Unnest Transform

NOTE: Transforms are a part of the underlying language that is not directly accessible to users. This content is maintained for reference purposes only.

Unpacks nested data from an Array or Object column to create new rows or columns based on the keys in the source data.

This transform works differently on columns of Array or Object type.

The unnest transform must include keys that you specify as part of the transform step. To unnest a column of array data that contains no keys, use the flatten transform. See Flatten Transform.

This transform might be automatically applied as one of the first steps of your recipe. See Initial Parsing Steps.

Basic Usage

unnest col: myObj keys:'sourceA','sourceB' pluck:true markLineage:true

Output:

  • Extracts from the myObj column the corresponding values for the keys sourceA and sourceB into two new columns.
  • Since markLineage is true, these new column names are prepended with the source name: sourceA_column1 and sourceB_column2.
  • Any non-missing values from the source columns are added to the corresponding new columns and are removed from the source column, since pluck is true.

Parameters

unnest col:column_ref keys:'key1','key2' [pluck:true|false] [markLineage:true|false]

TokenRequired?Data TypeDescription
unnestYtransformName of the transform
colYstringSource column name
keysYstringComma-separated list of quoted key names. See below for examples.
pluckNbooleanIf true, any values unnested from the source are also removed from the source. Default is false.
markLineageNbooleanIf true, the names of new columns are prepended with the name of the source column.

For more information on syntax standards, see Language Documentation Syntax Notes.

col

Identifies the column to which to apply the transform. You can specify only one column.

Usage Notes:

Required?Data Type
YesString (column name)

keys

Comma-separated list of keys to use to extract data from the specified source column.

  • Key values must be quoted. (e.g 'key1','key2'). Any quoted value is considered the path to a single key.
  • Key values are case-sensitive.
  • Each key must be listed. A range of keys cannot be specified.

NOTE: Keys that contain non-alphanumeric values, such as spaces, must be enclosed in square brackets and quotes. Values with underscores do not require this bracketing.

The comma-separated list of keys determines the columns to generate from the source data. If you specify three values for keys, the three new columns contain the corresponding values from the source column.

This parameter has different syntax to use for single-level and multi-level nested data. There are also variations in syntax between Object and Array data type.

Usage Notes:

Required?Data Type
Yes

Comma-separated String values.

Syntax examples are provided below.

Keys for Object data - single-level

NOTE: Key names are case-sensitive.

For a single, top-level key in an Object field, you can specify the key as a simple quoted string:

unnest col:myCol keys: 'myObjKey'

The above looks for the key myObjKey among the top-level keys in the Object and returns the corresponding value for the new column. You can also bracket this key in square brackets:

unnest col:myCol keys: '[myObjKey]'

To specify multiple first-level keys, use the following:

unnest col:myCol keys:'myObjKey','my2ndObjKey'

The above generates two new columns ( myObjKey and my2ndObjKey) containing the corresponding values for the keys.

Keys for Object data - multi-level

You can also reference keys that are below the first level in the Object.

Example data:

{ "Key1" :
  { "Key1A" :

  }
}
{ "Key2" :
  { "Key2A" :

  }
}
{ "Key3" :
  { "Key3A" :

  }
}

To acquire the data for the Key1A key, use the following:

unnest col: myCol keys: 'Key1[Key1A]'

In the new column, the displayed value is the following:

To unnest a third-layer value, use a transform similar to the following:

unnest col: myCol keys: 'Key2[Key2A][Key2A1]'

In the new column, this transform generates a value of Value2.

Keys for Array data - single level

You can reference array elements using zero-based indexes or key names.

NOTE: All references to Array keys must be bracketed. Array keys can be referenced by index number only.

Example array data:

["red","orange","yellow","green","blue","indigo","violet"]

unnest col: myCol keys:'[1]'

The above transform retrieves the value orange from the array.

unnest col: myCol keys:'[1]','[3]'

Returned values: orange and green.

Keys for Array data - multi-level

The following example nested Array data matches the structure of the Object data in the previous example:

[ [ "Item1", ["Item1A", ["Item1A1","Value1"] ] ], [ "Item2", ["Item2A",  ["Item2A1","Value2"] ] ], [ "Item3", ["Item3A",["Item3A1","Value3"] ] ] ] 

To unnest the value for Items2A:

unnest col:myCol keys:'[1][0]'

The value inserted into the new column is the following:

["Item2A1","Value2"]

To unnest from the third level:

unnest col:myCol keys:'[2][0][0]'

The inserted value is Item3A.

pluck

Indicates whether any values added from source to output columns should be removed from the source.

  • Set to true to remove values from source after they have been added to output columns.
  • (Default) Set to false to leave source columns untouched.

Usage Notes:

Required?Data Type
NoBoolean

markLineage

When set to true, the names of new columns are prepended with the name of the source column. Example:

Source ColumnOutput Column
mySourceColumnmySourceColumn_column1

Nested key references are appended to the column name:

Source ColumnKey ValueOutput Column
mySourceColumnkeys: '[Key1][Key2]'mySourceColumn_Key1_Key2

NOTE: If your unnest transform does not change the number of rows, you can still access source row number information in the data grid, assuming it was still available when the transform was executed.

Usage Notes:

Required?Data Type
NoBoolean

Examples

Example - Unnest an Object

You have the following dataset. The Sizes column contains Object data on available sizes.

Source:

ProdIdProdNameSizes
1001Hat{'Small':'N','Medium':'Y','Large':'Y','Extra-Large':'Y'}
1002Shirt{'Small':'N','Medium':'Y','Large':'Y','Extra-Large':'N'}
1003Pants{'Small':'Y','Medium':'Y','Large':'Y','Extra-Large':'N'}

Transform:

NOTE: Depending on the format of your source data, you might need to perform some replacements in the Sizes column in order to make it inferred as proper Object type values. The final format should look like the above.

If it is not inferred already, set the type of the Sizes column to Object:

settype col: Sizes type: 'Object'

Unnest the data into separate columns. The following prepends Sizes_ to the newly generated column name.

unnest col:Sizes keys:'Small','Medium','Large','Extra-Large' markLineage:true

You might find it useful to add pluck:true to the above transform. When added, values that are un-nested are removed from the source, leaving only the values that weren't processed:

unnest col:Sizes keys:'Small','Medium','Large','Extra-Large' markLineage:true pluck:true

If all values have been processed, the Sizes column now contains a set of maps missing data. You can use the following to determine if the length of the remaining data is longer than two characters. This transform is a good one to just preview:

derive type:single value:(LEN(Sizes) > 2) as:'len_Sizes'

You can drop the source column:

drop col:Sizes

Results:

When you are finished, the dataset should look like the following:

ProdIdProdNameSizes_SmallSizes_MediumSizes_LargeSizes_Extra-Large
1001HatNYYY
1002ShirtNYYN
1003PantsYYYN

Example - Unnest an array

The following example demonstrates differences between the unnest and the flatten transform, including how you use unnest to flatten array data based on specified keys.

Source:

You have the following data on student test scores. Scores on individual scores are stored in the Scores array, and you need to be able to track each test on a uniquely identifiable row. This example has two goals:

  1. One row for each student test
  2. Unique identifier for each student-score combination
LastNameFirstNameScores
AdamsAllen[81,87,83,79]
BurnsBonnie[98,94,92,85]
CannonCharles[88,81,85,78]

Transform:

When the data is imported from CSV format, you must add a header transform and remove the quotes from the Scores column:

header

replace col:Scores with:'' on:`"` global:true

Validate test date: To begin, you might want to check to see if you have the proper number of test scores for each student. You can use the following transform to calculate the difference between the expected number of elements in the Scores array (4) and the actual number:

derive type:single value: (4 - ARRAYLEN(Scores)) as: 'numMissingTests'

When the transform is previewed, you can see in the sample dataset that all tests are included. You might or might not want to include this column in the final dataset, as you might identify missing tests when the recipe is run at scale.

Unique row identifier: The Scores array must be broken out into individual rows for each test. However, there is no unique identifier for the row to track individual tests. In theory, you could use the combination of LastName-FirstName-Scores values to do so, but if a student recorded the same score twice, your dataset has duplicate rows. In the following transform, you create a parallel array called Tests, which contains an index array for the number of values in the Scores column. Index values start at 0:

derive type:single value:RANGE(0,ARRAYLEN(Scores)) as:'Tests'

Also, we will want to create an identifier for the source row using the SOURCEROWNUMBER function:

derive type:single value:SOURCEROWNUMBER() as:'orderIndex'

One row for each student test: Your data should look like the following:

LastNameFirstNameScoresTestsorderIndex
AdamsAllen[81,87,83,79][0,1,2,3]2
BurnsBonnie[98,94,92,85][0,1,2,3]3
CannonCharles[88,81,85,78][0,1,2,3]4

Now, you want to bring together the Tests and Scores arrays into a single nested array using the ARRAYZIP function:

derive type:single value:ARRAYZIP([Tests,Scores])

Your dataset has been changed:

LastNameFirstNameScoresTestsorderIndexcolumn1
AdamsAllen[81,87,83,79][0,1,2,3]2[[0,81],[1,87],[2,83],[3,79]]
AdamsBonnie[98,94,92,85][0,1,2,3]3[[0,98],[1,94],[2,92],[3,85]]
CannonCharles[88,81,85,78][0,1,2,3]4[[0,88],[1,81],[2,85],[3,78]]

With the flatten transform, you can unpack the nested array:

flatten col: column1

Each test-score combination is now broken out into a separate row. The nested Test-Score combinations must be broken out into separate columns using unnest:

unnest col:column1 keys:'[0]','[1]'

After you drop column1, which is no longer needed you should rename the two generated columns:

rename mapping:[column_0,'TestNum']

rename mapping:[column_1,'TestScore']

Unique row identifier: You can do one more step to create unique test identifiers, which identify the specific test for each student. The following uses the original row identifier OrderIndex as an identifier for the student and the TestNumber value to create the TestId column value:

derive type:single value: (orderIndex * 10) + TestNum as: 'TestId'

The above are integer values. To make your identifiers look prettier, you might add the following:

merge col:'TestId00','TestId'


Extending:
You might want to generate some summary statistical information on this dataset. For example, you might be interested in calculating each student's average test score. This step requires figuring out how to properly group the test values. In this case, you cannot group by the LastName value, and when executed at scale, there might be collisions between first names when this recipe is run at scale. So, you might need to create a kind of primary key using the following:

merge col:'LastName','FirstName' with:'-' as:'studentId'

You can now use this as a grouping parameter for your calculation:

derive type:single value:AVERAGE(TestScore) group:studentId as:'avg_TestScore'

Results:

After you drop unnecessary columns and move your columns around, the dataset should look like the following:

TestIdLastNameFirstNameTestNumTestScorestudentIdavg_TestScore
TestId0021AdamsAllen081Adams-Allen82.5
TestId0022AdamsAllen187Adams-Allen82.5
TestId0023AdamsAllen283Adams-Allen82.5
TestId0024AdamsAllen379Adams-Allen82.5
TestId0031AdamsBonnie098Adams-Bonnie92.25
TestId0032AdamsBonnie194Adams-Bonnie92.25
TestId0033AdamsBonnie292Adams-Bonnie92.25
TestId0034AdamsBonnie385Adams-Bonnie92.25
TestId0041CannonChris088Cannon-Chris83
TestId0042CannonChris181Cannon-Chris83
TestId0043CannonChris285Cannon-Chris83
TestId0044CannonChris378Cannon-Chris83

Example - extracting key values from car data and then unnesting into separate columns

This example shows how you can unpack data nested in an Object into separate columns using the following transforms:

Source:

You have the following information on used cars. The VIN column contains vehicle identifiers, and the Properties column contains key-value pairs describing characteristics of each vehicle. You want to unpack this data into separate columns.

VINProperties
XX3 JT4522year=2004,make=Subaru,model=Impreza,color=green,mileage=125422,cost=3199
HT4 UJ9122year=2006,make=VW,model=Passat,color=silver,mileage=102941,cost=4599
KC2 WZ9231year=2009,make=GMC,model=Yukon,color=black,mileage=68213,cost=12899
LL8 UH4921year=2011,make=BMW,model=328i,color=brown,mileage=57212,cost=16999

Transform:

Add the following transform, which identifies all of the key values in the column as beginning with alphabetical characters.

  • The valueafter string identifies where the corresponding value begins after the key.
  • The delimiter string indicates the end of each key-value pair.

extractkv col:Properties key:`{alpha}+` valueafter:`=` delimiter:`,`

Now that the Object of values has been created, you can use the unnest transform to unpack this mapped data. In the following, each key is specified, which results in separate columns headed by the named key:

unnest col:extractkv_Properties keys:'year','make','model','color','mileage','cost'

Results:

When you drop the unnecessary Properties columns, the dataset now looks like the following:

VINyearmakemodelcolormileagecost
XX3 JT45222004SubaruImprezagreen1254223199
HT4 UJ91222006VWPassatsilver1029414599
KC2 WZ92312009GMCYukonblack6821312899
LL8 UH49212011BMW328ibrown5721216999

Was this page helpful? Let us know how we did:

Send feedback about...

Google Cloud Dataprep Documentation
Need help? Visit our support page.