This section describes some techniques for performing analysis across data stored in multiple columns. For example, you may want to analyze combinations of height and weight. Some options:
Consolidate dimensions to a single metric. For example, height and weight can be combined using a BMI (body mass index) calculation. Then, use available outlier analysis capabilities in Trifacta® Self-Managed Enterprise Edition. Below, you can review a method for bringing together similar data from multiple columns into a single column for easier analysis.
- Flag outlier values of individual columns, perhaps giving each column a weighting factor (e.g. 0.5). Sum the outliers and their weights together.
- Defer analysis until the data has arrived in the target system.
- Build a custom function in another programming language. See User-Defined Functions.
In some cases, you may need to identify outliers across multiple columns of data. For example, you have a dataset containing scores from three separate tests taken by a set of individuals. Your columns may look like the following:
You can download the Dataset-TestScores.csv dataset.Most calculations, such as standard deviation, work for a single column of data. To perform analysis across all three columns, you must reshape the above dataset to look like the following:
This steps below outline the workflow for this example. The full recipe is provided at the bottom of this section.
- Load the TestScores dataset into the Transformer page. It should already be split out into five separate columns.
The three columns listed side by side are data that has been organized in a pivot table. To break down this data, you must unpivot the data, which breaks down the data into a
TestScore1) and a
valuecolumn, which contains individual test scores.
Parameter: Group size
- Rename the generated column of test scores to
The numeric information in the
keycolumn values can be extracted using the following:
Extract text or pattern
Parameter: Column to extract from
Custom text or pattern
Parameter: Text to extract
key2column contains just the numeric data now. Rename this column to
TestNumber. You can delete the
The dataset does not contain a primary key, which field containing a unique identifier for each row. The combination of last name, first name, and test number is a unique identifier for each row in the dataset:
- Rename the new column to
TestID. Typically, primary keys are listed as the first field in a dataset. You might want to move the column before the
You may have noticed that the data is still organized by name (first and last) and test number, so that an individual's tests are scattered throughout the dataset. To reorganize the information, you can re-aggregate the data using the following:
Parameter: Row labels
Parameter: Max number of columns to create
Tip: The above retains all instances of tests that have been taken. If you are only interested in the average test score, you can remove the
TestIDgroupings and the change the
AVERAGE. In the results, you have one average for each test taker.
- You may want to rename the aggregation column. Your final dataset should look like the following:
Now that your columns of data have been consolidated to a single column, you can use the single-column transforms and functions to perform analysis.
For more information on identifying outliers in this data, see Locate Outliers.
This page has no comments.