I’ve been writing a fair bit lately about specific aspects of Power BI, and how they all fit together. I decided to put together a “real world” demonstration of as many aspects of the product as possible, and this article is the result. I don’t dive too deep on any of the specific areas, but hopefully this walkthrough will help give a sense of the power and simplicity of working with this (these?) tool (s).
In this example, we will explore some publicly available data published by the US National Oceanic and Atmospheric Administration (NOAA). To do so, we will first use Power Query to acquire and transform the data. Next, we will utilize Power Pivot to build our model, and then create a Power Pivot and a Power Map to visualize it. Finally, we will publish the data to a Power BI enabled site where it can be consumed by anyone with a browser or a mobile client.
Data Acquisition and Transformation with Power Query
The NOAA keeps a record from about 800 weather stations around the world. The data collected includes daily maximum and minimum temperatures, rainfall, and a host of other measures. All of this data is available through their web/ftp site here. Well it should be. As I write this, the site is offline, as are most other US government websites. Luckily, I have a local copy of all the files. Feel free to contact me for a copy if they’re not back soon. Update – For the interim, I’ve made the files available here.
The way that the data is provided is via a series of text files. Each text file in a folder represents a single weather station, and each row contains all the data for a given measure for a month. The data are not delimited, they are distinguished by their position in the row, with the data for each day in the month being presented as a separate measure. Normally, a tool like SSIS would be required to transform the data into something that could be analyzed, but as you’ll shortly see Power Query is up to the task.
Unfortunately, Power Query cannot work with FTP sources, but it’s a rather simple matter to copy the files from the FTP site to a local drive, and there are tools that can keep the content synchronized for refresh purposes. Note, if anyone from Microsoft is reading this, I believe that FTP and WebDAV folders would be a great supplement to the Folder data source, particularly when it comes to automatic refresh. Once we have the files locally, we can begin the import process. First, we open a new Excel workbook, select the Power Query tab, and then in the “From File” dropdown, we select the new feature “From Folder”.
Next, we select the folder that contains all of the data files. Power Query returns the data for the files in the folder.
In our case, the file name does correspond to the ID of the station, but the rest of the file metadata is not that useful to us. However, what is useful is the fact that using Power Query, we can drill into the files. Depending on the requirement, this can be done one of two ways. By clicking on the word “Binary” in the Content column, you will be able to drill into that file, and subsequent operations will be performed on that file. However, if you click on the little drilldown icon at the right of the Content column header, you will drill into all files in the folder. Subsequent operations will be performed on all files in the folder, and refresh operations will pick up any new files.
It should be noted that for this to work, all files in the folder must have the same schema. (In addition to data like this, this feature is particularly useful for parsing and iterating through server log files.) Since we want all station data, this is the operation that we will perform next. What we get back is a column of seemingly randomly structured data.
Luckily for us, there is a key to the data structure (found in the file readme.txt if you’re following along). The basic key pattern is:
Variable Columns Type
ID 1-11 Character
YEAR 12-15 Integer
MONTH 16-17 Integer
ELEMENT 18-21 Character
VALUE1 22-26 Integer
MFLAG1 27-27 Character
QFLAG1 28-28 Character
SFLAG1 29-29 Character
VALUE2 30-34 Integer
MFLAG2 35-35 Character
QFLAG2 36-36 Character
SFLAG2 37-37 Character
. . .
. . .
. . .
VALUE31 262-266 Integer
MFLAG31 267-267 Character
QFLAG31 268-268 Character
SFLAG31 269-269 Character
In plain English, this means that the first 11 characters are the ID (the station ID actually, from which the country can be extracted), the next 4 characters are the year, the next 2 the month, the next 4 the “element” (the measure name). The next 4 “variables” are the value (5 characters), and 3 different flags (one character each, which we are not interested in) for the first day of the month. This pattern of 4 variables is then repeated 30 more times, once for each day of the month. A value of –9999 represents a null value.
We have some parsing to do.
Luckily, Power Query makes this (relatively) easy. Right click on the column header, select split column, and then select number of characters.
Next, for our first column, we set the number of characters to 11, and we tell it to split once, as far left as possible. We’ll use this option for our remaining splits.
Click on OK, and we will now have 2 columns. We then right click on the leftmost column, and rename it to ID.
Next, we right click on our rightmost column header and repeat the process (using 4 characters this time), this time renaming it to Year. We then do the same for the month (2 characters) and element (4 characters) columns. Once this is done, we can click the dropdown on the elements column. Since we are only interested in the maximum and minimum daily temperatures, we deselect all, and reselect these two elements. This will limit the amount of data that we will pull into our model.
At this point, it is also worth noting the steps window on the right hand side of the query editor. This shows all of the transformation steps that have occurred to get us where we “are”. Any of these steps can be deleted, which is very useful if a mistake is made, and all subsequent steps will be affected by it. Clicking on a previous step takes you back to that point in the development of the transformation as well.
Once we have filtered our data, we continue on with the parsing. We then split off the next 5 columns, and rename the column to “1”. The next three columns we don’t need, so we first split them off (3 characters) and then delete the column by right clicking and selecting “Remove”. We then repeat this, renaming the value column to 2. Finally, we continue repeating this process until the value for 31 is complete. This is a tremendously tedious process, but it works rather well. I’ll make another feature request here for anyone in a position to do something about it. In the Power Query editor, it would be nice to be able to select a number of steps and either copy them forward, or be able to repeat them x times. This would save a great deal of time in these situations.
Its also a good idea to give the query an understandable name, so we right click on “Query 1” and rename it to “Temperature Data” .
As mentioned above, the figure –9999 represents a null value. Well, we know this, but the model won’t. At this point we need to convert these to proper nulls. The way that this is done is to select all of the numbered column headers, right click on them and then choose “Replace Value”. We replace the value –9999 with the word null.
Now that we have our nulls, our data is good. However, from an analysis perspective, the data is hardly in the best shape. What we need is to is to have each day’s data represented by a row, not a column. Once again, Power Query provides a method for us to accomplish this through the unpivot function. Simply select all of the numbered columns, right click, and choose “Unpivot Columns”.
Instantly, the numbered rows are transformed into rows, with the column titles in a column named “Attribute”. Simply rename it to “Day”, and we’re ready to import our data. To do so, click “Done” and Power Query will begin the extract, transform, and load process. In this case, it takes quite some time to iterate through all of the files (about 15 minutes) so be patient.
While the data is loading, you will need to make some adjustments to your Query Settings (to the right of your Excel window). Specifically, you’ll want to turn off the “Load to worksheet” option. As I’ve discussed previously this is important to keep the size of the actual worksheet small. in this case, it’s also important as the sheer volume of data exceeds Excel’s ability to store it. When complete, your worksheet should appear as follows.
You can see that there’s a fair amount of data here, about 31.5 million rows. Did I mention that we’re working with Excel here? We’ve disallowed loading to the worksheet, so at this point, we need to load into the embedded data model. Do that by clicking on the link at the bottom of the Query Settings screen.
There is also some more supporting data that we want to bring into our model. The following files need to be imported:
- ghcnd-stations.txt – contains metadata for all of the different weather stations
- ghcnd-countries.txt – contains metadata for all participating countries
- ghcnd-states – contains metadata for US states and Canadian provinces
The keys to these files are also in the readme.txt file, and we bring in their data using the same procedure that we used above for the actual weather data, with the exception that we use the “From file” data source. Finally, we give each query a logical name.
The beauty of all this is that anytime we wish, we can refresh the data. Any new or changed data will be picked up and added to the model following our transformation.
Model Editing In Power Pivot
Once this is complete, we are ready to edit our model. We do so by selecting the Power Pivot tab, and clicking “Manage” in the ribbon. First, we work on the actual temperature data tab. The temperature values reported are in tenths of a degree Celsius. This doesn’t mean much to most people, we should at least report the data in degrees C. To do this, Click on Add Column, and add the formula [Value]/10. Rename the column to Temperature.
We always want to be working with temperature averages. To do so, add a calculated measure named Average Temperature (C) (in the calculation area at the bottom). Its formula should be Average Temperature (C):=AVERAGE([Temperature]). We no longer need the value column or the temperature column for analysis, so we need to hide them from client tools. To do so, select both the Value header and the Temperature header, right click and select “Hide from Client Tools”. Finally, we set the formatting of our calculated measure to be Decimal Number.
Next, we want to establish a relationship between the data and the country table. We don’t have a column for country, but according to the data key, the first two characters of the country code correspond to the country code. We can therefore add another calculated column with the formula LEFT([ID],2). We then rename the column to Country Code. We only need this column for the relationship, so we can hide this from client tools as well.
To establish the relationship, change to the diagram view in the ribbon. Drag the Country Code field from the Temperature Data table onto the CountryCode field from the Countries table. While we’re in this view, we can drag the ID field from Temperature Data onto the ID field of the stations data. this will allow us to take advantage of the location of each of the stations. Finally, drag the State column from the Stations table onto the State Code column of the States table. At this point, we can go ahead and hide any unnecessary columns from client tools.
Our model is ready. At this point, we can move forward with analysis.
Create a Simple Power View Report
We want to create a simple Power View report to display temperature trends over time, Move back to the Excel Workbook and click the Insert tab in the ribbon. Then select the Power View button – this opens up a new Power View design surface. We won’t be needing the Filters window, so we can close that.
Right away, if you look at the Power View Fields window, you will notice that every table is duplicated – one with a “filled in” title bar, and one without. This window is displaying both defined Power Query queries, and the table contained within the model (Power Pivot). The “filled in” one is the one that we want to be working with. Using the other will cause duplicate tables to be created in the model – something that we obviously don’t want. This is due to an issue with the preview for Power Query and should be fixed in the near future.
We’ll be working with “Average Temperatures” as a measure so from the start, we’ll pick it and “Year” from the Temperature Data table. Year is a number, so Power View will want to total it by default. We don’t want this, so in the Fields selector, select the dropdown for year and select “Do Not Summarize”.
Next, resize the table on the design surface so that it fills the width of the report. Then, with the table selected, from the Design tab in the ribbon, select “Other Chart” and “Line”
What we get is a graph of overall average of all recorded temperatures starting around 1840. It might be more interesting to see the maximum and minimum temperatures, so to do that, just select the “Element” field, and Power View will add it to the Legend, showing both values, maximum temperature in blue, minimum in red. It would be nice if we could switch those colours, but unfortunately, all you can do at the moment is to select a different theme, none of which start with red.
Next, it would be nice to be able to see this data by individual reporting country. We can do this by first displaying a chart showing the overall average temperature per country. We click on the design surface outside of the temperature chart, and select the ”Country” field from the Country table, and the “Average Temperature” field from the Temperature Data table. Again,we resize the table to the width of the report, and then select “Column Chart”, “Clustered Column from the Design tab. We then get a scrollable list of countries with their overall average temperatures.
We could sort this by temperature, but since it will service as a picker, alphabetical makes more sense. Being from Canada, I’m naturally interested in our weather patterns, so I click on the Canada data bar (a negative value…shiver). Immediately the report above is filtered by Canadian data.
It seems that this whole climate change thing isn’t working out as well as I’d hoped….
Finally, we want to give the report a good title, and clean up some formatting. After we do this, we can move on to the next step, creating a Power Map report.
Create a Simple Power Map Report
Creating a Power Map is similar to a Power View. From the Insert tab on the ribbon, select Map, and then “Launch Power Map” from the dropdown.
At this point a new window will be launched where you can build your Power Map report. What we want to do is to show the relative temperatures for each country using a colour scale from low to high, and to show the average maximums and minimums for each recording station. To start with, for our first layer, we are going to use Country from the Countries table for our Geography data, so we select it. Every country with data will get a small point on the map.
Clicking next, we then get to select our data for the first layer. The data that we want to map is Average Temperature, So we select it. The resulting column visualization makes the earth look like a pincushion. What we want is to fill in the country boundaries with relative colour values. To do this, we select the region type for the visualization. This is better, but it fills the regions with shades of blue from light blue to dark blue for the warmest averages. It’s far more typical for red to represent heat, and we can do this by first selecting the settings icon.
We can then select a colour palette for our data, and red is a more logical color. Now you can spin the globe around to check out your next vacation hotspot. However, we want to get more specific than country, we also want to plot details for each reporting weather station. First, we need to add a layer. To do so, click on the Layers icon and select the add layer icon.
We then repeat the same steps that we did for the first layer, only instead of selecting country, we select Latitude and Longitude for the geography fields. After clicking Next, we leave the report as a Column type, and select Average Temperature as the height. Next, we select Element as the category option. We want the report to use clustered columns, so we now select that option to the right of the category title.
Finally, we want TMAX to be red, and TMIN to be blue, so we again click on the settings icon for this layer, and set the colours appropriately.
From this, it’s easy to see why Canada’s numbers are so low – there are a disproportionate number of stations in the far north! We’re now ready to share our report with Power BI.
Sharing Reports With BI Sites
Sharing your report is almost as simple as saving to a BI Sites enabled Office 365 library. We can either upload it directly, or save it to a SkyDrive Pro enabled folder, which is what I do. At this point, the file is 100 MB. On a side note…. this Excel file, that contains over 30 million rows of data is only 100 MB in size. That’s impressive. After the file appears in the O365 library, we can attempt to open it. However if we do, we’ll get a message saying that the file is to big to open in the browser.
That would be true in a regular Office 365 library, but it isn’t too large for Power BI. The problem is that first, the file needs to be enabled. More information of file size limits and enablement can be found here. To enable the workbook for Power BI, first navigate the the Power BI app, locate your file (it will be showing a small caution symbol), click the ellipsis, and select “Enable”.
Enabling the file will perform a few operations:
- The model will be extracted from the file and stored elsewhere, drastically reducing the file size
- A thumbnail is extracted from the file and used in the tile
It should be noted that if the file is uploaded via the Power BI app, the saving and enablement steps will be performed automatically.
Clicking on the file, either here, of within its actual library will now open the file, and we can also interact with the model (in this case by clicking on the data for Bermuda).
At the top, you’ll notice a warning message. It is letting you know that although this workbook can be viewed in a browser, and interacted with, it cannot be edited in a browser. This is because it is beyond the maximum edit size limit of 30 MB. You may also notice that we are using an HTML5 version of our Power View report. By default, Silverlight is used, but if you click on the new icon on the bottom left of the report, HTML5 will be used. It will also be used if you have a system that does not support Silverlight. This is true anywhere in Office 365, not just in Power BI enabled sites.
Finally, you may also notice that our Power Map report is missing from the equation. This is because for the moment, Power Map is not supported in the browser. A workaround for the time being is to export the Power Map report to an MP4 video file, and store it separately. MP4 files can be rendered in a browser, and on mobile devices.
Viewing In the Power BI Mobile Apps
Speaking of mobile devices the Power BI app for Windows 8 is currently available in the Windows store, with a version for iOS coming shortly. Once installed and run you’ll be shows a couple of sample reports from a demo site. To navigate to your own site, right click anywhere, or swipe up from the bottom, and select “Browse”.
On the next screen, we can select from our registered sites. If our site has not yet been registered, we can do so by again right-clicking, or swiping up. Then, we can enter the URL of your site. The URL to be entered here is that of the parent site of the library in which your report is stored. Once we load from this site, it will be automatically registered.
The next few screens allow us to navigate down to the file in question. Once the file appears in the list, selecting it (not opening it) gives us the option to save it as a favourite. If we mark a report as a favourite, it will appear on the home screen whenever the app is started.
Finally, opening it will open the first names object in the file. You can navigate to other named objects by right clicking, or swiping down from the top.
As you will notice, all names objects in the workbook appear by default. We can control what appears through the Excel publishing options, as outlined here.
As you can see, we have just taken a very large amount of publicly available data, transformed it and analyzed it within Excel, and published it out for everyone in the organization no matter what device they are using. I haven’t even gotten into the Q&A piece – primarily because it’s not available yet for custom data sets. I plan on updating this article when it is.
Power BI is powerful indeed.