Random Sampling in Google BigQuery
insights Stats
Data analytics with Google Cloud Platform (GCP). For BigQuery SQL mentioned in the articles, they are all using standard SQL dialect unless specified differently.
In databases like Teradata, SAMPLE clause can be used to sample data from a large data table. Google BigQuery SQL didn't support SAMPLE clause at the very beginning however now TABLESAMPLE clause is added to support querying random sample subset of a large data set.
Use RAND function
Before TABLESAMPLE is added, RAND function is used as an alternative to retrieve random sample subset. The querying cost is big as the whole table will be scanned to generate one random number for each record.
Example
SELECT col1, col2, RAND(5) AS rnd FROM `myproj.mydataset.mytable` ORDER BY rnd LIMIT 10;
The above query returns 10 random records.
Use TABLESAMPLE clause
TABLESAMPLE clause is used to sampling a subset from a large data set.
Example
SELECT * FROM `myproj.mydataset.mytable` TABLESAMPLE SYSTEM (5 PERCENT)
The above example samples 5 percent of the data available in the specified table. The clause doesn't support specifying records volume. For more information about table sampling, refer to Table sampling | BigQuery | Google Cloud.