This page refers to the
type
parameter that is part of a measure.
type
can also be used as part of a dimension or filter, described on the Dimension, filter, and parameter types documentation page.
type
can also be used as part of a dimension group, described on thedimension_group
parameter documentation page.
Usage
view: view_name { measure: field_name { type: measure_field_type } }
Hierarchy
type |
Possible Field Types
Measure
Accepts
A measure type
|
This page includes details about the various types that can be assigned to a measure. A measure can only have one type, and it defaults to string
if no type is specified.
Some measure types have supporting parameters, which are described within the appropriate section.
Measure type categories
Each measure type falls into one of the following categories. These categories determine whether the measure type performs aggregations, the type of fields that the measure type can reference, and whether you can filter the measure type using the filters
parameter:
- Aggregate measures: Aggregate measure types perform aggregations, such as
sum
andaverage
. Aggregate measures can reference only dimensions, not other measures. This is the only measure type that works with thefilters
parameter. - Non-aggregate measures: Non-aggregate measures are, as the name suggests, measure types that do not perform aggregations, such as
number
andyesno
. These measure types perform simple transformations, and since they do not perform aggregations, can reference only aggregate measures or previously-aggregated dimensions. You cannot use thefilters
parameter with these measure types. - Post-SQL measures: Post-SQL measures are special measure types that perform specific calculations after Looker has generated query SQL. They can reference only numeric measures or numeric dimensions. You cannot use the
filters
parameter with these measure types.
List of type definitions
Type | Category | Description |
---|---|---|
average |
Aggregate | Generates an average (mean) of values within a column |
average_distinct |
Aggregate | Properly generates an average (mean) of values when using denormalized data. See the average_distinct section for a complete description. |
count |
Aggregate | Generates a count of rows |
count_distinct |
Aggregate | Generates a count of unique values within a column |
date |
Non-aggregate | For measures that contain dates |
list |
Aggregate | Generates a list of the unique values within a column |
max |
Aggregate | Generates the maximum value within a column |
median |
Aggregate | Generates the median (midpoint value) of values within a column |
median_distinct |
Aggregate | Properly generates a median (midpoint value) of the values when a join causes a fanout. See the median_distinct section for a complete description. |
min |
Aggregate | Generates the minimum value within a column |
number |
Non-aggregate | For measures that contain numbers |
percent_of_previous |
Post-SQL | Generates the percent difference between displayed rows |
percent_of_total |
Post-SQL | Generates the percent of total for each displayed row |
percentile |
Aggregate | Generates the value at the specified percentile within a column |
percentile_distinct |
Aggregate | Properly generates the value at the specified percentile when a join causes a fanout. See the percentile_distinct section for a complete description. |
running_total |
Post-SQL | Generates the running total for each displayed row |
string |
Non-aggregate | For measures that contain letters or special characters (as with MySQL's GROUP_CONCAT function) |
sum |
Aggregate | Generates a sum of values within a column |
sum_distinct |
Aggregate | Properly generates a sum of values when using denormalized data.See the sum_distinct section for a complete description. |
yesno |
Non-aggregate | For fields that will show if something is true or false |
int |
Non-aggregate |
REMOVED 5.4
Replaced by type: number |
average
type: average
averages the values in a given field. It is similar to SQL's AVG
function. However, unlike with raw SQL, Looker will properly calculate averages even if your query's joins contain fanouts.
The sql
parameter for type: average
measures can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
type: average
fields can be formatted by using the value_format
or value_format_name
parameters.
For example, the following LookML creates a field called avg_order
by averaging the sales_price
dimension, then displays it in a money format ($1,234.56):
measure: avg_order {
type: average
sql: ${sales_price} ;;
value_format_name: usd
}
average_distinct
type: average_distinct
is for use with denormalized datasets. It averages the nonrepeated values in a given field, based on the unique values defined by the sql_distinct_key
parameter.
This is an advanced concept which may be more clearly explained with an example. Consider a denormalized table like this:
Order Item ID | Order ID | Order Shipping |
---|---|---|
1 | 1 | 10.00 |
2 | 1 | 10.00 |
3 | 2 | 20.00 |
4 | 2 | 20.00 |
5 | 2 | 20.00 |
In this situation you can see that there are multiple rows for each order. Consequently, if you added a simple type: average
measure for the order_shipping
column, you would get a value of 16.00, even though the actual average is 15.00.
# Will NOT calculate the correct average
measure: avg_shipping {
type: average
sql: ${order_shipping} ;;
}
To get an accurate result, you can define for Looker how it should identify each unique entity (in this case, each unique order) by using the sql_distinct_key
parameter. This will calculate the correct 15.00 amount:
# Will calculate the correct average
measure: avg_shipping {
type: average_distinct
sql_distinct_key: ${order_id} ;;
sql: ${order_shipping} ;;
}
Every unique value of sql_distinct_key
must have just one corresponding value in sql
. In other words, the preceding example works because every row with an order_id
of 1 has the same order_shipping
of 10.00, every row with an order_id
of 2 has the same order_shipping
of 20.00, and so on.
type: average_distinct
fields can be formatted by using the value_format
or value_format_name
parameters.
count
type: count
performs a table count, similar to SQL's COUNT
function. However, unlike with raw SQL, Looker will properly calculate counts even if your query's joins contain fanouts.
type: count
measures perform table counts that are based on the table's primary key, so type: count
measures don't support the sql
parameter.
If you want to perform a table count on a field other than the table's primary key, use a type: count_distinct
measure. Or, if you don't want to use count_distinct
, you can use a measure of type: number
(see the Community post How to count a non-primary key for more information).
For example, the following LookML creates a field number_of_products
:
view: products {
measure: number_of_products {
type: count
drill_fields: [product_details*] # optional
}
}
It is very common to provide a drill_fields
(for fields) parameter when defining a type: count
measure, so that users can see the individual records that make up a count when they click on it.
When you use a measure of
type: count
in an Explore, the visualization labels the resulting values with the view name rather than the word "Count." To avoid confusion, we recommend pluralizing your view name, selecting Show Full Field Name under Series in the visualization settings, or using aview_label
with a pluralized version of your view name.
You can add a filter to a measure of type: count
using the filters
parameter.
count_distinct
type: count_distinct
calculates the number of distinct values in a given field. It makes use of SQL's COUNT DISTINCT
function.
The sql
parameter for type: count_distinct
measures can take any valid SQL expression that results in a table column, LookML dimension, or combination of LookML dimensions.
For example, the following LookML creates a field number_of_unique_customers
, which counts the number of unique customer IDs:
measure: number_of_unique_customers {
type: count_distinct
sql: ${customer_id} ;;
}
You can add a filter to a measure of type: count_distinct
using the filters
parameter.
date
type: date
is used with fields that contain dates.
The sql
parameter for type: date
measures can take any valid SQL expression that results in a date. In practice, this type is rarely used, because most SQL aggregate functions do not return dates. One common exception is a MIN
or MAX
of a date dimension.
Creating a max or min date measure with type: date
If you want to create a measure of a maximum or minimum date, you might initially think it would work to use a measure of type: max
or of type: min
. However, these measure types are compatible only with numerical fields. Instead, you can capture a maximum or minimum date by defining a measure of type: date
and wrapping the date field that is referenced in the sql
parameter in a MIN()
or MAX()
function.
Suppose you have a dimension group of type: time
, called updated
:
dimension_group: updated {
type: time
timeframes: [time, date, week, month, raw]
sql: ${TABLE}.updated_at ;;
}
You can create a measure of type: date
to capture the maximum date of this dimension group as follows:
measure: last_updated_date {
type: date
sql: MAX(${updated_raw}) ;;
convert_tz: no
}
In this example, instead of using a measure of type: max
to create the last_updated_date
measure, the MAX()
function is applied in the sql
parameter. The last_updated_date
measure also has the convert_tz
parameter set to no
to prevent double time zone conversion in the measure, since time zone conversion has already occurred in the definition of the dimension group updated
. For more information, see the documentation on the convert_tz
parameter.
In the example LookML for the last_updated_date
measure, type: date
could be omitted, and the value would be treated as a string, because string
is the default value for type
. However, you will get better filtering capability for users if you use type: date
.
You may also notice that the last_updated_date
measure definition references the ${updated_raw}
timeframe instead of the ${updated_date}
timeframe. Because the value returned from ${updated_date}
is a string, it is necessary to use ${updated_raw}
to reference the actual date value instead.
You can also use the datatype
parameter with type: date
to enhance query performance by specifying the type of date data your database table uses.
Creating a max or min measure for a datetime column
Computing the maximum for a type: datetime
column is a little different. In this case, you want to create a measure without declaring the type, like this:
measure: last_updated_datetime {
sql: MAX(${TABLE}.datetime_string_field) ;;
}
list
type: list
creates a list of the distinct values in a given field. It is similar to MySQL's GROUP_CONCAT
function.
You do not need to include a sql
parameter for type: list
measures. Instead, you'll use the list_field
parameter to specify the dimension from which you want to create lists.
The usage is:
view: view_name { measure: field_name { type: list list_field: my_field_name } }
For example, the following LookML creates a measure name_list
based on the name
dimension:
measure: name_list {
type: list
list_field: name
}
Note the following for list
:
- The
list
measure type does not support filtering. You cannot use thefilters
parameter on atype: list
measure. - The
list
measure type cannot be referenced using the substitution operator ($). You cannot use the${}
syntax to refer to atype: list
measure.
Supported database dialects for list
For Looker to support type: list
in your Looker project, your database dialect must also support it. The following table shows which dialects support type: list
in the latest release of Looker:
Dialect | Supported? |
---|---|
Actian Avalanche | Yes |
Amazon Athena | Yes |
Amazon Aurora MySQL | Yes |
Amazon Redshift | Yes |
Apache Druid | No |
Apache Druid 0.13+ | No |
Apache Druid 0.18+ | No |
Apache Hive 2.3+ | Yes |
Apache Hive 3.1.2+ | Yes |
Apache Spark 3+ | Yes |
ClickHouse | Yes |
Cloudera Impala 3.1+ | No |
Cloudera Impala 3.1+ with Native Driver | No |
Cloudera Impala with Native Driver | No |
DataVirtuality | No |
Databricks | Yes |
Denodo 7 | No |
Denodo 8 | No |
Dremio | No |
Dremio 11+ | No |
Exasol | Yes |
Firebolt | Yes |
Google BigQuery Legacy SQL | Yes |
Google BigQuery Standard SQL | Yes |
Google Cloud PostgreSQL | Yes |
Google Cloud SQL | Yes |
Google Spanner | No |
Greenplum | Yes |
HyperSQL | Yes |
IBM Netezza | No |
MariaDB | Yes |
Microsoft Azure PostgreSQL | Yes |
Microsoft Azure SQL Database | No |
Microsoft Azure Synapse Analytics | No |
Microsoft SQL Server 2008+ | No |
Microsoft SQL Server 2012+ | No |
Microsoft SQL Server 2016 | No |
Microsoft SQL Server 2017+ | No |
MongoBI | No |
MySQL | Yes |
MySQL 8.0.12+ | Yes |
Oracle | Yes |
Oracle ADWC | Yes |
PostgreSQL 9.5+ | Yes |
PostgreSQL pre-9.5 | Yes |
PrestoDB | Yes |
PrestoSQL | Yes |
SAP HANA 2+ | No |
SingleStore | Yes |
SingleStore 7+ | Yes |
Snowflake | Yes |
Teradata | No |
Trino | Yes |
Vector | Yes |
Vertica | No |
max
type: max
finds the largest value in a given field. It makes use of SQL's MAX
function.
The sql
parameter for measures of type: max
can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
Since measures of type: max
are compatible only with numerical fields, you cannot use a measure of type: max
to find a maximum date. Instead, you can use the MAX()
function in the sql
parameter of a measure of type: date
to capture a maximum date, as shown previously in the examples in the date
section.
type: max
fields can be formatted by using the value_format
or value_format_name
parameters.
For example, the following LookML creates a field called largest_order
by looking at the sales_price
dimension, then displays it in a money format ($1,234.56):
measure: largest_order {
type: max
sql: ${sales_price} ;;
value_format_name: usd
}
You cannot currently use type: max
measures for strings or dates, but you can manually add the MAX
function to create such a field, like this:
measure: latest_name_in_alphabet {
type: string
sql: MAX(${name}) ;;
}
median
type: median
returns the midpoint value for the values in a given field. This is especially useful when the data has a few very large or small outlier values that would skew a simple average (mean) of the data.
Consider a table like this:
Order Item ID | Cost | Midpoint? |
---|---|---|
2 | 10.00 | |
4 | 10.00 | |
3 | 20.00 | Midpoint value |
1 | 80.00 | |
5 | 90.00 |
For easy viewing, the table is sorted by cost but that does not affect the result. While the average
type would return 42 (adding all the values and dividing by 5), the median
type would return the midpoint value: 20.00.
If there is an even number of values, then the median value is calculated by taking the mean of the two values closest to the midpoint. Consider a table like this with an even number of rows:
Order Item ID | Cost | Midpoint? |
---|---|---|
2 | 10 | |
3 | 20 | Closest before midpoint |
1 | 80 | Closest after midpoint |
4 | 90 |
The median, the middle value, is (20 + 80)/2 = 50
.
The median is also equal to the value at the 50th percentile.
The sql
parameter for type: median
measures can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
type: median
fields can be formatted by using the value_format
or value_format_name
parameters.
Example
For example, the following LookML creates a field called median_order
by averaging the sales_price
dimension, then displays it in a money format ($1,234.56):
measure: median_order {
type: median
sql: ${sales_price} ;;
value_format_name: usd
}
Things to consider for median
If you're using median
for a field that is involved in a fanout, Looker will attempt to use median_distinct
instead. However, medium_distinct
is supported only for certain dialects. If median_distinct
is not available for your dialect, Looker returns an error. Since the median
can be considered the 50th percentile, the error states that the dialect does not support distinct percentiles.
Supported database dialects for median
For Looker to support the median
type in your Looker project, your database dialect must also support it. The following table shows which dialects support the median
type in the latest release of Looker:
Dialect | Supported? |
---|---|
Actian Avalanche | No |
Amazon Athena | Yes |
Amazon Aurora MySQL | Yes |
Amazon Redshift | Yes |
Apache Druid | No |
Apache Druid 0.13+ | No |
Apache Druid 0.18+ | No |
Apache Hive 2.3+ | Yes |
Apache Hive 3.1.2+ | Yes |
Apache Spark 3+ | Yes |
ClickHouse | Yes |
Cloudera Impala 3.1+ | No |
Cloudera Impala 3.1+ with Native Driver | No |
Cloudera Impala with Native Driver | No |
DataVirtuality | No |
Databricks | Yes |
Denodo 7 | No |
Denodo 8 | No |
Dremio | No |
Dremio 11+ | No |
Exasol | Yes |
Firebolt | No |
Google BigQuery Legacy SQL | Yes |
Google BigQuery Standard SQL | Yes |
Google Cloud PostgreSQL | Yes |
Google Cloud SQL | Yes |
Google Spanner | No |
Greenplum | Yes |
HyperSQL | No |
IBM Netezza | No |
MariaDB | Yes |
Microsoft Azure PostgreSQL | Yes |
Microsoft Azure SQL Database | No |
Microsoft Azure Synapse Analytics | No |
Microsoft SQL Server 2008+ | No |
Microsoft SQL Server 2012+ | No |
Microsoft SQL Server 2016 | No |
Microsoft SQL Server 2017+ | No |
MongoBI | No |
MySQL | Yes |
MySQL 8.0.12+ | Yes |
Oracle | Yes |
Oracle ADWC | Yes |
PostgreSQL 9.5+ | Yes |
PostgreSQL pre-9.5 | Yes |
PrestoDB | Yes |
PrestoSQL | Yes |
SAP HANA 2+ | No |
SingleStore | No |
SingleStore 7+ | Yes |
Snowflake | Yes |
Teradata | No |
Trino | Yes |
Vector | No |
Vertica | Yes |
When there is a fanout involved in a query, Looker tries to convert the median
into median_distinct
. This is only successful in dialects that support median_distinct
.
median_distinct
Use type: median_distinct
when your join involves a fanout. It averages the nonrepeated values in a given field, based on the unique values defined by the sql_distinct_key
parameter. If the measure does not have a sql_distinct_key
parameter, then Looker tries to use the primary_key
field.
Consider the result of a query joining the Order Item and Order tables:
Order Item ID | Order ID | Order Shipping |
---|---|---|
1 | 1 | 10 |
2 | 1 | 10 |
3 | 2 | 20 |
4 | 3 | 50 |
5 | 3 | 50 |
6 | 3 | 50 |
In this situation you can see that there are multiple rows for each order. This query involved a fanout because each order maps to several order items. The median_distinct
takes this into consideration and finds the median between the distinct values 10, 20, and 50 so you would get a value of 20.
To get an accurate result, you can define for Looker how it should identify each unique entity (in this case, each unique order) by using the sql_distinct_key
parameter. This will calculate the correct amount:
measure: median_shipping {
type: median_distinct
sql_distinct_key: ${order_id} ;;
sql: ${order_shipping} ;;
}
Every unique value of sql_distinct_key
must have just one corresponding value in the measure's sql
parameter. In other words, the preceding example works because every row with an order_id
of 1 has the same order_shipping
of 10, every row with an order_id
of 2 has the same order_shipping
of 20, and so on.
type: median_distinct
fields can be formatted by using the value_format
or value_format_name
parameters.
Things to consider for median_distinct
The medium_distinct
measure type is supported only for certain dialects. If median_distinct
is not available for the dialect, Looker returns an error. Since the median
can be considered the 50th percentile, the error states that the dialect does not support distinct percentiles.
Supported database dialects for median_distinct
For Looker to support the median_distinct
type in your Looker project, your database dialect must also support it. The following table shows which dialects support the median_distinct
type in the latest release of Looker:
Dialect | Supported? |
---|---|
Actian Avalanche | No |
Amazon Athena | No |
Amazon Aurora MySQL | Yes |
Amazon Redshift | No |
Apache Druid | No |
Apache Druid 0.13+ | No |
Apache Druid 0.18+ | No |
Apache Hive 2.3+ | No |
Apache Hive 3.1.2+ | No |
Apache Spark 3+ | No |
ClickHouse | No |
Cloudera Impala 3.1+ | No |
Cloudera Impala 3.1+ with Native Driver | No |
Cloudera Impala with Native Driver | No |
DataVirtuality | No |
Databricks | No |
Denodo 7 | No |
Denodo 8 | No |
Dremio | No |
Dremio 11+ | No |
Exasol | No |
Firebolt | No |
Google BigQuery Legacy SQL | Yes |
Google BigQuery Standard SQL | Yes |
Google Cloud PostgreSQL | Yes |
Google Cloud SQL | Yes |
Google Spanner | No |
Greenplum | Yes |
HyperSQL | No |
IBM Netezza | No |
MariaDB | Yes |
Microsoft Azure PostgreSQL | Yes |
Microsoft Azure SQL Database | No |
Microsoft Azure Synapse Analytics | No |
Microsoft SQL Server 2008+ | No |
Microsoft SQL Server 2012+ | No |
Microsoft SQL Server 2016 | No |
Microsoft SQL Server 2017+ | No |
MongoBI | No |
MySQL | Yes |
MySQL 8.0.12+ | Yes |
Oracle | No |
Oracle ADWC | No |
PostgreSQL 9.5+ | Yes |
PostgreSQL pre-9.5 | Yes |
PrestoDB | No |
PrestoSQL | No |
SAP HANA 2+ | No |
SingleStore | No |
SingleStore 7+ | No |
Snowflake | No |
Teradata | No |
Trino | No |
Vector | No |
Vertica | No |
min
type: min
finds the smallest value in a given field. It makes use of SQL's MIN
function.
The sql
parameter for measures of type: min
can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
Since measures of type: min
are compatible only with numerical fields, you cannot use a measure of type: min
to find a minimum date. Instead, you can use the MIN()
function in the sql
parameter of a measure of type: date
to capture a minimum, just as you can use the MAX()
function with a measure of type: date
to capture a maximum date. This is shown previously on this page in the date
section, which includes examples of using the MAX()
function in the sql
parameter to find a maximum date.
type: min
fields can be formatted by using the value_format
or value_format_name
parameters.
For example, the following LookML creates a field called smallest_order
by looking at the sales_price
dimension, then displays it in a money format ($1,234.56):
measure: smallest_order {
type: min
sql: ${sales_price} ;;
value_format_name: usd
}
You cannot currently use type: min
measures for strings or dates, but you can manually add the MIN
function to create such a field, like this:
measure: earliest_name_in_alphabet {
type: string
sql: MIN(${name}) ;;
}
number
type: number
is used with numbers or integers. A measure of type: number
does not perform any aggregation, and is meant to perform simple transformations on other measures. If you are defining a measure based on another measure, the new measure must be of type: number
to avoid nested-aggregation errors.
The sql
parameter for type: number
measures can take any valid SQL expression that results in a number or an integer.
type: number
fields can be formatted by using the value_format
or value_format_name
parameters.
For example, the following LookML creates a measure called total_gross_margin_percentage
based on the total_sale_price
and total_gross_margin
aggregate measures, then displays it in a percentage format with two decimals (12.34%):
measure: total_sale_price {
type: sum
value_format_name: usd
sql: ${sale_price} ;;
}
measure: total_gross_margin {
type: sum
value_format_name: usd
sql: ${gross_margin} ;;
}
measure: total_gross_margin_percentage {
type: number
value_format_name: percent_2
sql: ${total_gross_margin}/ NULLIF(${total_sale_price},0) ;;
}
This example also uses the NULLIF()
SQL function to remove the possibility of division-by-zero errors.
Things to consider for type: number
There are several important things to keep in mind when using type: number
measures:
- A measure of
type: number
can perform arithmetic only on other measures, not on other dimensions. - Looker's symmetric aggregates will not protect aggregate functions in the SQL of a measure
type: number
when computed across a join. - The
filters
parameter cannot be used withtype: number
measures, but thefilters
documentation explains a workaround. type: number
measures will not provide suggestions to users.
percent_of_previous
type: percent_of_previous
calculates the percent difference between a cell and the previous cell in its column.
The sql
parameter for type: percent_of_previous
measures must reference another numeric measure.
type: percent_of_previous
fields can be formatted by using the value_format
or value_format_name
parameters. However, the percentage formats of the value_format_name
parameter do not work with type: percent_of_previous
measures. These percentage formats multiply values by 100, which skews results of a percent of previous calculation.
This example LookML creates a count_growth
measure that is based on the count
measure:
measure: count_growth {
type: percent_of_previous
sql: ${count} ;;
}
Note that percent_of_previous
values depend on sort order. If you change the sort, you must rerun the query to recalculate the percent_of_previous
values. In cases where a query is pivoted, percent_of_previous
runs across the row instead of down the column. You cannot currently change this behavior.
Additionally, percent_of_previous
measures are calculated after data is returned from your database. This means that you should not reference a percent_of_previous
measure within another measure; since they might be calculated at different times, you may not get accurate results. It also means that percent_of_previous
measures cannot be filtered on.
percent_of_total
type: percent_of_total
calculates a cell's portion of the column total. The percentage is calculated against the total of the rows returned by your query, and not the total of all possible rows. However, if the data returned by your query exceeds a row limit, the field's values will appear as nulls, since it needs the full results to calculate the percent of total.
The sql
parameter for type: percent_of_total
measures must reference another numeric measure.
type: percent_of_total
fields can be formatted by using the value_format
or value_format_name
parameters. However, the percentage formats of the value_format_name
parameter do not work with type: percent_of_total
measures. These percentage formats multiply values by 100, which skews results of a percent_of_total
calculation.
This example LookML creates a percent_of_total_gross_margin
measure that is based on the total_gross_margin
measure:
measure: percent_of_total_gross_margin {
type: percent_of_total
sql: ${total_gross_margin} ;;
}
In cases where a query is pivoted, percent_of_total
runs across the row instead of down the column. If this is not desired, add direction: "column"
to the measure definition.
Additionally, percent_of_total
measures are calculated after data is returned from your database. This means that you should not reference a percent_of_total
measure within another measure; since they might be calculated at different times, you may not get accurate results. It also means that percent_of_total
measures cannot be filtered on.
percentile
type: percentile
returns the value at the specified percentile of values in a given field. For example, specifying the 75th percentile will return the value that is greater than 75% of the other values in the dataset.
To identify the value to return, Looker calculates the total number of data values and multiplies the specified percentile times the total number of data values. Regardless of how the data is actually sorted, Looker identifies the data values' relative order in increasing value. The data value that Looker returns depends on whether the calculation results in an integer or not, as discussed in the following two sections.
If the calculated value is not an integer
Looker rounds the calculated value up and uses it to identify the data value to return. In this example set of 19 test scores, the 75th percentile would be identified by 19 * .75 = 14.25, which means that 75% of the values are in the first 14 data values -- below the 15th position. Thus, Looker returns the 15th data value (87) as being larger than 75% of the data values.
If the calculated value is an integer
In this slightly more complex case, Looker returns an average of the data value at that position and the following data value. To understand this, consider a set of 20 test scores, the 75th percentile would be identified by 20 * .75 = 15, which means that the data value at the 15th position is part of the 75th percentile and we need to return a value that is above 75% of the data values. By returning the average of the values at the 15th position (82) and the 16th position (87), Looker ensures that 75%. That average (84.5) does not exist in the set of data values but would be larger than 75% of the data values.
Required and optional parameters
Use the percentile:
keyword to specify the fractional value, meaning the percent of the data that should be below the returned value. For example, use percentile: 75
to specify the value at the 75th percentile in the order of data, or percentile: 10
to return the value at the 10th percentile. If you want to find the value at the 50th percentile, you can specify percentile: 50
or simply use the median type.
The sql
parameter for type: percentile
measures can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
type: percentile
fields can be formatted by using the value_format
or value_format_name
parameters.
Example
For example, the following LookML creates a field called test_scores_75th_percentile
which returns the value at the 75th percentile in the test_scores
dimension:
measure: test_scores_75th_percentile {
type: percentile
percentile: 75
sql: ${TABLE}.test_scores ;;
}
Things to consider for percentile
If you're using percentile
for a field that is involved in a fanout, Looker will attempt to use percentile_distinct
instead. If percentile_distinct
is not available for the dialect, Looker returns an error. For more information, see the supported dialects for percentile_distinct
.
Supported database dialects for percentile
For Looker to support the percentile
type in your Looker project, your database dialect must also support it. The following table shows which dialects support the percentile
type in the latest release of Looker:
Dialect | Supported? |
---|---|
Actian Avalanche | No |
Amazon Athena | Yes |
Amazon Aurora MySQL | Yes |
Amazon Redshift | Yes |
Apache Druid | No |
Apache Druid 0.13+ | No |
Apache Druid 0.18+ | No |
Apache Hive 2.3+ | Yes |
Apache Hive 3.1.2+ | Yes |
Apache Spark 3+ | Yes |
ClickHouse | Yes |
Cloudera Impala 3.1+ | No |
Cloudera Impala 3.1+ with Native Driver | No |
Cloudera Impala with Native Driver | No |
DataVirtuality | No |
Databricks | Yes |
Denodo 7 | No |
Denodo 8 | No |
Dremio | No |
Dremio 11+ | No |
Exasol | Yes |
Firebolt | No |
Google BigQuery Legacy SQL | Yes |
Google BigQuery Standard SQL | Yes |
Google Cloud PostgreSQL | Yes |
Google Cloud SQL | Yes |
Google Spanner | No |
Greenplum | Yes |
HyperSQL | No |
IBM Netezza | No |
MariaDB | Yes |
Microsoft Azure PostgreSQL | Yes |
Microsoft Azure SQL Database | No |
Microsoft Azure Synapse Analytics | No |
Microsoft SQL Server 2008+ | No |
Microsoft SQL Server 2012+ | No |
Microsoft SQL Server 2016 | No |
Microsoft SQL Server 2017+ | No |
MongoBI | No |
MySQL | Yes |
MySQL 8.0.12+ | Yes |
Oracle | Yes |
Oracle ADWC | Yes |
PostgreSQL 9.5+ | Yes |
PostgreSQL pre-9.5 | Yes |
PrestoDB | Yes |
PrestoSQL | Yes |
SAP HANA 2+ | No |
SingleStore | No |
SingleStore 7+ | Yes |
Snowflake | Yes |
Teradata | No |
Trino | Yes |
Vector | No |
Vertica | Yes |
percentile_distinct
The type: percentile_distinct
is a specialized form of percentile and should be used when your join involves a fanout. It uses the nonrepeated values in a given field, based on the unique values defined by the sql_distinct_key
parameter. If the measure does not have a sql_distinct_key
parameter, then Looker tries to use the primary_key
field.
Consider the result of a query joining the Order Item and Order tables:
Order Item ID | Order ID | Order Shipping |
---|---|---|
1 | 1 | 10 |
2 | 1 | 10 |
3 | 2 | 20 |
4 | 3 | 50 |
5 | 3 | 50 |
6 | 3 | 50 |
7 | 4 | 70 |
8 | 4 | 70 |
9 | 5 | 110 |
10 | 5 | 110 |
In this situation you can see that there are multiple rows for each order. This query involved a fanout because each order maps to several order items. The percentile_distinct
takes this into consideration and finds the percentile value using the distinct values 10, 20, 50, 70, and 110. The 25th percentile will return the second distinct value, or 20, while the 80th percentile will return the average of the fourth and fifth distinct values, or 90.
Required and optional parameters
Use the percentile:
keyword to specify the fractional value. For example, use percentile: 75
to specify the value at the 75th percentile in the order of data, or percentile: 10
to return the value at the 10th percentile. If you are trying to find the value at the 50th percentile, you can use the median_distinct
type instead.
To get an accurate result, specify how Looker should identify each unique entity (in this case, each unique order) by using the sql_distinct_key
parameter.
Here's an example of using percentile_distinct
to return the value at the 90th percentile:
measure: order_shipping_90th_percentile {
type: percentile_distinct
percentile: 90
sql_distinct_key: ${order_id} ;;
sql: ${order_shipping} ;;
}
Every unique value of sql_distinct_key
must have just one corresponding value in the measure's sql
parameter. In other words, the preceding example works because every row with order_id
of 1 has the same order_shipping
of 10, every row with an order_id
of 2 has the same order_shipping
of 20, and so on.
type: percentile_distinct
fields can be formatted by using the value_format
or value_format_name
parameters.
Things to consider for percentile_distinct
If percentile_distinct
is not available for the dialect, Looker returns an error. For more information, see the supported dialects for percentile_distinct
.
Supported database dialects for percentile_distinct
For Looker to support the percentile_distinct
type in your Looker project, your database dialect must also support it. The following table shows which dialects support the percentile_distinct
type in the latest release of Looker:
Dialect | Supported? |
---|---|
Actian Avalanche | No |
Amazon Athena | No |
Amazon Aurora MySQL | Yes |
Amazon Redshift | No |
Apache Druid | No |
Apache Druid 0.13+ | No |
Apache Druid 0.18+ | No |
Apache Hive 2.3+ | No |
Apache Hive 3.1.2+ | No |
Apache Spark 3+ | No |
ClickHouse | No |
Cloudera Impala 3.1+ | No |
Cloudera Impala 3.1+ with Native Driver | No |
Cloudera Impala with Native Driver | No |
DataVirtuality | No |
Databricks | No |
Denodo 7 | No |
Denodo 8 | No |
Dremio | No |
Dremio 11+ | No |
Exasol | No |
Firebolt | No |
Google BigQuery Legacy SQL | Yes |
Google BigQuery Standard SQL | Yes |
Google Cloud PostgreSQL | Yes |
Google Cloud SQL | Yes |
Google Spanner | No |
Greenplum | Yes |
HyperSQL | No |
IBM Netezza | No |
MariaDB | Yes |
Microsoft Azure PostgreSQL | Yes |
Microsoft Azure SQL Database | No |
Microsoft Azure Synapse Analytics | No |
Microsoft SQL Server 2008+ | No |
Microsoft SQL Server 2012+ | No |
Microsoft SQL Server 2016 | No |
Microsoft SQL Server 2017+ | No |
MongoBI | No |
MySQL | Yes |
MySQL 8.0.12+ | Yes |
Oracle | No |
Oracle ADWC | No |
PostgreSQL 9.5+ | Yes |
PostgreSQL pre-9.5 | Yes |
PrestoDB | No |
PrestoSQL | No |
SAP HANA 2+ | No |
SingleStore | No |
SingleStore 7+ | No |
Snowflake | No |
Teradata | No |
Trino | No |
Vector | No |
Vertica | No |
running_total
type: running_total
calculates a cumulative sum of the cells along a column. It cannot be used to calculate sums along a row, unless the row has resulted from a pivot.
The sql
parameter for type: running_total
measures must reference another numeric measure.
type: running_total
fields can be formatted by using the value_format
or value_format_name
parameters.
The following LookML example creates a cumulative_total_revenue
measure that is based on the total_sale_price
measure:
measure: cumulative_total_revenue {
type: running_total
sql: ${total_sale_price} ;;
value_format_name: usd
}
Note that running_total
values depend on sort order. If you change the sort, you must re-run the query to re-calculate the running_total
values. In cases where a query is pivoted, running_total
runs across the row instead of down the column. If this is not desired, add direction: "column"
to the measure definition.
Additionally, running_total
measures are calculated after data is returned from your database. This means that you should not reference a running_total
measure within another measure; since they might be calculated at different times, you may not get accurate results. It also means that running_total
measures cannot be filtered on.
string
type: string
is used with fields that contain letters or special characters.
The sql
parameter for type: string
measures can take any valid SQL expression that results in a string. In practice, this type is rarely used, because most SQL aggregate functions do not return strings. One common exception is MySQL's GROUP_CONCAT
function, although Looker provides type: list
for that use case.
For example, the following LookML creates a field category_list
by combining the unique values of a field called category
:
measure: category_list {
type: string
sql: GROUP_CONCAT(${category}) ;;
}
In this example type: string
could be omitted, because string
is the default value for type
.
sum
type: sum
adds up the values in a given field. It is similar to SQL's SUM
function. However, unlike with raw SQL, Looker will properly calculate sums even if your query's joins contain fanouts.
The sql
parameter for type: sum
measures can take any valid SQL expression that results in a numeric table column, LookML dimension, or combination of LookML dimensions.
type: sum
fields can be formatted by using the value_format
or value_format_name
parameters.
For example, the following LookML creates a field called total_revenue
by adding up the sales_price
dimension, then displays it in a money format ($1,234.56):
measure: total_revenue {
type: sum
sql: ${sales_price} ;;
value_format_name: usd
}
sum_distinct
type: sum_distinct
is for use with denormalized datasets. It adds up the nonrepeated values in a given field, based on the unique values defined by the sql_distinct_key
parameter.
This is an advanced concept which may be more clearly explained with an example. Consider a denormalized table like this:
Order Item ID | Order ID | Order Shipping |
---|---|---|
1 | 1 | 10.00 |
2 | 1 | 10.00 |
3 | 2 | 20.00 |
4 | 2 | 20.00 |
5 | 2 | 20.00 |
In this situation you can see that there are multiple rows for each order. Consequently, if you added a simple type: sum
measure for the order_shipping
column, you would get a total of 80.00, even though the total shipping collected is actually 30.00.
# Will NOT calculate the correct shipping amount
measure: total_shipping {
type: sum
sql: ${order_shipping} ;;
}
To get an accurate result, you can define for Looker how it should identify each unique entity (in this case, each unique order) by using the sql_distinct_key
parameter. This will calculate the correct 30.00 amount:
# Will calculate the correct shipping amount
measure: total_shipping {
type: sum_distinct
sql_distinct_key: ${order_id} ;;
sql: ${order_shipping} ;;
}
Every unique value of sql_distinct_key
must have just one corresponding value in sql
. In other words, the preceding example works because every row with an order_id
of 1 has the same order_shipping
of 10.00, every row with an order_id
of 2 has the same order_shipping
of 20.00, and so on.
type: sum_distinct
fields can be formatted by using the value_format
or value_format_name
parameters.
yesno
type: yesno
creates a field that indicates if something is true or false. The values appear as Yes and No in the Explore UI.
The sql
parameter for a type: yesno
measure takes a valid SQL expression that evaluates to TRUE
or FALSE
. If the condition evaluates to TRUE
, Yes is displayed to the user; otherwise, No is displayed.
The SQL expression for type: yesno
measures must include only aggregations, which means SQL aggregations or references to LookML measures. If you want to create a yesno
field that includes a reference to a LookML dimension or a SQL expression that is not an aggregation, use a dimension with type: yesno
, not a measure.
Similar to measures with type: number
, a measure with type: yesno
doesn't do any aggregations; it just references other aggregations.
For example, the following total_sale_price
measure example is a sum of the total sale price of order items in an order. A second measure called is_large_total
is type: yesno
. The is_large_total
measure has a sql
parameter that evaluates whether the total_sale_price
value is greater than $1,000.
measure: total_sale_price {
type: sum
value_format_name: usd
sql: ${sale_price} ;;
drill_fields: [detail*]
}
measure: is_large_total {
description: "Is order total over $1000?"
type: yesno
sql: ${total_sale_price} > 1000 ;;
}
If you want to reference a type: yesno
field in another field, you should treat the type: yesno
field as a boolean (in other words, as if it contains a true or false value already). For example:
measure: is_large_total {
description: "Is order total over $1000?"
type: yesno
sql: ${total_sale_price} > 1000 ;;
}
# This is correct:
measure: reward_points {
type: number
sql: CASE WHEN ${is_large_total} THEN 200 ELSE 100 END ;;
}
# This is NOT correct:
measure: reward_points {
type: number
sql: CASE WHEN ${is_large_total} = 'Yes' THEN 200 ELSE 100 END ;;
}