This Power Monday trick is about random sample with Power Query. This is based on my experience of working with large volumes of data.
The other day I have been building a hotel dashboard (more on this later). As part of the dashboard, I wanted to show a random sample of user reviews. Reviews database had quite a few rows, so I wanted to extract a randomized sample of 100 reviews and show them in the report. When you refresh the report (Data > Refresh), then a new set of reviews will be fetched and shown.
Let’s learn how to generate a random sample with Power Query in this article.
This tutorial works in Power Query for Excel or Power BI. In case of Excel, the output sample will be either loaded as table or to data model. In case of Power BI, output goes to your data model.
If you want to get random sample with Excel formulas, read this.
5 Steps to create random sample with Power Query
Step 1: Get your data to Power Query
Simple. Grab the data you want to sample and bring it to PQ. At this point, you will get something like this:
Step 2: Add Random Numbers as a column
Go to “Add Column” > Custom Column and add this formula.
=Number.Random()
Remember: Power Query formulas are case-sensitive. So type exactly. Name this column “Random”
But Power Query gives same random number in all rows …
That is right. As Power Query is a parallel language, each row gets same random number (unlike Excel’s RAND() filled down a column).
Note: your experience with Number.Random() could be different, but as you build transformations, at some point PQ will replace all numbers with same value.
So how to get different numbers per row? Simple, we force PQ to evaluate something per row. A simple thing like index number column will do. This will force PQ to run random formula for all rows.
Hat tip to Gil Raviv for suggesting this technique in a forum post.
Step 3: Add Index Number column & Sort the random numbers
Go to “Add column” > Index number. Now that we have index numbers in a column, this will force PQ to regenerate the random number per row.
Select the random number column and sort it.
Note: You may need to switch Steps 2 & 3 if the random numbers are same all the way thru.
Step 4: Keep top 100 rows
Go to Home > Keep Rows > Keep Top Rows. Enter the sample size you want (100) and Click OK. Your sample is ready.
Step 5: Remove the Random & Index columns
Now that our sample is ready, let’s remove the random & index number columns. We do not need them in the final output (or model). Click on Save & Load (or Close & Apply).
Enjoy the sample.
How to get random sample with repetitions?
The above technique gives a sample without repetitions. What if you need a sample with repetitions (ie memory-less sampling). For example, a series of dice throws or coin tosses?
We can use Power Query to get such samples too. This is slightly complicated compared to first technique, but fun to try.
- Load your source to PQ
- Group the data so you can get row count (while still keeping the data). Like this:
- Add a custom column with a list of 100 numbers =List.Numbers(1,100)
- Expand the list to new rows
- Add a column with random number between 0 & row count-1 =Number.RandomBetween(0,[Count]-1))
- Add index column
- Change random number to whole number
- Extract the random row number from [Data] to a new column =[Data]{[Random]}
- Remove all other columns except this new column in #8
- Expand the column
- Your sample with possible repetitions is ready.
Here is the full M code for you to customize.
let Source = Excel.CurrentWorkbook(){[Name="myData"]}[Content], #"Grouped Rows" = Table.Group(Source, {}, {{"Count", each Table.RowCount(_), type number}, {"Data", each _, type table}}), #"Added Custom" = Table.AddColumn(#"Grouped Rows", "List", each List.Numbers(1,100)), #"Expanded List" = Table.ExpandListColumn(#"Added Custom", "List"), #"Added Custom1" = Table.AddColumn(#"Expanded List", "Random", each Number.RandomBetween(0,[Count]-1)), #"Added Index" = Table.AddIndexColumn(#"Added Custom1", "Index", 0, 1), #"Changed Type" = Table.TransformColumnTypes(#"Added Index",{{"Random", Int64.Type}}), #"Added Custom2" = Table.AddColumn(#"Changed Type", "Custom", each [Data]{[Random]}), #"Removed Columns" = Table.RemoveColumns(#"Added Custom2",{"Data"}), #"Removed Columns1" = Table.RemoveColumns(#"Removed Columns",{"Count", "List", "Random", "Index"}), #"Expanded Custom" = Table.ExpandRecordColumn(#"Removed Columns1", "Custom", {"Review Text", "Rating"}, {"Review Text", "Rating"}) in #"Expanded Custom"
Answers to your questions about sampling…
How to get another sample?
Simple. Just refresh your Power Query connection. You will get another sample.
How to change the sample size?
In the M code, where it says 100 replace with another number or parameter.
Use Excel Cell to tell Power Query how big a sample you want…
You can even use an Excel named cell to tell PQ what sample size you want. Assuming named cell sample.size has the size, use this M code =Excel.CurrentWorkbook(){[Name=”sample.size“]}[Content][Column1]{0} to get the value in your query. Use it as part of other steps and bingo, your sample size changes.
Other questions…?
Struggle sampling some sensible set? Post your sample problem in comments so I or one of our excellent readers can help you.
Download sample file and get your samples…
Excuse the pun, but here is a sample file with all the M code for making your own samples. Examine the queries to learn how this is done.
How do you sample?
Excel’s Rand() is my favorite way to sample. But now that I am spending more time with Power Query & Power BI, I needed another way to sample the data. This post outlines my preferred approach (unless I am dealing with very large volumes of data) For large volumes of data, I suggest sampling at server-side thru SQL.
What about you? How do you sample? Share your approach or troubles in the comments.
30 Responses to “Rescue oddly shaped data – Battle between Formulas, VBA and Power Query”
Nice use of Power Query! Power Query is simply awesome! But somehow a lot of people are punishing themselves by not using it (not learning it).
An imperfect 4th approach for consideration... no codes at all...
Select myrange.
Go to Special --> Blank
Delete Cell --> Shift cell left
90% done... now we just need to move the data of 2nd column to the bottom of 1st column
Of course... Power Query is the best.
Cheers,
There is another way but it involves multiple steps:
Copy the values in column E, move the cursor to F5, Paste Special with Skip Blanks, OK
Copy the values in column D, move the cursor to F8, Paste Special with Skip Blanks, OK
And so on.
This works perfectly, albeit a little clumsily apart from the values in B17 and C16, which can be moved with simple copy and paste
Power Query Forever! I do not know how I survived for so long without knowing and using this tool, I can not recommend it to my colleagues, but by the way they prefer to suffer to learn.
My congratulations here from Brazil.
I rolled my eyes when I saw that data
Using decimal places is a nice trick to order data, thanks for that
And tweaking the first formula a bit, you can use OFFSET instead of INDIRECT
=OFFSET($A$1, MIN(IF(myrange, ROW(myrange)), ROWS(A$1:A1))-1, RIGHT(TEXT(MIN(IF(myrange, ROW(myrange) + COLUMN(myrange)*0.00001), ROWS(A$1:A1)), ".00000"), 5)-1)
Tried the above formula with the downloaded oddly shaped data file and I could not get it to work. I get #value without ctrl+shift+enter, and #ref with ctrl+shift+enter.
Sorry, it was SMALL, not MIN.
Add with CTRL+SHIFT+ENTER.
Thank you for your formula. Like the indirect formula I tested this one in older versions of EXCEL and it worked without ALTERATION in EXCEL 95. Very impressive.
Too complicated
Use =Sum to summarize all the sells to the left and Bobs Your Uncle
@Bertie... I am afraid that won't work when you have more than one value in a row.
I tested this formula in versions of Excel all the way back to Excel 95
=IF(ISERROR(INDIRECT("R"&SUBSTITUTE(TEXT(SMALL(IF(MyRange"",ROW(MyRange)+COLUMN(MyRange)*0.00001),ROWS(A$1:A9)),"00000.00000"),".","C"),FALSE)),"",(INDIRECT("R"&SUBSTITUTE(TEXT(SMALL(IF(MyRange"",ROW(MyRange)+COLUMN(MyRange)*0.00001),ROWS(A$1:A9)),"00000.00000"),".","C"),FALSE)))
So there are multiple ways of cleaning up messy data by formulas.
Wow.. Excel 95. Who knew people still use that. But as you have shown, Excel has all these beautiful and powerful functions for 23 years. It has data sciency stuff before DS was even a thing.
I had a problem with pasting the formula in the original post.
Formula should be: =IF(ISERROR(INDIRECT("R"&SUBSTITUTE(TEXT(SMALL(IF(myrange"",ROW(myrange)+COLUMN(myrange)*0.00001),ROWS(A$1:A1)),"00000.00000"),".","C"),FALSE)),"",(INDIRECT("R"&SUBSTITUTE(TEXT(SMALL(IF(myrange"",ROW(myrange)+COLUMN(myrange)*0.00001),ROWS(A$1:A1)),"00000.00000"),".","C"),FALSE)))
EXCEL even in a 16 bit version, is a very robust and capable program.
I don't like the VBA code. If you have a blank row in MyRange, the last entry in the range is doubled up in the paste.here range.
Not really. The macro is writing one cell at a time from paste.here. You have to clean the range before, which I was too lazy to write. But a line like Range(range("paste.here"), range("paste.here").end(xldown)).clearcontents should do the trick.
Adding Range(range("paste.here"), range("paste.here").end(xldown)).clearcontents fixed the problem.
for step split column by delimiter i am not getting option of split into rows or columns. Can you help me in this
Thanks Chandoo for promoting Power Query.
To simplify further, you can "Unpivot Columns" instead of right click on the newly created column and split it by comma in to rows in step 3 of Power Query.
i used
=LOOKUP(10000,B5:F5)
and got the answers. I just plagiarized this formula somewhere and use it, maybe you can explain why it works.
Regards
@Johan... I am not sure if the formula works correctly. When I tested it with the sample data in this post, it showed #N/As in two cells. Essentially, it will only give first value in each row. So if a row has multiple values, then subsequent values are missed. LOOKUP() function goes thru a list and finds the first value that is less than or equal to the input - in this case 10000 in B5:F5.
I have the need to convert pdf's to excel on occasion and they often come out a mess like this. I have used:
Cell G2 =COUNT(myrange)
Cell G3 =IFERROR(IF(G2-1<1,"",G2-1),"") copied down to G100
Cell H2 =IFERROR(LARGE(myrange,G2),"") copied down to H100
Waouw...
=IFERROR(INDIRECT("R" & SUBSTITUTE(TEXT(SMALL(IF(myrange "", ROW(myrange) + COLUMN(myrange)*0.00001),
ROWS(A$1:A1)), "00000.00000"), ".", "C"), FALSE), "")
but CTRL Shift Enter with {} before and after 🙂 😀
Here's a way with pivot table
https://www.bookkempt.com/2018/02/aligning-non-contiguous-data.html
This is brilliant. Bookmarked 🙂
Another possibility.
This assumes that you have a row index 'k' to use in the SMALL function and a column index 'h' to identify the columns of 'myRange'.
If you define 'coord' to refer to
=k+h/10 [assuming h<10]
then it will be possible to recover values later based upon location within 'myRange'. The formula 'nb' that identifies non-blanks by coordinates is given by
= SMALL( IF(myRange"", coord), k )
Finally, to unpick the pieces
= INDEX( myRange, INT(nb), 10*MOD(nb, 1) )
Whilst I am here and making trouble the PQ solution is also a tad over-complicated. All that is needed is to unpivot the entire table and remove the Attribute column.
The advanced editor would show
let
Source = Excel.CurrentWorkbook(){[Name="myRange"]}[Content],
#"Unpivoted Columns" = Table.UnpivotOtherColumns(Source, {}, "Attribute", "Value"),
#"Removed Columns" = Table.RemoveColumns(#"Unpivoted Columns",{"Attribute"})
in
#"Removed Columns"
1.fill the blank cells with 0
2.the requested column value=sum of those mess number column
but this can be used in only one column has value
Chandoo
And if we use the formula SEARCH (100000000, B5: F5)
JC
Another approach with Power Query, it will still work if the number of columns changed:
let
Source = Excel.CurrentWorkbook(){[Name="myrange"]}[Content],
#"Added Custom" = Table.AddColumn(Source, "List", each Record.ToList(_)),
#"Removed Other Columns" = Table.SelectColumns(#"Added Custom",{"List"}),
#"Expanded LIst" = Table.ExpandListColumn(#"Removed Other Columns", "List"),
#"Filtered Rows" = Table.SelectRows(#"Expanded LIst", each ([List] null))
in
#"Filtered Rows"
Cool idea to use Record.ToList as added column. Thanks for sharing this.
Nowadays, you can just use TOCOL on Excel 2024, MS 365, and Web Excel. It has a parameter to ignore blanks/errors/both.