per month, and you have a separate table listing competitive products that has its own primary key, so you need to match on product name to get the key value to use as a foreign key.
Let’s start by extracting the data from the spreadsheet into a more usable form. Microsoft Access can import data in many different formats, so let’s fire up the Import tool to import a spreadsheet. In the first step, you identify the file and tell Access what you want to do with the output (import into a new table, append the data to an existing table, or link as a read-only table).
When you go to the next step, Access shows you a grid with a sample of the data it found, as shown in Figure 3.2. Because it determined that the first row might very well be usable as column names, it has used the names it found and has assigned generated names to the blank columns.
)
Figure 3.2 The Import Spreadsheet utility performing an initial analysis of the data
In the following step, Access shows you a display where you can select columns one at a time, tell Access to skip unimportant columns, and fix the data type that the utility has assumed. Figure 3.3 on the next page shows one of the data columns selected. The utility has assumed that the numbers, because they contain decimal points, should be imported as the very flexible Double data type. We know that these are all dollar sales figures, so it makes sense to change the data type to Currency to make it easier to work with the data. You can also see the “Do not import” check box (behind the drop-down) that you can select for columns that you want to ignore.
)
Figure 3.3 Selecting columns to skip and choosing a data type
The next step in the utility lets you pick a column to act as the primary key, ask the utility to generate an ID column with incrementing integers, or assign no primary key to the table. The final step allows you to name the table (the default is the name of the worksheet) and to invoke another utility after importing the table to perform further analysis and potentially reload the data into a more normalized table design. If you choose to run the Table Analyzer, Access presents you with a design tableau as shown in Figure 3.4. In the figure, we have already dragged and dropped the Product column into a separate table and named both tables. As you can see, the utility automatically generates a primary key in the product table and provides a matching foreign key in the sales data table.
)
Figure 3.4 Using the Table Analyzer to break out products into a separate table
Even after using the Table Analyzer, you can see that there is still plenty of work to do to further normalize the sales data into one row per month. You can “unpivot” the sales data by using a UNION query to turn the columns into rows, as shown in Listing 3.5. (See also Item 21, “Use UNION statements to ‘unpivot’ non-normalized data.”)
Listing 3.5 Using a UNION query to “unpivot” a repeating group
SELECT '2015-01-01' AS SalesMonth, Product, Jan AS SalesAmt FROM tblPostSales UNION ALL SELECT '2015-02-01' AS SalesMonth, Product, Feb AS SalesAmt FROM tblPostSales UNION ALL ... etc. for all 12 months.
The tools in Microsoft Access are fairly simple (for example, they cannot handle totals rows), but they give you an idea of the amount of work that can be saved when trying to perform ETL to load external data into your database. As noted earlier, most database systems provide similar—and in some cases more powerful—tools that you can use. Examples include Microsoft SQL Server Integration Services (SSIS), Oracle Data Integrator (ODI), and IBM InfoSphere DataStage. Commercial tools are available from vendors such as Informatica, SAP, and SAS, and you can also find a number of open-source tools available on the Web.
The main point here is that you should use those tools so that your data conforms to the data model that your business needs, not the other way around. A common mistake is to build tables that fit the incoming data as is and then use it directly in applications. The investment made to transform data will result in a database that is easy to understand and maintain in spite of the divergent data sources from which it may collect the raw input.
Things to Remember
ETL tools allow you to import nonrelational data into your database with less effort.
ETL tools help you reformat and rearrange imported data so that you can turn it into information.
Most database systems offer some level of ETL tools, and there are also commercial tools available.