##### Page tree

Release 5.1

Contents:

Contents:

Computes the rolling average of values forward or backward of the current row within the specified column.
• If an input value is missing or null, it is not factored in the computation. For example, for the first row in the dataset, the rolling average of previous values is the value in the first row.
• The row from which to extract a value is determined by the order in which the rows are organized based on the `order` parameter.

If you are working on a randomly generated sample of your dataset, the values that you see for this function might not correspond to the values that are generated on the full dataset during job execution.

• The function takes a column name and two optional integer parameters that determine the window backward and forward of the current row.
• The default integer parameter values are `-1` and `0`, which computes the rolling average from the current row back to the first row of the dataset.
• This function works with the following transforms:

For more information on a non-rolling version of this function, see AVERAGE Function.

## Basic Usage

Column example:

derive type:single value:ROLLINGAVERAGE(myCol)

Output: Generates a new column containing the rolling average of all values in the `myCol` column from the first row of the dataset to the current one.

Rows before example:

window value:ROLLINGAVERAGE(myNumber, 3)

Output: Generates the new column, which contains the rolling average of the current row and the three previous row values in the `myNumber` column.

Rows before and after example:

window value:ROLLINGAVERAGE(myNumber, 3, 2)

Output: Generates the new column , which contains the rolling average of the three previous row values, the current row value, and the two rows after the current one in the `myNumber` column.

## Syntax and Arguments

window value:ROLLINGAVERAGE(col_ref, rowsBefore_integer, rowsAfter_integer) order: order_col [group: group_col]

ArgumentRequired?Data TypeDescription
col_refYstringName of column whose values are applied to the function
rowsBefore_integerNintegerNumber of rows before the current one to include in the computation
rowsAfter_integerNintegerNumber of rows after the current one to include in the computation

For more information on the `order` and `group` parameters, see Window Transform.

### col_ref

Name of the column whose values are used to compute the rolling average.

• Multiple columns and wildcards are not supported.

Usage Notes:

Required?Data TypeExample Value
YesString (column reference to Integer or Decimal values)`myColumn`

### rowsBefore_integer, rowsAfter_integer

Integers representing the number of rows before or after the current one from which to compute the rolling average, including the current row. For example, if the first value is `5`, the current row and the five rows before it are used in the computation. Negative values for `k` compute the rolling average from rows preceding the current one.

• `rowBefore=0` generates the current row value only.
• `rowBefore=-1`  uses all rows preceding the current one.
• If `rowsAfter` is not specified, then the value `0` is applied.
• If a `group` parameter is applied, then these parameter values should be no more than the maximum number of rows in the groups.

Usage Notes:

Required?Data TypeExample Value
NoInteger`4`

## Examples

### Example - Compute prior quarter values

The following dataset contains order information for the preceding 12 months. You want to compare the current month's average against the preceding quarter.

Source:

DateAmount
12/31/15118
11/30/156
10/31/15443
9/30/15785
8/31/1577
7/31/15606
6/30/15421
5/31/15763
4/30/15305
3/31/15824
2/28/15135
1/31/15523

Transform:

Using the `ROLLINGAVERAGE` function, you can generate a column containing the rolling average of the current month and the two previous months:

window value: ROLLINGAVERAGE(Amount, 3, 0) order: -Date

Note the sign of the second parameter and the `order` parameter. The sort is in the reverse order of the `Date` parameter, which preserves the current sort order. As a result, the second parameter, which identifies the number of rows to use in the calculation, must be positive to capture the previous months.

Technically, this computation does not capture the prior quarter, since it includes the current quarter as part of the computation. You can use the following column to capture the rolling average of the preceding month, which then becomes the true rolling average for the prior quarter. The `window` column refers to the name of the column generated from the previous step:

window value: NEXT(window, 1) order: -Date

Note that the order parameter must be preserved. This new column, `window1`, contains your prior quarter rolling average:

rename col:window1 to:'Amount_PriorQtr'

You can reformat this numeric value:

set col:Amount_PriorQtr value:NUMFORMAT(Amount_PriorQtr, '###.00')

You can use the following transform to calculate the net change. This formula computes the change as a percentage of the prior quarter and then formats it as a two-digit percentage.

derive type:single value:NUMFORMAT(((Amount - Amount_PriorQtr) / Amount_PriorQtr) * 100, '##.##') as:'NetChangePct_PriorQtr'

Results:

NOTE: You might notice that there are computed values for `Amount_PriorQtr` for February and March. These values do not factor in a full three months because the data is not present. The January value does not exist since there is no data preceding it.

DateAmountAmount_PriorQtrNetChangePct_PriorQtr
12/31/15118411.33-71.31
11/30/156435.00-98.62
10/31/15443489.33-9.47
9/30/15785368.00113.32
8/31/1577596.67-87.1
7/31/15606496.3322.1
6/30/15421630.67-33.25
5/31/15763421.3381.09
4/30/15305494.00-38.26
3/31/15824329.00150.46
2/28/15135523.00-.74.19
1/31/15523

### Example - Rolling window functions

This example describes how to use the rolling computational functions:
• `ROLLINGSUM` - computes a rolling sum from a window of rows before and after the current row. See ROLLINGSUM Function.
• `ROLLINGAVERAGE` - computes a rolling average from a window of rows before and after the current row. See ROLLINGAVERAGE Function.
• `ROWNUMBER` - computes the row number for each row, as determined by the ordering column. See ROWNUMBER Function.

The following dataset contains sales data over the final quarter of the year.

Source:

DateSales
10/2/16200
10/9/16500
10/16/16350
10/23/16400
10/30/16190
11/6/16550
11/13/16610
11/20/16480
11/27/16660
12/4/16690
12/11/16810
12/18/16950
12/25/161020
1/1/17680

Transform:

First, you want to maintain the row information as a separate column. Since data is ordered already by the `Date` column, you can use the following:

window value:ROWNUMBER() order:Date

Rename this column to `rowId` for week of quarter.

Now, you want to extract month and week information from the `Date` values. Deriving the month value:

derive type:single value:MONTH(Date) as:'Month'

Deriving the quarter value:

derive type:single value:(1 + FLOOR(((month-1)/3))) as:'QTR'

Deriving the week-of-quarter value:

window value:ROWNUMBER() order:Date group:QTR

Rename this column `WOQ` (week of quarter).

Deriving the week-of-month value:

window value:ROWNUMBER() group:Month order:Date

Rename this column `WOM` (week of month).

Now, you perform your rolling computations. Compute the running total of sales using the following:

window value: ROLLINGSUM(Sales, -1, 0) order: Date group:QTR

The `-1` parameter is used in the above computation to gather the rolling sum of all rows of data from the current one to the first one. Note that the use of the `QTR` column for grouping, which moves the value for the `01/01/2017` into its own computational bucket. This may or may not be preferred.

Rename this column `QTD` (quarter to-date). Now, generate a similar column to compute the rolling average of weekly sales for the quarter:

window value: ROUND(ROLLINGAVERAGE(Sales, -1, 0)) order: Date group:QTR

Since the `ROLLINGAVERAGE` function can compute fractional values, it is wrapped in the `ROUND` function for neatness. Rename this column `avgWeekByQuarter`.

Results:

When the unnecessary columns are dropped and some reordering is applied, your dataset should look like the following:

DateWOQSalesQTDavgWeekByQuarter
10/2/161200200200
10/9/162500700350
10/16/1633501050350
10/23/1644001450363
10/30/1651901640328
11/6/1665502190365
11/13/1676102800400
11/20/1684803280410
11/27/1696603940438
12/4/16106904630463
12/11/16118105440495
12/18/16129506390533
12/25/161310207410570
1/1/171680680680

### Example - Rolling computations for racing splits

This example describes how to use the rolling computational functions:
• `ROLLINGAVERAGE` - computes a rolling average from a window of rows before and after the current row. See ROLLINGAVERAGE Function.
• `ROLLINGMIN` - computes a rolling minimum from a window of rows. See ROLLINGMIN Function.
• `ROLLINGMAX` - computes a rolling maximum from a window of rows.  See ROLLINGMAX Function.
• `ROLLINGSTDEV` - computes a rolling standard deviation from a window of rows. See ROLLINGSTDEV Function.
• `ROLLINGVAR` - computes a rolling variance from a window of rows. See ROLLINGVAR Function.

Source:

In this example, the following data comes from times recorded at regular intervals during a three-lap race around a track. The source data is in cumulative time in seconds (`time_sc`). You can use ROLLING and other windowing functions to break down the data into more meaningful metrics.

lapquartertime_sc
100.000
1119.554
1239.785
1360.021
2080.950
21101.785
22121.005
23141.185
30162.008
31181.887
32200.945
33220.856

Transform:

Primary key: Since the quarter information repeats every lap, there is no unique identifier for each row. The following steps create this identifer:

settype col: lap,quarter type: 'String'

derive type:single value: MERGE(['l',lap,'q',quarter]) as: 'splitId'

Get split times: Use the following transform to break down the splits for each quarter of the race:

derive type:single value: ROUND(time_sc - PREV(time_sc, 1), 3) order: splitId as: 'split_time_sc'

Compute rolling computations: You can use the following types of computations to provide rolling metrics on the current and three previous splits:

derive type:single value: ROLLINGAVERAGE(split_time_sc, 3) order: splitId as: 'ravg'

derive type:single value: ROLLINGMAX(split_time_sc, 3) order: splitId as: 'rmax'

derive type:single value: ROLLINGMIN(split_time_sc, 3) order: splitId as: 'rmin'

derive type:single value: ROUND(ROLLINGSTDEV(split_time_sc, 3), 3) order: splitId as: 'rstdev'

derive type:single value: ROUND(ROLLINGVAR(split_time_sc, 3), 3) order: splitId as: 'rvar'

Results:

When the above transforms have been completed, the results look like the following:

lapquartersplitIdtime_scsplit_time_scrvarrstdevrminrmaxravg
10l1q00
11l1q120.09620.0960020.09620.09620.096
12l1q240.5320.4340.0290.16920.09620.43420.265
13l1q361.03120.5010.0310.17720.09620.50120.344
20l2q081.08720.0560.0390.19820.05620.50120.272
21l2q1101.38320.2960.0290.1720.05620.50120.322
22l2q2122.09220.7090.0590.24220.05620.70920.39
23l2q3141.88619.7940.1130.33719.79420.70920.214
30l3q0162.58120.6950.1390.37319.79420.70920.373
31l3q1183.01820.4370.1380.37119.79420.70920.409
32l3q2203.49320.4750.1130.33619.79420.69520.35
33l3q3222.89319.40.2520.50219.420.69520.252

You can reduce the number of steps by applying a `window` transform such as the following:

window value: window1 = lap,rollingaverage(split_time_sc, 0, 3), rollingmax(split_time_sc, 0, 3),rollingmin(split_time_sc, 0, 3),round(rollingstdev(split_time_sc, 0, 3), 3),round(rollingvar(split_time_sc, 0, 3), 3) group: lap order: lap

However, you must rename all of the generated `windowX` columns.