Aggregate functions

COUNT

Counting the number of rows in the table (if * or constant is specified as the argument) or non-empty values in a table column (if the column name is specified as an argument).

Like other aggregate functions, it can be combined with GROUP BY to get statistics on the parts of the table that correspond to the values in the columns being grouped. Use the modifier DISTINCT to count distinct values.

Examples

SELECT COUNT(*) FROM my_table;

SELECT key, COUNT(value) FROM my_table GROUP BY key;

SELECT COUNT(DISTINCT value) FROM my_table;

MIN and MAX

Minimum or maximum value.

As an argument, you may use an arbitrary computable expression with a numeric result.

Examples

SELECT MIN(value), MAX(value) FROM my_table;

SUM

Sum of the numbers.

As an argument, you may use an arbitrary computable expression with a numeric result.

Integers are automatically expanded to 64 bits to reduce the risk of overflow.

SELECT SUM(value) FROM my_table;

AVG

Arithmetic average.

As an argument, you may use an arbitrary computable expression with a numeric result.

Integer values and time intervals are automatically converted to Double.

Examples

SELECT AVG(value) FROM my_table;

COUNT_IF

Number of rows for which the expression specified as the argument is true (the expression's calculation result is true).

The value NULL is equated to false (if the argument type is Bool?).

The function does not do the implicit type casting to Boolean for strings and numbers.

Examples

SELECT
  COUNT_IF(value % 2 == 1) AS odd_count

SUM_IF and AVG_IF

Sum or arithmetic average, but only for the rows that satisfy the condition passed by the second argument.

Therefore, SUM_IF(value, condition) is a slightly shorter notation for SUM(IF(condition, value)), same for AVG. The argument's data type expansion is similar to the same-name functions without a suffix.

Examples

SELECT
    SUM_IF(value, value % 2 == 1) AS odd_sum,
    AVG_IF(value, value % 2 == 1) AS odd_avg,
FROM my_table;

When you use aggregation factories, a Tuple containing a value and a predicate is passed as the first AGGREGATE_BY argument.

$sum_if_factory = AggregationFactory("SUM_IF");
$avg_if_factory = AggregationFactory("AVG_IF");

SELECT
    AGGREGATE_BY(AsTuple(value, value % 2 == 1), $sum_if_factory) AS odd_sum,
    AGGREGATE_BY(AsTuple(value, value % 2 == 1), $avg_if_factory) AS odd_avg
FROM my_table;

SOME

Get the value for an expression specified as an argument, for one of the table rows. Gives no guarantee of which row is used. It's similar to the any() function in ClickHouse.

Because of no guarantee, SOME is computationally cheaper than MIN / MAX often used in similar situations.

Examples

SELECT
  SOME(value)
FROM my_table;

CountDistinctEstimate, HyperLogLog, and HLL

Approximating the number of unique values using the HyperLogLog algorithm. Logically, it does the same thing as COUNT(DISTINCT ...), but runs much faster at the cost of some error.

Arguments:

1. Estimated value
2. Accuracy (4 to 18 inclusive, 14 by default).

By selecting accuracy, you can trade added resource and RAM consumption for decreased error.

All the three functions are aliases at the moment, but CountDistinctEstimate may start using a different algorithm in the future.

Examples

SELECT
  CountDistinctEstimate(my_column)
FROM my_table;

SELECT
  HyperLogLog(my_column, 4)
FROM my_table;

AGGREGATE_LIST

Get all column values as a list. When combined with DISTINCT, it returns only distinct values. The optional second parameter sets the maximum number of values to be returned. A zero limit value means unlimited.

If you know already that you have few distinct values, use the AGGREGATE_LIST_DISTINCT aggregate function to build the same result in memory (that might not be enough for a large number of distinct values).

The order of elements in the result list depends on the implementation and can't be set externally. To return an ordered list, sort the result, for example, with ListSort.

To return a list of multiple values from one line, DO NOT use the AGGREGATE_LIST function several times, but add all the needed values to a container, for example, via AsList or AsTuple, then pass this container to a single AGGREGATE_LIST call.

For example, you can combine it with DISTINCT and the function String::JoinFromList (it's an equivalent of ','.join(list) in Python) to output to a string all the values found in the column after GROUP BY.

Examples

SELECT
   AGGREGATE_LIST( region ),
   AGGREGATE_LIST( region, 5 ),
   AGGREGATE_LIST( DISTINCT region ),
   AGGREGATE_LIST_DISTINCT( region ),
   AGGREGATE_LIST_DISTINCT( region, 5 )
FROM users

-- An equivalent of GROUP_CONCAT in MySQL
SELECT
    String::JoinFromList(CAST(AGGREGATE_LIST(region, 2) AS List<String>), ",")
FROM users

These functions also have a short notation: AGG_LIST and AGG_LIST_DISTINCT.

MAX_BY and MIN_BY

Return the value of the first argument for the table row where the second argument is minimum/maximum.

You can optionally specify the third argument N that affects behavior if the table has multiple rows with the same minimum or maximum value:

• If N is omitted, the value of one of the rows is returned, and the other rows are discarded.
• If N is specified, the list is returned with all values, but their number can't exceed N. All values after the number are discarded.

When choosing N, we recommend that you don't exceed several hundreds or thousands to avoid issues with the limited memory available on YDB clusters.

If your task needs absolutely all values, and their number is measured in dozens of thousands or more, then instead of those aggregate functions better use JOIN on the source table with a subquery doing GROUP BY + MIN/MAX on the desired columns of this table.

When you use aggregation factories, a Tuple containing a value and a key is passed as the first AGGREGATE_BY argument.

Examples

SELECT
  MIN_BY(value, LENGTH(value)),
  MAX_BY(value, key, 100)
FROM my_table;

$min_by_factory = AggregationFactory("MIN_BY");
$max_by_factory = AggregationFactory("MAX_BY", 100);

SELECT
    AGGREGATE_BY(AsTuple(value, LENGTH(value)), $min_by_factory),
    AGGREGATE_BY(AsTuple(value, key), $max_by_factory)
FROM my_table;

TOP and BOTTOM

Return a list of the maximum/minimum values of an expression. The first argument is an expression, the second argument limits the number of items.

Examples

SELECT
    TOP(key, 3),
    BOTTOM(value, 3)
FROM my_table;

$top_factory = AggregationFactory("TOP", 3);
$bottom_factory = AggregationFactory("BOTTOM", 3);

SELECT
    AGGREGATE_BY(key, $top_factory),
    AGGREGATE_BY(value, $bottom_factory)
FROM my_table;

TOP_BY and BOTTOM_BY

Return a list of values of the first argument for the rows containing the maximum/minimum values of the second argument. The third argument limits the number of items in the list.

When you use aggregation factories, a Tuple containing a value and a key is passed as the first AGGREGATE_BY argument. In this case, the limit for the number of items is passed by the second argument at factory creation.

Examples

SELECT
    TOP_BY(value, LENGTH(value), 3),
    BOTTOM_BY(value, key, 3)
FROM my_table;

$top_by_factory = AggregationFactory("TOP_BY", 3);
$bottom_by_factory = AggregationFactory("BOTTOM_BY", 3);

SELECT
    AGGREGATE_BY(AsTuple(value, LENGTH(value)), $top_by_factory),
    AGGREGATE_BY(AsTuple(value, key), $bottom_by_factory)
FROM my_table;

TOPFREQ and MODE

Getting an approximate list of the most common values in a column with an estimation of their count. Returns a list of structures with two fields:

• Value: the frequently occurring value that was found.
• Frequency: An estimated value occurrence in the table.

Required argument: the value itself.

Optional arguments:

1. For TOPFREQ, the desired number of items in the result. MODE is an alias to TOPFREQ with this argument set to 1. For TOPFREQ, this argument is also 1 by default.
2. The number of items in the buffer used: lets you trade memory consumption for accuracy. Default: 100.

Examples

SELECT
    MODE(my_column),
    TOPFREQ(my_column, 5, 1000)
FROM my_table;

STDDEV and VARIANCE

Standard deviation and variance in a column. Those functions use a single-pass parallel algorithm, whose result may differ from the more common methods requiring two passes through the data.

By default, the sample variance and standard deviation are calculated. Several write methods are available:

• with the POPULATION suffix/prefix, for example: VARIANCE_POPULATION, POPULATION_VARIANCE calculates the variance or standard deviation for the population.
• With the SAMPLE suffix/prefix or without a suffix, for example, VARIANCE_SAMPLE, SAMPLE_VARIANCE, SAMPLE calculate sample variance and standard deviation.

Several abbreviated aliases are also defined, for example, VARPOP or STDDEVSAMP.

If all the values passed are NULL, it returns NULL.

Examples

SELECT
  STDDEV(numeric_column),
  VARIANCE(numeric_column)
FROM my_table;

CORRELATION and COVARIANCE

Correlation and covariance between two columns.

Abbreviated versions are also available: CORR or COVAR. For covariance, there are also versions with the SAMPLE/POPULATION suffix that are similar to VARIANCE above.

Unlike most other aggregate functions, they don't skip NULL, but accept it as 0.

When you use aggregation factories, a Tuple containing two values is passed as the first AGGREGATE_BY argument.

Examples

SELECT
  CORRELATION(numeric_column, another_numeric_column),
  COVARIANCE(numeric_column, another_numeric_column)
FROM my_table;

$corr_factory = AggregationFactory("CORRELATION");

SELECT
    AGGREGATE_BY(AsTuple(numeric_column, another_numeric_column), $corr_factory)
FROM my_table;

PERCENTILE and MEDIAN

Calculating percentiles using the amortized version of the TDigest algorithm. MEDIAN: An alias for PERCENTILE(N, 0.5).

SELECT
    MEDIAN(numeric_column),
    PERCENTILE(numeric_column, 0.99)
FROM my_table;

HISTOGRAM

Plotting an approximate histogram based on a numeric expression with automatic selection of buckets.

Auxiliary functions

Basic settings

You can limit the number of buckets using an optional argument. The default value is 100. Keep in mind that added accuracy costs you more computing resources and may negatively affect the query execution time. In extreme cases, it may affect your query success.

Support for weights

You can specify a "weight" for each value used in the histogram. To do this, pass to the aggregate function the second argument with an expression for calculating the weight. The weight of 1.0 is always used by default. If you use non-standard weights, you may also use the third argument to limit the number of buckets.

If you pass two arguments, the meaning of the second argument is determined by its type (if it's an integer literal, it limits the number of buckets, otherwise it's used as a weight).

If you need an accurate histogram

1. You can use the aggregate functions described below with fixed bucket grids: LinearHistogram or LogarithmicHistogram.
2. You can calculate the bucket number for each row and apply to it GROUP BY.

When you use aggregation factories, a Tuple containing a value and a weight is passed as the first AGGREGATE_BY argument.

Examples

SELECT
    HISTOGRAM(numeric_column)
FROM my_table;

SELECT
    Histogram::Print(
        HISTOGRAM(numeric_column, 10),
        50
    )
FROM my_table;

$hist_factory = AggregationFactory("HISTOGRAM");

SELECT
    AGGREGATE_BY(AsTuple(numeric_column, 1.0), $hist_factory)
FROM my_table;

LinearHistogram, LogarithmicHistogram, and LogHistogram

Plotting a histogram based on an explicitly specified fixed bucket scale.

Arguments:

1. Expression used to plot the histogram. All the following arguments are optional.
2. Spacing between the LinearHistogram buckets or the logarithm base for LogarithmicHistogram/LogHistogram (those are aliases). In both cases, the default value is 10.
3. Minimum value. By default, it's minus infinity.
4. Maximum value. By default, it's plus infinity.

The format of the result is totally similar to adaptive histograms, so you can use the same set of auxiliary functions.

If the spread of input values is uncontrollably large, we recommend that you specify the minimum and maximum values to prevent potential failures due to high memory consumption.

Examples

SELECT
    LogarithmicHistogram(numeric_column, 2)
FROM my_table;

BOOL_AND, BOOL_OR and BOOL_XOR

Apply the relevant logical operation (AND/OR/XOR) to all values in a Boolean column or expression.

These functions don't skip NULL during aggregation, with even one NULL turning the result into NULL. To skip NULLs during aggregation, you can use the functions MIN/MAX or BIT_AND/BIT_OR/BIT_XOR.

Examples

SELECT
  BOOL_AND(bool_column),
  BOOL_OR(bool_column),
  BOOL_XOR(bool_column)
FROM my_table;

BIT_AND, BIT_OR and BIT_XOR

Apply the relevant bitwise operation to all values of a numeric column or expression.

Examples

SELECT
    BIT_XOR(unsigned_numeric_value)
FROM my_table;

SessionStart

No arguments. It's allowed only if there is SessionWindow in GROUP BY / PARTITION BY.
Returns the value of the SessionWindow key column. If SessionWindow has two arguments, it returns the minimum value of the first argument within the group/section.
In the case of the expanded version SessionWindow, it returns the value of the second element from the tuple returned by <calculate_lambda>, for which the first tuple element is True.

AGGREGATE_BY and MULTI_AGGREGATE_BY

Applying an aggregation factory to all values of a column or expression. The MULTI_AGGREGATE_BY function requires that the value of a column or expression has a structure, tuple, or list, and applies the factory to each individual element, placing the result in a container of the same format. If different values of a column or expression contain lists of different length, the resulting list will have the smallest of the source lengths.

1. Column, DISTINCT column or expression.
2. Factory.

Examples

$count_factory = AggregationFactory("COUNT");

SELECT
    AGGREGATE_BY(DISTINCT column, $count_factory) as uniq_count
FROM my_table;

SELECT
    MULTI_AGGREGATE_BY(nums, AggregationFactory("count")) as count,
    MULTI_AGGREGATE_BY(nums, AggregationFactory("min")) as min,
    MULTI_AGGREGATE_BY(nums, AggregationFactory("max")) as max,
    MULTI_AGGREGATE_BY(nums, AggregationFactory("avg")) as avg,
    MULTI_AGGREGATE_BY(nums, AggregationFactory("percentile", 0.9)) as p90
FROM my_table;