It’s no secret that bad data leads to bad decisions and poor results.  However, how do you prevent dirty data from taking up residency in your data store?  Some might argue that it’s the responsibility of the person sending you the data.  While that may be true, in practice that will rarely hold up.  It doesn’t matter how many times you ask, you will get the data however they decide to provide it.

So now you have bad data.  What constitutes bad data?  There are quite a few valid answers, for example:

• Invalid date values
• Inappropriate characters
• Wrong data
• Values that exceed a pre-set threshold

While it is certainly possible to write your own scripts and custom SQL to identify and deal with these data anomalies, that effort often takes too long and becomes difficult to maintain.  Instead, leveraging an ETL tool like expressor Studio makes the data cleansing process much easier and faster.  Below are some tips for leveraging expressor to get your data into tip-top shape.

Tip 1:     Build reusable data objects with embedded cleansing rules

One of the new features in expressor Studio 3.2 is the ability to define constraints at the metadata level.  Using expressor’s concept of Semantic Types, you can define reusable data objects that have embedded logic such as constraints for dealing with dirty data.  Once defined, they can be saved as a shared atomic type and then re-applied to other data attributes in other schemas.

As you can see in the figure above, I’ve defined a constraint on zip code.  I can then save the constraint rules I defined for zip code as a shared atomic type called zip_type for example.   The next time I get a different data source with a schema that also contains a zip code field, I can simply apply the shared atomic type (shown below) and the previously defined constraints will be automatically applied.

Tip 2:     Unlock the power of regular expressions in Semantic Types

Another powerful feature introduced in expressor Studio 3.2 is the option to use regular expressions as a constraint.   A regular expression is used to identify patterns within data.   The patterns could be something as simple as a date format or something much more complex such as a street address.  For example, I could define that a valid IP address should be made up of 4 numbers, each 0 to 255, and separated by a period.  So 192.168.23.123 might be a valid IP address whereas 888.777.0.123 would not be.   How can I account for this using regular expressions?

A very simple regular expression that would look for any 4 sets of 3 digits separated by a period would be:  ^[0-9]{1,3}\.[0-9]{1,3}\.[0-9]{1,3}\.[0-9]{1,3}$

Alternatively, the following would be the exact check for truly valid IP addresses as we had defined above:  ^(25[0-5]|2[0-4][0-9]|1[0-9]{2}|[1-9]?[0-9])\.(25[0-5]|2[0-4][0-9]|1[0-9]{2}|[1-9]?[0-9])\.(25[0-5]|2[0-4][0-9]|1[0-9]{2}|[1-9]?[0-9])\.(25[0-5]|2[0-4][0-9]|1[0-9]{2}|[1-9]?[0-9])$ .  In expressor, we would enter this regular expression as a constraint like this:

Here we select the corrective action to be ‘Escalate’, meaning that the expressor Dataflow operator will decide what to do.  Some of the options include rejecting the offending record, skipping it, or aborting the dataflow.

Tip 3:     Email pattern expressions that might come in handy

In the example schema that I am using, there’s a field for email.  Email addresses are often entered incorrectly because people are trying to avoid spam.  While there are a lot of different ways to define what constitutes a valid email address, a quick search online yields a couple of really useful regular expressions for validating email addresses:

This one is short and sweet:  \b[A-Z0-9._%+-]+@[A-Z0-9.-]+\.[A-Z]{2,4}\b
(Source: http://www.regular-expressions.info/)

This one is more specific about which characters are allowed:  ^([a-zA-Z0-9_\-\.]+)@((\[[0-9]{1,3}\.[0-9]{1,3}\.[0-9]{1,3}\.)|(([a-zA-Z0-9\-]+\.)+))([a-zA-Z]{2,4}|[0-9]{1,3})(\]?)$
(Source: http://regexlib.com/REDetails.aspx?regexp_id=26 )

Tip 4:     Reject “dirty data” for analysis or further processing

Yet another feature introduced in expressor Studio 3.2 is the ability to reject records based on constraint violations.  To capture reject records on input, simply specify Reject Record in the Error Handling setting for the Read File operator.  Then attach a Write File operator to the reject port of the Read File operator as such:

Next, in the Write File operator, you can configure the expressor operator in a similar way to the Read File.  The key difference would be that the schema needs to be derived from the upstream operator as shown below:

Once configured, expressor will output rejected records to the file you specified.  In addition to the rejected records, expressor also captures some diagnostic information that will be helpful towards identifying why the record was rejected.  This makes diagnosing errors much easier!

Tip 5:    Use a Filter or Transform after the initial cleansing to finish the job

Sometimes you may want to predicate the data cleansing on a more complex set of conditions.  For example, I may only be interested in processing data containing males over the age of 25 in certain zip codes.  Using an expressor Filter operator, you can define the conditional logic which isolates the records of importance away from the others.

Alternatively, the expressor Transform operator can be used to alter the input value via a user defined algorithm or transformation.  It also supports the use of conditional logic and data can be rejected based on constraint violations.

However, the best tip I can leave you with is to not constrain your solution design approach – expressor operators can be combined in many different ways to achieve the desired results.  For example, in the expressor Dataflow below, I can post-process the reject data from the Filter which did not meet my pre-defined criteria and, if successful, Funnel it back into the flow so that it gets written to the target table.

I continue to be impressed that expressor offers all this functionality as part of their FREE expressor Studio desktop ETL tool, which you can download from here.  Their Studio ETL tool is absolutely free and they are very open about saying that if you want to deploy their software on a dedicated Windows Server, you need to purchase their server software, whose pricing is posted on their website.

Reference: Pinal Dave (http://blog.SQLAuthority.com)