In my consulting expertise working with clients utilizing Energy BI, many challenges that Energy BI builders face are because of negligence to information sorts. Listed here are some frequent challenges which might be the direct or oblique outcomes of inappropriate information sorts and information sort conversion:
- Getting incorrect outcomes whereas all calculations in your information mannequin are appropriate.
- Poor performing information mannequin.
- Bloated mannequin measurement.
- Difficulties in configuring user-defined aggregations (agg consciousness).
- Difficulties in organising incremental information refresh.
- Getting clean visuals after the primary information refresh in Energy BI service.
On this blogpost, I clarify the frequent pitfalls to forestall future challenges that may be time-consuming to determine and repair.
Background
Earlier than we dive into the subject of this weblog put up, I wish to begin with a little bit of background. Everyone knows that Energy BI will not be solely a reporting device. It’s certainly a knowledge platform supporting numerous elements of enterprise intelligence, information engineering, and information science. There are two languages we should study to have the ability to work with Energy BI: Energy Question (M) and DAX. The aim of the 2 languages is sort of totally different. We use Energy Question for information transformation and information preparation, whereas DAX is used for information evaluation within the Tabular information mannequin. Right here is the purpose, the 2 languages in Energy BI have totally different information sorts.
The most typical Energy BI improvement eventualities begin with connecting to the information supply(s). Energy BI helps a whole lot of knowledge sources. Most information supply connections occur in Energy Question (the information preparation layer in a Energy BI resolution) until we join reside to a semantic layer corresponding to an SSAS occasion or a Energy BI dataset. Many supported information sources have their very own information sorts, and a few don’t. As an illustration, SQL Server has its personal information sorts, however CSV doesn’t. When the information supply has information sorts, the mashup engine tries to determine information sorts to the closest information sort out there in Energy Question. Despite the fact that the supply system has information sorts, the information sorts won’t be suitable with Energy Question information sorts. For the information sources that don’t assist information sorts, the matchup engine tries to detect the information sorts primarily based on the pattern information loaded into the information preview pane within the Energy Question Editor window. However, there is no such thing as a assure that the detected information sorts are appropriate. So, it’s best observe to validate the detected information sorts anyway.
Energy BI makes use of the Tabular mannequin information sorts when it masses the information into the information mannequin. The information sorts within the information mannequin might or is probably not suitable with the information sorts outlined in Energy Question. As an illustration, Energy Question has a Binary information sort, however the Tabular mannequin doesn’t.
The next desk exhibits Energy Question’s datatypes, their representations within the Energy Question Editor’s UI, their mapping information sorts within the information mannequin (DAX), and the inner information sorts within the xVelocity (Tabular mannequin) engine:
Because the above desk exhibits, in Energy Question’s UI, Entire Quantity, Decimal, Fastened Decimal and Share are all in sort quantity within the Energy Question engine. The kind names within the Energy BI UI additionally differ from their equivalents within the xVelocity engine. Allow us to dig deeper.
Information Varieties in Energy Question
As talked about earlier, in Energy Question, now we have just one numeric datatype: quantity whereas within the Energy Question Editor’s UI, within the Rework tab, there’s a Information Kind drop-down button displaying 4 numeric datatypes, as the next picture exhibits:
In Energy Question system language, we specify a numeric information sort as sort quantity or Quantity.Kind. Allow us to have a look at an instance to see what this implies.
The next expression creates a desk with totally different values:
#desk({"Worth"}
, {
{100}
, {65565}
, {-100000}
, {-999.9999}
, {0.001}
, {10000000.0000001}
, {999999999999999999.999999999999999999}
, {#datetimezone(2023,1,1,11,45,54,+12,0)}
, {#datetime(2023,1,1,11,45,54)}
, {#date(2023,1,1)}
, {#time(11,45,54)}
, {true}
, {#length(11,45,54,22)}
, {"It is a textual content"}
})The outcomes are proven within the following picture:
Now we add a brand new column that exhibits the information sort of the values. To take action, use the Worth.Kind([Value]) operate returns the kind of every worth of the Worth column. The outcomes are proven within the following picture:
To see the precise sort, we must click on on every cell (not the values) of the Worth Kind column, as proven within the following picture:
With this methodology, now we have to click on every cell in to see the information sorts of the values that isn’t superb. However there’s at present no operate out there in Energy Question to transform a Kind worth to Textual content. So, to point out every sort’s worth as textual content in a desk, we use a easy trick. There’s a operate in Energy Question returning the desk’s metadata: Desk.Schema(desk as desk). The operate ends in a desk revealing helpful details about the desk used within the operate, together with column Identify, TypeName, Variety, and so forth. We need to present TypeName of the Worth Kind column. So, we solely want to show every worth right into a desk utilizing the Desk.FromValue(worth as any) operate. We then get the values of the Variety column from the output of the Desk.Schema() operate.
To take action, we add a brand new column to get textual values from the Variety column. We named the brand new column Datatypes. The next expression caters to that:
Desk.Schema(
Desk.FromValue([Value])
)[Kind]{0}
The next picture exhibits the outcomes:
Because the outcomes present, all numeric values are of sort quantity and the way in which they’re represented within the Energy Question Editor’s UI doesn’t have an effect on how the Energy Question engine treats these sorts. The information sort representations within the Energy Question UI are one way or the other aligned with the kind aspects in Energy Question. A side is used so as to add particulars to a sort sort. As an illustration, we are able to use aspects to a textual content sort if we need to have a textual content sort that doesn’t settle for null. We will outline the worth’s sorts utilizing sort aspects utilizing Aspect.Kind syntax, corresponding to utilizing In64.Kind for a 64-bit integer quantity or utilizing Share.Kind to point out a quantity in proportion. Nevertheless, to outline the worth’s sort, we use the sort typename syntax corresponding to defining quantity utilizing sort quantity or a textual content utilizing sort textual content. The next desk exhibits the Energy Question sorts and the syntax to make use of to outline them:
Sadly, the Energy Question Language Specification documentation doesn’t embody aspects and there will not be many on-line sources or books that I can reference right here apart from Ben Gribaudo’s weblog who completely defined aspects intimately which I strongly suggest studying.
Whereas Energy Question engine treats the values primarily based on their sorts not their aspects, utilizing aspects is advisable as they have an effect on the information when it’s being loaded into the information mannequin which raises a query: what occurs after we load the information into the information mannequin? which brings us to the following part of this weblog put up.
Information sorts in Energy BI information mannequin
Energy BI makes use of the xVelocity in-memory information processing engine to course of the information. The xVelocity engine makes use of columnstore indexing know-how that compresses the information primarily based on the cardinality of the column, which brings us to a essential level: though the Energy Question engine treats all of the numeric values as the kind quantity, they get compressed in another way relying on their column cardinality after loading the values within the Energy BI mannequin. Due to this fact, setting the right sort side for every column is necessary.
The numeric values are one of the frequent datatypes utilized in Energy BI. Right here is one other instance displaying the variations between the 4 quantity aspects. Run the next expression in a brand new clean question within the Energy Question Editor:
// Decimal Numbers with 6 Decimal Digits
let
Supply = Checklist.Generate(()=> 0.000001, every _ <= 10, every _ + 0.000001 ),
#"Transformed to Desk" = Desk.FromList(Supply, Splitter.SplitByNothing(), null, null, ExtraValues.Error),
#"Renamed Columns" = Desk.RenameColumns(#"Transformed to Desk",{{"Column1", "Supply"}}),
#"Duplicated Supply Column as Decimal" = Desk.DuplicateColumn(#"Renamed Columns", "Supply", "Decimal", Decimal.Kind),
#"Duplicated Supply Column as Fastened Decimal" = Desk.DuplicateColumn(#"Duplicated Supply Column as Decimal", "Supply", "Fastened Decimal", Foreign money.Kind),
#"Duplicated Supply Column as Share" = Desk.DuplicateColumn(#"Duplicated Supply Column as Fastened Decimal", "Supply", "Share", Share.Kind)
in
#"Duplicated Supply Column as Share"The above expressions create 10 million rows of decimal values between 0 and 10. The ensuing desk has 4 columns containing the identical information with totally different aspects. The primary column, Supply, accommodates the values of sort any, which interprets to sort textual content. The remaining three columns are duplicated from the Supply column with totally different sort aspects, as follows:
- Decimal
- Fastened decimal
- Share
The next screenshot exhibits the ensuing pattern information of our expression within the Energy Question Editor:
Now click on Shut & Apply from the House tab of the Energy Question Editor to import the information into the information mannequin. At this level, we have to use a third-party neighborhood device, DAX Studio, which could be downloaded from right here.
After downloading and putting in, DAX Studio registers itself as an Exterior Software within the Energy BI Desktop as the next picture exhibits:
Click on the DAX Studio from the Exterior Instruments tab which mechanically connects it to the present Energy BI Desktop mannequin, and observe these steps:
- Click on the Superior tab
- Click on the View Metrics button
- Click on Columns from the VertiPaq Analyzer part
- Have a look at the Cardinality, Col Dimension, and % Desk columns
The next picture exhibits the previous steps:
The outcomes present that the Decimal column and Share consumed probably the most vital a part of the desk’s quantity. Their cardinality can also be a lot greater than the Fastened Decimal column. So right here it’s now extra apparent that utilizing the Fastened Decimal datatype (side) for numeric values may also help with information compression, decreasing the information mannequin measurement and growing the efficiency. Due to this fact, it’s sensible to all the time use Fastened Decimal for decimal values. Because the Fastened Decimal values translate to the Foreign money datatype in DAX, we should change the columns’ format if Foreign money is unsuitable. Because the identify suggests, Fastened Decimal has mounted 4 decimal factors. Due to this fact, if the unique worth has extra decimal digits after conversion to the Fastened Decimal, the digits after the fourth decimal level shall be truncated.
That’s the reason the Cardinality column within the VertiPaq Analyzer in DAX Studio exhibits a lot decrease cardinality for the Fastened Decimal column (the column values solely hold as much as 4 decimal factors, no more).
Obtain the pattern file from right here.
So, the message is right here to all the time use the datatype that is sensible to the enterprise and is environment friendly within the information mannequin. Utilizing the VertiPaq Analyzer in DAX Studio is nice for understanding the varied elements of the information mannequin, together with the column datatypes. As a knowledge modeler, it’s important to grasp how the Energy Question sorts and aspects translate to DAX datatypes. As we noticed on this weblog put up, information sort conversion can have an effect on the information mannequin’s compression fee and efficiency.
