You might encounter problems with how data has been structured or formatted that you must fix prior to providing the content to your target system. You can use the methods in this section to locate problems with the content or data typing of your data.
Locate mismatched values
When Trifacta® Self-Managed Enterprise Edition evaluates a dataset sample, it interprets the values in a column against its expectations for the values. Based on the column's specified data type and internal pattern matching, values are categorized as valid, mismatched, or missing. These value categories are represented in a slender bar at the top of each column.
- A mismatched value is any value that seems to be of a different data type than the type specified for the column. For example, if the value
San Franciscoappears in a column of Zip Code type, it would be marked as a mismatched value.
In the data quality bar, mismatched values are identified in red:
Tip: The type against which values are checked is displayed to the upper left of the data quality bar. Below, the data type is
ZIP for U.S. Zip code data.
Tip: Before you start performing transformations on your data based on mismatched values, you should check the data type for these columns to ensure that they are correct. For more information, see Supported Data Types.
Tip: From the Transformer page, click the mismatched values in a column's data quality bar to see their count, highlight them in the rows of the data grid, and trigger a set of suggestions for your review.
Mismatched values can be sourced from a variety of issues:
- Values may be miskeyed into the source system.
- The source system may introduce errors in output, particularly if the data is generated for export using a customized structure.
- Incorrect use of column delimiters may create offsets among fields in individual rows.
- Data may be badly structured across a set of rows.
- The column may be assigned the wrong data type.
Tip: When cleaning up bad data, you should look to work from bigger problems to smaller problems. If a higher percentage of a column's values have been categorized as mismatched data, it may indicate a wider problem with the data. In affected rows, verify if other columns' values are also mismatched. These rows should be reviewed and fixed first. When fixed, other mismatches may be fixed in other rows, too.
Methods for fixing mismatched data
When you discover mismatched data in your dataset, you have the following basic methods of fixing it:
Change the data type. If the percentage of mismatched rows is significant, you may need to change the data type for a better match.
Replace the values with constant values. This method works if it is clear to you that the values should be a single, consistent value. Select the mismatched values in the column, and then select one of the highlighted mismatched values. Use the
replacetransform to change the mismatched values to corrected values.
Tip: One easy way to fix isolated problems with mismatched values is to highlight a mismatched value in the data grid. A new set of suggestions is displayed. You can select the
replacesuggestion and then modify it to include the replacement value.
Set values with other columns' values. You can use the
settransform to fix mismatched values by replacing them with the corresponding values from other columns.
Use functions. Data can be fixed by using a function in conjunction with the
settransform to replace mismatched values.
- Delete rows. Select the mismatched values and use the
deletetransform to remove the problematic rows.
- Hide the column for now. You can remove the column from display if you want to focus on other things. Select Hide from the column drop-down. Note that hidden columns appear in any generated output.
- Delete the column. If the column data is unnecessary or otherwise unusable, you can delete the entire column from your dataset. Select Delete from the column drop-down.
Tip: Delete unnecessary columns as early as possible. Less data is easier to work with in the application and improves job execution performance.
NOTE: You might need to review and fixed mismatched data problems multiple times in your dataset. For example, if you unnest the data, additional mismatches might be discovered. Similarly, joins and lookups can reveal mismatches in data typing.
Mismatched values in transform code
In your transforms, mismatched data can be identified references as in the following:
Note that the single quotes are important around the value, which identifies the value as a constant.
Tip: In the above, note that the value
Zipcode identifies the data type that is used for matching the column values. In this case, for greater specificity, you might want to identify the mismatched values in the column against the data type
Integer, since all U.S. postal codes are positive integers. For more information on how to explicitly reference data types in your steps, see Valid Data Type Strings.
Tip: To trim whitespace out of a column, use the following transformation:
This step may increase the number of missing values (for values that contain only whitespace characters) and the number of instances of matching values (for values that have spaces before and after an alphanumeric value).
Set values using other columns
You can use values from other columns to replace mismatched values in your current column. Using the previous example, mismatched postal codes are replaced by the corresponding value in the parent entity's postal code column (
Use functions to fix mismatched values
In your transforms, you can insert a predefined function to replace mismatched data values. In the following example, the value for mismatched values in the
score column are computed as the average of all values in the column:
Tip: You can also use the
IFMISMATCHED function to test for mismatched values. Unlike the above construction, however,
IFMISMATCHED does not support an else clause when the value does match the listed data type. For more information, see IFMISMATCHED Function.
Bad data typing
Tip: Particularly for dates, data is often easiest to manage as String data type. Trifacta Self-Managed Enterprise Edition has a number of functions that you can deploy to manage strings. After the data has been properly formatted, you can change it to the proper data type. If you change data type immediately, you may have some challenges in reformatting and augmenting it. Do this step last.
For columns that have a high percentage of mismatched values, the column's data type may have been mis-assigned. In the following example, a column containing data on precipitation in inches has been mis-typed as Boolean data:
To change a column's data type, click the type identifier at the top of the column and select a new type. In this case, you would select
NOTE: After you change the type, review the data quality bar again. If there are still mismatched values, review them to see if you can categorize the source of the mismatch.
As you can see in the previous example, the precipitation column contains values set to
T, which may be short for
true. When the data type is set to
Decimal, these values now register as mismatched data. To fix, you can replace all
T values with
1.0 using the
Select an instance of
T in the column and click the Set suggestion card. Click Modify. For the
value in the transform, enter
1.0. Your transform should look like the following:
Tip: If possible, you should review and refer to an available schema of your dataset, as generated from the source system. If the data has also been mis-typed in the source system, you should fix it there as well, so any future exports from that system show the correct type.
NOTE: As needed, you can create custom data types for use in Trifacta Self-Managed Enterprise Edition, which enable you to validate column data against a custom set of values. See Create Custom Data Types.
This page has no comments.