Please note javascript is required for full website functionality.
MVP

Blog

Power Query: Finding Fuzzy

12 June 2019

Welcome to our Power Query blog.  After last week’s adventure in Power BI, I discover the fuzzy features of Excel Power Query.

 

Last week I looked at how the ‘Merge’ window in Power BI Power Query includes options for fuzzy matching, as the following screenshot shows:

In Excel’s Power Query, the ‘Merge’ screen does not have these options:

If this was the end of the story, this would be a very short blog!  The fuzzy merge options do actually exist, but to get to them, I have to use M.  If I type

=  Table.FuzzyJoin

in the ‘Next Step’ window in the Power Query Editor, I can see that the functionality is available.

The M function Table.FuzzyJoin() is described in full, and if I scroll down, I can see what parameters are required to use this M function.

The description of Table.FuzzyJoin() in the Microsoft help pages is as follows:

Table.FuzzyJoin(table1 as table, key1 as any, table2 as table, key2 as any, optional joinKind as nullable number, optional joinOptions as nullable record) as table


This joins the rows of table1 with the rows of table2 based on a fuzzy matching of the values of the key columns selected by key1 (for table1) and key2 (for table2).

Fuzzy matching is a comparison based on similarity of text rather than equality of text.

By default, an inner join is performed, however an optional joinKind may be included to specify the type of join.  Options include:

  • JoinKind.Inner
  • JoinKind.LeftOuter
  • JoinKind.RightOuter
  • JoinKind.FullOuter
  • JoinKind.LeftAnti
  • JoinKind.RightAnti

An optional set of joinOptions may be included to specify how to compare the key columns.  Options include:

  • ConcurrentRequests
  • Culture
  • IgnoreCase
  • IgnoreSpace
  • NumberOfMatches
  • Threshold
  • TransformationTable

These options are the same as the options available for fuzzy matching on the Power BI ‘Merge’ screen.

    ConcurrentRequests (default is 1)            

    Values can be between 1 and 8.                 

    The ConcurrentRequests option supports parallelizing the join operation by specifying the number of parallel threads to use.

     

    Culture (default is neutral)

    Values are valid culture names.                   

    The Culture option allows matching records based on culture-specific rules. 
    For example, a Culture option of 'ja-JP' matches records based on the Japanese language.

     

    IgnoreCase (default is true)                        

    Values are true or false.  

    The IgnoreCase option allows matching records irrespective of the case of the text. 
    For example, 'Grapes' (sentence case) is matched with 'grapes' (lower case) if the IgnoreCase option is set to true.

     

    IgnoreSpace (default is true)

    Values are true or false.  

    The IgnoreSpace option allows combining text parts in order to find matches. 
    For example, 'Micro soft' is matched with both 'Microsoft' and 'Micro soft' if the IgnoreSpace option is set to true.

     

    NumberOfMatches (default is 2,147,483,647)

    Values are between 0 and 2,147,483,647.  

    The NumberOfMatches option specifies the maximum number of matching rows that can be returned.

     

    Threshold (default is 0.80)                         

    Values are between 0.00 and 1.00.

    The similarity Threshold option provides the ability to match records above a given similarity score.  A Threshold of 1.00 is the same as specifying an exact match criterion. 
    For example, 'Grapes' matches with 'Graes' (missing 'p') only if the Threshold is set to less than 0.90.

     

    TransformationTable                                

    A valid table with at least two columns named 'From' and 'To'.      

    The TransformationTable option allows matching records based on custom value mappings.  For example, 'Grapes' is matched with 'Raisins' if a TransformationTable is provided with the 'From' column containing 'Grapes' and the 'To' column containing 'Raisins'. 

There is another very similar option, Table.FuzzyNestedJoin() – the main difference is that if I use Table.FuzzyNestedJoin(), the second table that I have joined to won’t have been expanded for me.  I will show an example later.

I will use Table.FuzzyJoin() to link ‘Contacts’ to the ‘Company’ query.  A reminder of the data from the Excel worksheet is shown next:

I will start from a new blank query by choosing the ‘New Query’ option from the ‘Get & Transform’ section of the Data tab and selecting ‘Blank Query’ in the ‘From Other Sources’ option.

I have created some M code which uses the Table.FuzzyJoin() function:

= Table.FuzzyJoin(Contacts, "Company", Company, "Actual_Company", JoinKind.LeftOuter, [Threshold = 0.5 ])

I have created a fuzzy join between Contacts and Company – I have used a left outer join so that the correct version of the company name will appear.  The threshold I have chosen is the same one I found worked last week for my data.  Other than the specified options, I have taken the defaults.  One point to note, is that this function will not create a table where two columns have the same name – so I had to change the name of the column in Company to Actual_Company so that I could join it to a table that already had the name Company.

I can also use Table.FuzzyNestedJoin(), if I want to keep my second table in a separate column.  For some reason, in this case the join is represented by a number – I have given the join type a value of 1 which is a left outer join type.

The M code is

= Table.FuzzyNestedJoin(Contacts, "Company", Company, "Actual_Company", "Actual Company", 0, [Threshold = 0.5 ])

I have named the column that holds my table; this is a required parameter. 

The advantage of this method is that the tables can have the same column names and it can be managed by Power Query when the table is expanded.

Once expanded, I get the same results.  I can also choose to use a transformation table as one of my options, which I will set up in the Excel workbook.

I am going to use ‘Company_Translation’ to get the head company name, which will allow me to link from ‘Contacts’ to ‘More_Companies

The M code is

= Table.FuzzyNestedJoin(Contacts, "Company", Company, "Actual_Company", "Actual Company", 0, [Threshold = 0.5 ])

I have named the column that holds my table; this is a required parameter. 

Come back next time for more ways to use Power Query!

Newsletter