Overview of Pattern Matching
Dataprep by Trifacta utilizes columnar pattern matching to identify data patterns of interest to you and to surface them in the interface for use in building your recipes. Additionally, in your recipe steps, you can apply regular expressions or Wrangle to locate patterns and transform the matching data in your datasets.
Overview
A pattern is a combination of abstracted character sets and literal characters that can summarize data patterns in a column. Patterns can be applied through one of two methods:
Regular expressions are a standardized method of matching data. The syntax of regular expressions is both powerful and not easy to understand.
Alteryx patterns are pattern-matching widgets that provide a layer of abstraction on top of regular expressions. Instead of having to specify the sometimes complex underlying regular expression, you can specify a simple token to represent the underlying expression.
Tip
While regular expressions are a widely used standard, Wrangle are powerful simplifications that can limit the sometimes "greedy" matching issues in regular expressions.
For more information on the supported patterns, see Text Matching.
This section provides an overview of the pattern matching features of the platform.
Example Patterns
Within a row, multiple patterns may be applied at different levels of abstraction to describe the data in all fields (columns) of the row. Suppose you have two records like the following:
[cz.laping@gmail.com,3987,1446319063821] [ajuneauk@gmail.com,5289,1447275151508]
The above records can be described by any of the following patterns:
[{alpha-numeric}+,{4-digits},{13-digits}]
[{email},{4-digits},{13-digits}]
[{alpha-numeric}+@gmail.com,{4-digits},{13-digits}]Note
The above patterns utilize the syntax of Wrangle . Regular expressions can be used to describe them as well.
In the above case, all three pattern sets capture the data completely. However, please note the differences between the patterns for column 1:
Pattern | Description |
|---|---|
{alpha-numeric}+ | This pattern captures alpha-numeric values of one or more characters. So, entries that match on this pattern do not need to be valid email addresses. |
{email} | This pattern ensures matching only on valid email addresses. So, values that do not match this pattern are likely to be flagged as mismatched within the platform. |
{alpha-numeric}+@gmail.com | This partial pattern ensures that the only matches are from gmail.com. |
Depending on the specific meaning of the data for your use, any of the above may apply.
Patterns in the Platform
Column Profiling
Pattern matching applied to columns can permit users to see the most common patterns and anomalous patterns of data in a column across the entire sample. Since patterns presented to the user encompass the entire set of values in the sample, you can gather detailed information about the consistency of data in the column across the column.
Tip
Column pattern profiling is especially useful after you have addressed the mismatched values in the column.
Based on the patterns surfaced for the column, you can take any of the following actions:
Filtering a subset of records. For example, you can review patterns for a column of addresses and filter the rows of data where no street number is provided, based on patterns you select.
Standardize values. You can make selections of patterns for the different patterns for phone numbers. See Pattern Matching by Data Type below.
Extract values. You can break apart column values based on mismatches in structure. For example, apartment numbers from an address field can be extracted into a new column.
Variable levels of abstraction. As demonstrated in the previous example, you may be able to select from multiple matching patterns to determine which one is the best fit for the row values of interest.
Machine Learning
Additionally, Dataprep by Trifacta collects aggregated information about patterns applied by all users. These patterns are given weight in the set of suggested patterns presented to each user.
Pattern Matching by Data Type
As part of pattern matching, the platform evaluates the data against the specified data type for the column. Type-specific pattern matching applies to the following data types:
Datetime
Phone
Using Patterns
In the application, patterns can be used as the starting point in building your next recipe step, and you can modify or iterate on a pattern definition to preview the results of the specified transformation. Patterns are used in the following actions:
Select text to trigger a pattern-based suggestion or suggestions
Select patterns of varying level of abstraction to modify column data
Selecting Data
When you select a value in the data grid, your options include pattern-based suggestions. In this manner, you indicate something of interest and enable the platform to interpret your specific interest or broader goal for the selected data. These broader changes are surfaced as pattern-based suggestions in the context panel.
See Explore Suggestions.
For more information on how the platform predicts suggestion cards based on selection, see Overview of Predictive Transformation.
Patterns in Column Details
In the Column Details panel, you can review sets of patterns that describe subsets of the values in the column. When you select one of the patterns, you are prompted with a set of suggested transform steps to apply to the data. See Column Details Panel.
Advanced Uses
In addition to the above basic uses, patterns can be used as the basis for the following advanced uses and more.
Use | Description |
|---|---|
Standardize records | Match values based on a pattern and then change values to fit this pattern. See Standardize Using Patterns. |
Filter records | Keep or delete records based on patterns of values found in row data. See Filter Data. |
Extract values | Extract values matching a pattern from one column and insert them into a new column of data. See Extract Values. |
Generate function outputs | Use patterns to generate function outputs in new columns. |
User-Defined Patterns
In your recipe steps, you can specify patterns using either of the following methods.
Regular expressions
Regular expressions (regexes) are sequence of characters that can be used to define a pattern. This pattern can be used in the transformations that support regex to identify patterns in your data of interest to you. Example:
replace col: myCol with:/$1/ on:/^\((\d\d\d)\)/ global: false
In the above step, the matching pattern expressed in the on clause evaluates in the following manner:
The forward slashes around the pattern indicate that it is a regular expression.
^indicates the start of the value in themyColcolumn. So, the matching is only made at the beginning of the column.\(and\)are representations in regular expressions of the literal values for parentheses. So, matches are made on those specific characters.The interior set of parentheses are used to define a capture group of values. These values, which correspond to three digits, are captured and inserted as the replacement.
So, the net effect is to search the beginning of a field for values like (555) and replace them with just the digits: 555. This replacement removes the parentheses from the area code part of a phone number.
Note
Regular expressions are very powerful tools for matching patterns. They can also cause unexpected results. Use of regular expressions is considered a developer-level skill. You should use the Wrangle described below instead.
Dataprep by Trifacta implements a version of regular expressions based off of RE2 and PCRE regular expressions.
Patterns
Use Wrangle to quickly assemble sophisticated patterns to match in your data. The following example includes the equivalent Wrangle as the previous regular expression:
replace col: myCol with:`$1` on:`^\(({digit}{3})\)` global: falseThe back-ticks around the pattern indicate that it is a Wrangle .