Forum Discussion
Shaping data and calculate weighted average
- 3 years ago
Here's example code for normalising your raw data in PQ:
let Source = Table.FromRows(Json.Document(Binary.Decompress(Binary.FromText("i45WCkgtKs7PU3BU0lEyMgISbkBsCOIAsTEQmyjF6sBVOYEkDKGqjJFUoqhyBgmaQlWZoKhEUuUCEjADEr5QMwzhqpFUuYIkzZFUmUBVoqhyg7neFy4Ld1csAA==", BinaryEncoding.Base64), Compression.Deflate)), let _t = ((type nullable text) meta [Serialized.Text = true]) in type table [Name = _t, Age = _t, Gender = _t, #"Q1. Family" = _t, #"Q1. Money" = _t, #"Q1. Health" = _t, #"Q1. Friends" = _t]), unpivOthCols = Table.UnpivotOtherColumns(Source, {"Name", "Age", "Gender"}, "Option", "Response"), addWeight = Table.AddColumn(unpivOthCols, "Weight", each (4 - Number.From([Response])) + 1, type number) in addWeightIt's literally just unpivoting and adding a semi-hardcoded [Weight] column.
In a production model, you would further normalise this by removing all the 'Person' fields except the person identifier (in this case [Name]), and using a related dimension table for the person information fields (in this case [Age] and [Gender]). There's an example of this in the working PBIX I've attached below if you want to take this extra step (you should).
Once you have your data in this efficient format, you just need a few small measures:
// Count the number of responses: _noofResponses = COUNTROWS(factTable) // Get the sum of weight value _sumWeight = SUM(factTable[Weight]) // Calculate weighted average: _weightedAvg = DIVIDE([_sumWeight], [_noofResponses], 0)Example Output:
The beauty of setting it up like this is that you can now calculate any of your measures over any dimensions, so you're not stuck with a single hard-coded data format. It's also very fast and HDD size efficient.
Let me know how you get on.
Pete
- 3 years ago
Yes, that's possible. Use this in your [Weight] column instead:
( Number.From(List.Max(previousStepName[Response])) - Number.From([Response]) ) + 1Pete
Hi Keith011 ,
Are you bringing the data into Power BI to be reported on, or is it all staying in Excel?
The reason I ask is that the most efficient way to create your output would be to normalise your raw data then create measures for the values, but this is harder in Excel so would stick to a PQ-only solution if you're not using Power BI.
Pete
hi BA_Pete
I'm doing it in Power BI by using Power Query , is this possible?
when you said normalise my raw data meaning to prepare my raw data like how i intend it to be in excel only then i visualize it in power bi?
- BA_Pete3 years agoSuper User
No, I mean that we'll leave your raw data in Excel, import it into PBI Power Query, transform it there into the most efficient structure for storage and scanning, then write the measures over it that will allow you to visualise it in Power BI how you want it.
Leave it with me, I'll have a look first thing this morning and get back to you.
Pete
- BA_Pete3 years agoSuper User
Here's example code for normalising your raw data in PQ:
let Source = Table.FromRows(Json.Document(Binary.Decompress(Binary.FromText("i45WCkgtKs7PU3BU0lEyMgISbkBsCOIAsTEQmyjF6sBVOYEkDKGqjJFUoqhyBgmaQlWZoKhEUuUCEjADEr5QMwzhqpFUuYIkzZFUmUBVoqhyg7neFy4Ld1csAA==", BinaryEncoding.Base64), Compression.Deflate)), let _t = ((type nullable text) meta [Serialized.Text = true]) in type table [Name = _t, Age = _t, Gender = _t, #"Q1. Family" = _t, #"Q1. Money" = _t, #"Q1. Health" = _t, #"Q1. Friends" = _t]), unpivOthCols = Table.UnpivotOtherColumns(Source, {"Name", "Age", "Gender"}, "Option", "Response"), addWeight = Table.AddColumn(unpivOthCols, "Weight", each (4 - Number.From([Response])) + 1, type number) in addWeightIt's literally just unpivoting and adding a semi-hardcoded [Weight] column.
In a production model, you would further normalise this by removing all the 'Person' fields except the person identifier (in this case [Name]), and using a related dimension table for the person information fields (in this case [Age] and [Gender]). There's an example of this in the working PBIX I've attached below if you want to take this extra step (you should).
Once you have your data in this efficient format, you just need a few small measures:
// Count the number of responses: _noofResponses = COUNTROWS(factTable) // Get the sum of weight value _sumWeight = SUM(factTable[Weight]) // Calculate weighted average: _weightedAvg = DIVIDE([_sumWeight], [_noofResponses], 0)Example Output:
The beauty of setting it up like this is that you can now calculate any of your measures over any dimensions, so you're not stuck with a single hard-coded data format. It's also very fast and HDD size efficient.
Let me know how you get on.
Pete
- Keith0113 years agoHelper III
BA_Pete Worked like a charm!! thanks for this Pete.
I have a question , the added column in PQ in your solution (like below), can the "4" be swaped with a MAX formula? For instance, another survey might have a rating from 1 to 13 rather than just 1 to 4. So if I could tell PQ to always take the MAX number instead of a hard-coded "4"
each (4 - Number.From([Response])) + 1
is that possible?