Data Transformation Strategies
Within any BI project, it is essential that the data you are working with has been properly scrubbed to ensure accurate results on your reports and dashboards. Applying data cleansing business rules, also known as transforms, is the primary method for correcting inaccurate or malformed data, but the process can often be the most time-consuming part of any corporate BI solution. However, the data transformation capabilities built into Power BI are both very powerful and user-friendly. Using the Power Query Editor, tasks that would typically be difficult or time-consuming in an enterprise BI tool are as simple as right-clicking on a column and selecting the appropriate transform for the field. While interacting with the user interface, the Power Query Editor automatically writes queries using a language called M behind the scenes.
Through the course of this chapter, you will explore some of the most common features of the Power Query Editor that make it so highly regarded by its users. Since one sample dataset cannot provide all the problems you will run into, you will be provided with several small, disparate examples to show you what is possible. This chapter will detail the following topics:
- The Power Query Editor
- Transform basics
- Advanced data transformation options
- Leveraging R
- AI Insights
- The M formula language
To get started, let's get familiar with the interface known as the Power Query Editor.
The Power Query Editor
The Power Query Editor is the primary tool that you will utilize for applying transformations and cleansing processes to your data. This editor can be launched as part of establishing a connection to your data, or by simply clicking Transform Data on the Home ribbon of the Power BI Desktop. When the Power Query Editor is opened, you will notice that it has its own separate environment for you to work in. The environment encapsulates a user-friendly method for working with all of the queries that you will define. Before you dive deep into the capabilities of the Power Query Editor, let's first start by reviewing the key areas of the Power Query Editor interface, as shown in Figure 2.1:
Figure 2.1: First view of the Power Query Editor
Following the numbered figures, let's review some of the most important features of the Power Query Editor:
- New Source: This launches the interface to establish your connection details, which is the same interface as the Get data button that you learned about in Chapter 1, Getting Started with Importing Data Options.
- The Queries pane: A list of all the queries that you have connected to. From here, you can rename a query, disable the load and modify report refresh capabilities, and organize your queries into groups.
- Query Settings: Within this pane, you can rename the query, but more importantly, you can see and change the list of steps, or transforms, that have been applied to your query. If you ever accidentally close this pane, you can relaunch it from the View menu.
- Advanced Editor: By launching the Advanced Editor, you can see the M query that is automatically written for you by the Power Query Editor.
- Close & Apply: Choosing this option will close the Power Query Editor and load the results into the data model.
With this basic navigation understood, let's start to discuss some of the basics of working with various transforms.
Applying data transformations within the Power Query Editor can be a surprisingly simple thing to do. However, there are a few things to consider as we begin this process. The first is that there are multiple ways to solve a problem. As you work your way through this book, the authors have tried to show you the fastest and easiest methods of solving the problems that are presented, but these solutions will certainly not be the only ways to reach your goals.
The next thing you should understand is that every click you do inside the Power Query Editor is automatically converted into a formula language called M. Virtually all the basic transforms you will need can be accomplished by simply interacting with the Power Query Editor user interface, but for more complex business problems there is a good chance you may have to modify the M queries that are written for you by the editor. You will learn more about M later in this chapter.
Finally, the last important consideration to understand is that all transforms that are created within the editor are stored in the Query Settings pane under a section called Applied Steps. Why is this important to know? The Applied Steps section has many features, but here are some of the most critical to know for now:
- Deleting transforms: If you make a mistake and need to undo a step, you can click the Delete button next to a step.
- Modifying transforms: This can be done with any step that has a gear icon next to it.
- Changing the order of transforms: If you realize that it is better for one step to execute before another one, you can change the order of how the steps are executed.
- Selecting previous steps: Clicking on any step prior to the current one will allow you to see how your query results would change one step earlier in the process.
With this understanding, you will now get hands-on with applying several basic transforms inside the Power Query Editor. The goal of these first sets of examples is to get you comfortable with the Power Query Editor user interface before the more complex use cases are covered.
Use First Row as Headers
Organizing column names or headers is often an important first task when managing your dataset. Providing relevant column names makes many of the downstream processes, such as building reports, much easier. Often, column headers are automatically imported from your data source, but sometimes you may be working with a more unique data source that makes it difficult for Power BI to capture the column header information. This walkthrough will show how to deal with such a scenario:
- Launch Power BI Desktop, and click Get data on the Home ribbon.
- Choose Excel, then navigate to and select Open on the
Failed Bank List.xlsxfile that is available in the book source files.
- In the Navigator window, select the table called Data, then choose Transform Data. When the Power Query Editor launches, you should notice that the column headers are not automatically imported. In fact, the column headers are in the first row of the data.
- To push the column names that are in the first row of data to the header section, select the transform called Use First Row as Headers from the Home ribbon as shown in Figure 2.2:
Figure 2.2: Leveraging the Use First Row as Headers transform
Once complete, you will see the first row of the dataset has been promoted to the column header area. This is a very common transform that you can expect to use often with flat files. Next, let's look at another commonly used transform, Remove Columns.
Often, the data sources you will connect to will include many columns that are not necessary for the solution you are designing. It is important to remove these unnecessary columns from your dataset because these unused columns needlessly take up space inside your data model. There are several different methods for removing columns in the Power Query Editor. This example will show one of these methods using the same dataset from the previous demonstration:
- Multi-select (Ctrl + click) the column headers of the columns you wish to keep as part of your solution. In this scenario, select the columns Bank Name, City, ST, and Closing Date.
- With these four columns selected, right-click on any of the selected columns and choose Remove Other Columns, as shown in Figure 2.3:
Figure 2.3: Selecting the Remove Other Columns transform
Once this transform is completed, you should be left with only the columns you need.
Another popular method for removing columns is clicking the Choose Columns button on the Home ribbon of the Power Query Editor. This option provides a list of all the columns, and you can choose the columns you wish to keep or exclude.
You can also select the columns you wish to remove; right-click on one of the selected columns and click Remove. This seems like the more obvious method. However, this option is not as user-friendly in the long run because it does not provide an option to edit the transform in the Applied Steps section like the first two methods do.
With any data cleansing tool, data type manipulation is critical and can help save you from many headaches later in the development of your solution. In the next section, you will learn about how to change data types.
Defining column data types properly early on in your data scrubbing process can help to ensure proper business rules can be applied and data is presented properly in reports. The Power Query Editor has various numeric, text, and date-time data types for you to choose from. In our current example, all of the data types were automatically interpreted correctly by the Power Query Editor, but let's look at where you could change this if necessary:
- Locate the data type indicator on the column header to the left of the column name.
- Click the data type icon, and a menu will open that allows you to choose whichever data type you desire, as shown in Figure 2.4:
Figure 2.4: Choosing a different data type
Another method you can use for changing column data types is to right-click on the column you wish to change, then select Change Type and choose the new data type. You should always be careful when changing data types to ensure your data supports the change. For instance, if you change a column data type to a Whole Number while it has letters stored in it, Power BI will produce an error.
If you want to change multiple column data types at once, you can multi-select the necessary columns, then select the new data type from the Data Type property on the Home ribbon.
Many of the transforms you will encounter in the future are contextually based on the column data types you are working with. For example, if you have a column that is a date, then you will be provided with special transforms that can only be executed against a date data type, such as extracting the month name from a date column.
Understanding how to properly set data types in Power BI is often the first step to using more exciting transforms. In the next section, you will learn how Power BI can read from an example you provide to automatically create transform rules.
Column From Examples
One option that can make complex data transformations seem simple is the feature called Add Column From Examples. Using Add Column From Examples, you can provide the Power Query Editor with a sample of what you would like your data to look like, and it can then automatically determine which transforms are required to accomplish your goal. Continuing with the same failed banks example, let's walk through a simple example of how to use this feature:
- Find and select the Add Column tab in the Power Query Editor ribbon.
- Select the Column From Examples button and, if prompted, choose From All Columns. This will launch a new Add Column From Examples interface:
Figure 2.5: Choosing the Column from Examples transform
- Our goal is to leverage this feature to combine the
STcolumns together. In the first empty cell, type
Barboursville, WVand then hit Enter. In Figure 2.5 you will notice that the text you typed has automatically been translated into an M query and applied for every row in the dataset.
- Once you click OK, the transform is finalized and automatically added to the overall M query that has been built through the user interface. The newly merged column will be added with the rest of your columns and you can optionally rename the column something more appropriate by double-clicking on the column header:
Figure 2.6: Adding Column from Examples
As you can see, the Add Column from Examples feature is great because you don't have to be an expert in which transforms are appropriate because Power BI will automatically choose them for you!
Sometimes, you may encounter scenarios where the Add Column From Examples feature needs more than one example to properly translate your example into an M query function that accomplishes your goal. If this happens, simply provide additional examples of how you would like the data to appear in different rows, and the Power Query Editor should adjust to account for outliers.
Now that you have learned some basic transforms, let's explore some more complex design patterns that are still used quite frequently.
Advanced data transformation options
Now that you should be more comfortable working within the Power Query Editor, let's take the next step and discuss more advanced options. Often, you will find the need to go beyond these basic transforms when dealing with data that requires more care. In this section, you will learn about some common advanced transforms that you may have a need for, which include Conditional Columns, Fill Down, Unpivot, Merge Queries, and Append Queries.
Using the Power Query Editor Conditional Columns functionality is a great way to add new columns to your query that follow logical
else statements. This concept of
else is common across many programming languages, including Excel formulas. Let's review a real-world scenario where you would be required to do some data cleansing on a file before it could be used. In this example, you will be provided with a file of all the counties in the United States, and you must create a new column that extracts the state name from the county column and places it in its own column:
- Start by connecting to the
FIPS_CountyName.txtfile that is found in the book files using the Text/CSV connector.
- Launch the Power Query Editor by selecting Transform Data, then start by changing the data type of
Column1to Text. When you do this, you will be prompted to replace an existing type conversion. You can accept this by clicking Replace current.
- Now, on
Column2, filter out the value UNITED STATES from the column by clicking the arrow next to the column header and unchecking UNITED STATES. Then, click OK.
- Remove the state abbreviation from
Column2by right-clicking on the column header and selecting Split Column | By Delimiter. Choose -- Custom -- for the delimiter type, and type , before then clicking OK, as shown in Figure 2.7:
Figure 2.7: Splitting a column based on a delimiter
- Next, rename the column names
Column 2.1, and
Column 2.2, to
County Name, and
State Abbreviation, respectively.
- To isolate the full state name into its own column, you will need to implement Conditional Column. Go to the Add Column ribbon and select Conditional Column.
- Change the New column name property to
State Nameand implement the logic
If State Abbreviation equals null Then return County Name Else return nullas shown in Figure 2.8. To return the value from another column, you must select the icon in the Output property, then choose Select a column. Once this is complete, click OK:
Figure 2.8: Adding a conditional column
This results in a new column called State Name, which has the fully spelled-out state name only appearing on rows where the
State Abbreviation is
Figure 2.9: End result of following these steps
This is only setting the stage to fully scrub this dataset. To complete the data cleansing process for this file, read on to the next section about Fill Down. However, for the purposes of this example, you have now learned how to leverage the capabilities of the Conditional Column transform in the Power Query Editor.
Fill Down is a rather unique transform in how it operates. By selecting Fill Down on a particular column, a value will replace all
null values below it until another non-null appears. When another non-null value is present, that value will then fill down to all subsequent
null values. To examine this transform, you will pick up from where you left off in the Conditional Column example in the previous section:
- Right-click on the State Name column header and select Transform | Capitalize Each Word. This transform should be self-explanatory.
- Next, select the State Name column and, in the Transform ribbon, select Fill | Down. This will take the value in the State Name column and replace all non-null values until there is another State Name value that it can switch to. After performing this transform, scroll through the results to ensure that the value of
- To finish this example, filter out any
nullvalues that appear in the State Abbreviation column. The final result should look like Figure 2.10, as follows:
Figure 2.10: End result of following these steps
In this example, you learned how you can use Fill Down to replace all of the
null values below a non-null value. You can also use Fill Up to do the opposite, which would replace all the
null values above a non-null value. One important thing to note is that the data must be sorted properly for Fill Down or Fill Up to be successful. In the next section, you will learn about another advanced transform, known as Unpivot.
The Unpivot transform is an incredibly powerful transform that allows you to reorganize your dataset into a more structured format best suited for BI. Let's discuss this by visualizing a practical example to help understand the purpose of Unpivot. Imagine you are provided with a file that contains the populations of US states over the last three years, and looks as in Figure 2.11:
Figure 2.11: Example data that will cause problems in Power BI
The problem with data stored like this is you cannot very easily answer simple questions. For example, how would you answer questions like, What was the total population for all states in the US in 2018? or What was the average state population in 2016? With the data stored in this format, simple reports are made rather difficult to design. This is where the Unpivot transform can be a lifesaver. Using Unpivot, you can change this dataset into something more acceptable for an analytics project, as shown in Figure 2.12:
Figure 2.12: Results of unpivoted data
Data stored in this format can now easily answer the questions posed earlier by simply dragging a few columns into your visuals. To accomplish this in other programming languages would often require fairly complex logic, while the Power Query Editor does it in just a few clicks.
There are three different methods for selecting the Unpivot transform that you should be aware of, and they include the following options:
- Unpivot Columns: Turns any selected columns, headers into row values and the data in those columns into a corresponding row. With this selection, any new columns that may get added to the data source will automatically be included in the Unpivot transform.
- Unpivot Other Columns: Turns all column headers that are not selected into row values and the data in those columns into a corresponding row. With this selection, any new columns that may get added to the data source will automatically be included in the Unpivot transform.
- Unpivot Only Selected Columns: Turns any selected columns' headers into row values and the data in those columns into a corresponding row. With this selection, any new columns that may get added to the data source will not be included in the Unpivot transform.
Let's walk through two examples of using the Unpivot transform to show you the first two of these methods, and provide an understanding of how this complex problem can be solved with little effort in Power BI. The third method mentioned for doing Unpivot will not be shown since it's so similar to the first option:
- Launch a new instance of the Power BI Desktop, and use the Excel connector to import the workbook called
Income Per Person.xlsxfound in the book source files. Once you select this workbook, choose the spreadsheet called Data in the Navigator window, and then select Transform Data to launch the Power Query Editor. Figure 2.13 shows what our data looks like before the Unpivot operation:
Figure 2.13: Example before Unpivot is performed
- Now, make the first row of data into column headers by selecting the transform called Use First Row as Headers on the Home ribbon.
- Rename the GDP per capita PPP, with projections column to Country.
- If you look closely at the column headers, you can tell that most of the column names are actually years and the values inside those columns are the income for those years. This is not the ideal way to store this data because it would be incredibly difficult to answer a question like, What is the average income per person for Belgium? To make it easier to answer this type of question, right-click on the Country column and select Unpivot Other Columns.
- Rename the columns
- To finish this first example, you should also rename this query Income. The results of these first steps can be seen in Figure 2.14:
Figure 2.14: Results of unpivoted data
This first method walked you through what can often be the fastest method for performing an Unpivot transform, which is by using the Unpivot Other Columns option. In this next example, you will learn how to use the Unpivot Columns method as well:
- Remain in the Power Query Editor, and select New Source from the Home ribbon. Use the Excel connector to import the
Total Population.xlsxworkbook from the book source files. Once you select this workbook, choose the spreadsheet called Data in the Navigator window, and then select OK. Figure 2.15 shows the dataset before Unpivot has been added:
Figure 2.15: Example before Unpivot is performed
- Like the last example, you will again need to make the first row of data into column headers by selecting the transform called Use First Row as Headers on the Home ribbon.
- Then, rename the column Total population to Country.
- This time, multi-select all the columns except Country, then right-click on one of the selected columns and choose Unpivot Other Columns as shown in Figure 2.16. The easiest way to multi-select these columns is to select the first column then hold Shift before clicking the last column:
Figure 2.16: Using the Unpivot Other Columns transform
- Rename the columns from Attribute and Value to Year and Population, respectively, to see the result showing in Figure 2.17:
Figure 2.17: Shows the final result of these steps
In this section, you learned about two different methods for performing an Unpivot. To complete the data cleansing process on these two datasets, it's recommended that you continue through the next section on merging queries.
Another common requirement when building BI solutions is the need to join two tables together to form a new outcome that includes some columns from both tables in the result. Fortunately, Power BI makes this task very simple with the Merge Queries feature. Using this feature requires that you select two tables and then determine which column or columns will be the basis of how the two queries are merged. After determining the appropriate columns for your join, you will select a join type. The join types are listed here with the description that is provided within the product:
- Left Outer (all rows from the first table, only matching rows from the second)
- Right Outer (all rows from the second table, only matching rows from the first)
- Full Outer (all rows from both tables)
- Inner (only matching rows from both tables)
- Left Anti (rows only in the first table)
- Right Anti (rows only in the second table)
Many of you may already be very familiar with these different join terms from SQL programming you have learned in the past. However, if these terms are new to you, I recommend reviewing Visualizing Merge Join Types in Power BI, courtesy of Jason Thomas in the Power BI Data Story Gallery: https://community.powerbi.com/t5/Data-Stories-Gallery/Visualizing-Merge-Join-Types-in-Power-BI/m-p/219906. This visual aid is a favorite of many users that are new to these concepts.
To examine the Merge Queries option, you will pick up from where you left off with the Unpivot examples in the previous section:
- With the Population query selected, find and select Merge Queries | Merge Queries as New on the Home ribbon.
- In the Merge dialog box, select the Income query from the drop-down selection in the middle of the screen.
- Then, multi-select the Country and Year columns on the Population query, and do the same under the Income query. This defines which columns will be used to join the two queries together. Ensure that the number indicators next to the column headers match, as demonstrated in Figure 2.18. If they don't, you could accidentally attempt to join on the incorrect columns.
- Next, select Inner (only matching rows) for Join Kind. This join type will return rows only when the columns you chose to join on have values that exist in both queries. Before you click OK, confirm that your screen matches Figure 2.18:
Figure 2.18: Configuring a merge between two queries
- Once you select OK, this will create a new query called
Merge1that combines the results of the two queries. Go ahead and rename this query
- You will also notice that there is a column called
Incomethat has a value of
Tablefor each row. This column is actually representative of the entire
Incomequery that you joined to. To choose which columns you want from this query, click the Expand button on the column header. After clicking the Expand button, uncheck Country, Year, and Use original column name as prefix, then click OK.
- Rename the column called
Income. Figure 2.19 shows this step completed:
Figure 2.19: Configuring a merge between two queries
- Finally, since you chose the option Merge Queries as New in Step 1, you can disable the load option for the original queries that you started with. To do this, right-click on the Income query in the Queries pane and click Enable load to disable it. Do the same thing for the Population query as shown in Figure 2.20. Disabling these queries means that the only query that will be loaded into your Power BI data model is the new one, called Country Stats:
Figure 2.20: Uncheck to disable the loading of this query into the data model
To begin using this dataset in a report, you would click Close & Apply. You will learn more about building reports in Chapter 5, Visualizing Data.
By default, merging queries together relies on exact matching values between your join column(s). However, you may work with data that does not always provide perfect matching values. For example, a user enters data and misspells their country as "Unite States" instead of
United States. In those cases, you may consider the more advanced feature called Fuzzy Matching. With Fuzzy Matching, Power BI can perform an approximate match and still join on these two values based on the similarity of the values. In this section, you learned how the Merge Queries option is ideal for joining two queries together. In the next section, you will learn how you could solve the problem of performing a union of two or more queries.
Occasionally, you will work with multiple datasets that need to be appended to each other. Here's a scenario: you work for a customer service department for a company that provides credit or loans to customers. You are regularly provided with
.xlsx files that give summaries of customer complaints regarding credit cards and student loans. You would like to analyze both of these data extracts at the same time but, unfortunately, the credit card and student loan complaints are provided in two separate files. In this example, you will learn how to solve this problem by performing an append operation on these two different files:
- Launch a new instance of the Power BI Desktop, and use the Excel connector to import the workbook called
Student Loan Complaints.xlsxfound in the book source files. Once you select this workbook, choose the spreadsheet called Student Loan Complaints in the Navigator window, and then select Transform Data to launch the Power Query Editor.
- Next, import the credit card data by selecting New Source | Text/CSV, then choose the file called
Credit Card Complaints.csvfound in the book source files. Click OK to bring this data into the Power Query Editor.
- With the
Credit Card Complaintsquery selected, find and select Append Queries | Append Queries as New on the Home ribbon.
- Select Student Loan Complaints as the table to append to, then select OK as shown in Figure 2.21:
Figure 2.21: Configuring an append between two queries
- Rename the newly created query All Complaints and view the results as seen in Figure 2.22:
Figure 2.22: Configuring an append between two queries
- Similar to the previous example, you would likely want to disable the load option for the original queries that you started with. To do this, right-click on the Student Load Complaints query in the Queries pane and click Enable load to disable it.
- Do the same to the Credit Card Complaints query, and then select Close & Apply.
Now that you have learned about the various methods for combining data, the next section will discuss a more advanced method of working with data using the R programming language.
R is a very powerful scripting language that is primarily used for advanced analytics tools, but also has several integration points within Power BI. One such integration is the ability to apply business rules to your data with the R language. Why is that important? Well, with this capability you can extend beyond the limits of the Power Query Editor and call functions and libraries from R to do things that would not normally be possible. In the next two sections, you will explore how to set up your machine to leverage R within Power BI and then walk through an example of using an R script transform.
There are many additional books and references you can read to learn more about the R scripting language, but for the purposes of this book, our goal is to inform you of what is possible when R and Power BI are combined.
Installation and configuration
To use R within Power BI, you must first install an R distribution for you to run and execute scripts against. In this book, we will leverage Microsoft's distribution, Microsoft R Open. It is an open source project and free for anyone to use. Once Microsoft R Open has been installed, you can then configure Power BI to recognize the home directory where R libraries may be installed. Let's walk through these setup steps together:
- Navigate to the website https://mran.microsoft.com/download/ to download and install Microsoft R Open.
- For the purposes of our example, you will select Download next to Windows.
- Once the download has completed, run the installation and accept all default settings and user agreements.
- Next, launch a new instance of Power BI Desktop to set up the R integration with Power BI. Click the menu options File | Options and settings | Options.
- Choose the R scripting section and ensure that the Detected R home directories property is filled with the R instance you just installed, as shown in Figure 2.23:
Figure 2.23: Mapping the R home directories in Power BI
- Once this is completed, click OK.
With this setup now complete, let's see how we can take advantage of R within Power BI.
The R script transform
With the R distribution now installed and configured to integrate with Power BI, you are now ready to see what's possible with these new capabilities. In this example, you will be looking at data from the European stock market. The problem with this dataset, which calls for it to be corrected with R, is that the file provided to you has missing values for certain days. So, to get a more accurate reading of the stock market, you will use an R package called
MICE to impute the missing values:
- Before beginning in Power BI, you should ensure that the
MICElibrary is installed and available in the R distribution you set up in the last section. To do this, launch Microsoft R Open from your device. This is the basic RGui that was installed for you to run R scripts with.
For many developers, the preferred method for writing
Rscripts is a free open source tool called RStudio. RStudio includes a code editor, debugging, and visualization tools that many find easier to work with. You can download RStudio from https://www.rstudio.com/.
- Type the following script in the R Console window, and then hit Enter:
This input is illustrated in the following screenshot:
Figure 2.24: Running the library install in RGui
- You can close the R Console window and return to Power BI Desktop after it returns an output like the following:
package 'mice' successfully unpacked and MD5 sums checked.
- In Power BI Desktop, start by connecting to the required data source called
EuStockMarkets_NA.csvfrom the book source files. Once you connect to the file, click Transform Data to launch the Power Query Editor.
- You will notice that there are a few days missing values in the SMI (Stock Market Index) column. We would like to replace values that show NA with approximate values using an R script. Go to the Transform ribbon, and select the Run R Script button on the far right.
- Use the following R script to call the
MICElibrary that you recently installed to detect what the missing values in this dataset should be:
# 'dataset' holds the input data for this script library(mice) tempData <- mice(dataset,m=1,maxit=50,meth='pmm',seed=100) completedData <- complete(tempData,1) output <- dataset output$completedValues <- completedData$"SMI missing values"
- Click OK. If you are prompted with a warning indicating Information is required about data privacy click Continue.
- Next, click on the hyperlink on the table value next to the
completedDatarow to see the result of the newly implemented transform for detecting missing values.
This new output has replaced the missing values with new values that were detected based on the algorithm used within the R script. To now build a set of report visuals on this example, you can click Close & Apply on the Home ribbon.
This is just one simple way that R can be used with Power BI. You should note that in addition to using R as a transform, it can also be used as a data source and as a visual within Power BI.
While this book highlights the programming language R to extend the capabilities of Power BI, some might prefer Python. Python is another programming language that allows for extensibility into Power BI to create new data connectors, transforms, and visuals. So, should you choose R or Python? That depends on which you are more comfortable with. If you have already spent time learning
Python, then stick with that! In the next section of this chapter, you will learn about Power BI's AI integration features, which give you the ability to call on components of Azure Cognitive Services with the Power Query Editor.
As you learned in the previous section, Power BI integrates and takes advantage of outside tools to enhance the capabilities within itself. That continues to be the case with the AI Insights features. Leveraging the AI Insights capabilities gives you the ability to tap into core features and algorithms within Azure Cognitive Services and expose them within Power BI. So how can this be useful to you?
Imagine you work for a company that runs a vacation rentals website. Customers can book travel and post reviews of their trips on your website. With thousands of customers and hundreds of rental homes, it can be difficult to manage all the reviews that come in to make sure your locations are all meeting the standards your customers expect. With AI Insights you can run algorithms that can perform sentiment analysis, key phrase extraction, language detection, and even image tagging. So, if you have international customers that post reviews, you can use language detection to understand what language the post was written in. Then you can use sentiment analysis to capture whether the review was positive or negative. Finally, using phrase extraction, you can pull out key terms in the reviews to see if the same locations continue to receive feedback regarding similar problems. Furthermore, if your feedback system allows photos to be posted in the reviews, the image tagging capabilities can return a list of characteristics found in the images posted. This would allow for automated categorization of images using AI.
As you can see, these are very powerful features that take your analytics processing to the next level. There are limitations, however, that you should be aware of before exploring these features. As of the time that this book was published, Cognitive Services integration is only supported for Power BI Premium capacity nodes
P1 and above. This means if your company is not currently leveraging Power BI Premium, then these features are not available to you.
Before using the AI Insights features in Power BI, you will need to change the capacity settings in the Power BI admin portal to enable the AI workload. After turning on the AI workload setting, you can also set the maximum amount of memory you would like to give the workload. The general recommendation is a memory limit of 20%.
In the next section, you will learn how to leverage an AI Insights Text Analytics feature called Sentiment Analysis.
Sentiment Analysis with Text Analytics
The Text Analytics features within the AI Insights features can be incredible time-savers. Imagine having to read paragraphs of information and conclude what was important or whether it was written in a positive or negative light. These are exactly the type of things that this feature can do for you. In this next example, you are going to test out one of these features by running a sentiment analysis algorithm on hotel reviews to see how customers feel about staying at your hotel locations:
- Launch a new instance of Power BI Desktop, and use the Excel connector to import the workbook called
Hotel Ratings.xlsxfound in the book source files. Once you select this workbook, choose the spreadsheet called Reviews in the Navigator window, and then select Transform Data to launch the Power Query Editor.
- Select Text Analytics on the Home ribbon of the Power Query Editor. If this is your first time using this feature, you may be prompted to sign into a Power BI account that has Power BI Premium capacity assigned to it.
- Next, you will be prompted to choose which Text Analytics algorithm you would like to use. Select Score sentiment, as shown in Figure 2.25, and ensure the ReviewText field is the Text that will be analyzed. Then click OK:
Figure 2.25: Using the Text Analytics feature
- If prompted with a data privacy warning, click Continue and then select Ignore Privacy Levels check for this file before clicking Save. This type of warning can occur when you combine two disparate sources or services together and is to ensure it is OK for these data sources to be combined.
This transform will produce a new numeric column with a value between
1 for every row in the dataset. A sentiment score of
.50 is considered neutral, while any score lower is negative and any score higher is generally positive:
Figure 2.26: Results of Score sentiment
Looking at Figure 2.26, it looks like the AI integration, with a few exceptions, did a good job determining how to rate each review.
Next, in the final section of this chapter, you will be introduced to the M formula language.
The M formula language
The Power Query Editor is the user interface that is used to design and build data imports. However, you should also know that every transform you apply within this editor is actually, quietly and behind the scenes, writing an M query for you. The letter M here is a reference to the language's data mashup capabilities.
For simple solutions, it is unlikely that you will ever need to even look at the M query that is being written, but there are some more complex cases where it's helpful to understand how to read and write your own M. For the purposes of this book, covering just the Power BI essentials, you will learn how to find the M query editor within your solution and then understand how to read what it is doing for you.
For the purposes of this example, you can open up any previously built example, however, the screenshot used here is from the very first example in this chapter on basic transforms:
- Using any Power BI solution you have designed, launch the Power Query Editor.
- On the Home ribbon, select Advanced Editor to see the M query that has been written by the user interface. Figure 2.27 shows an example of what your Advanced Editor might show:
Figure 2.27: Understanding the elements of M
This query has been formatted to make it easier to read. Let's review the key elements that are present here:
- The let expression: Encapsulates a set of values or named expressions to be computed.
- Named expressions or variables: The name given to a set of operations in a step. These names can be anything, but you should note that if you wish to have a space in the name of a step then it must be surrounded by
#"". For example, if I wanted something to be called Step 1, then I would have to name an expression
#"Step 1". If a space is not required in the name of your step then the double quotes are not required.
- M functions: The operations that are used to manipulate the data source.
- Prior step reference: The M query language generally executes its functions as serial operations, meaning each operation is executed one after the other sequentially. You can see this when you look at a query because each call to an M function always references the prior-named expression, to pick up where it left off.
- The in expression: Oddly, the
inexpression is actually a reference to what the query will output. Whichever named expression is referenced in the
inexpression will be what is returned back in the
Power Query Editorpreview.
It is important to realize that M is case-sensitive. That means if you ever make a change to a query or write one from scratch, you should be careful because there is a difference between "a" and "A."
As mentioned previously, this book will not dive deep into writing your own M queries since that would be far beyond the essentials of Power BI. However, there is a great method for exploring the M functions that are available, and how to use them. Within the Power Query Editor, you can use the
#shared function to return documentation on every available function in the M library. Let's walk through how you can leverage this tool:
- In a new instance of Power BI Desktop, select Get data and then choose Blank Query. This will launch the Power Query Editor with an empty formula bar waiting for you to provide your own M.
- In this formula bar, type
= #shared, then hit Enter. Remember that M is case-sensitive so you must use a lowercase
- This will return a list of all the available M functions. By selecting the cell that has the hyperlink text of a certain function, you can see documentation on how to use each function. Figure 2.28 shows an example of this:
Figure 2.28: Example of function documentation
This is a great method for learning what M functions are available, and how each may be used. If you are stumped on how to solve a problem using M then make this your first stop to explore what options you have.
In this chapter, you learned that the Power Query Editor is an extremely powerful tool for applying business rules to incoming data. Implementing data cleansing techniques can be as simple as right-clicking on a column, or more complex, such as when building a conditional column. While the Power Query Editor does have a vast library of transforms available, you also learned that you can tap into the capabilities of R to extend what's possible when designing queries. Finally, this chapter discussed the AI capabilities within the Power Query Editor that allow you to leverage several algorithms available within Azure Cognitive Services. In the next chapter, on building the data model, you will learn about proper techniques for building a well-designed Power BI data model to ensure your solutions can solve all your reporting needs.