window_funnel

Description

Searches for an event chain in a sliding window and calculates the maximum number of consecutive events in the event chain. This function is commonly used for analyzing conversion rate. It is supported from v2.3.

This function works according to the following rules:

It starts the count from the first event in the event chain. If the first event is found, the event counter is set to 1 and the sliding window starts. If the first event is not found, 0 is returned.
In the sliding window, the counter is incremented if the events in the event chain occur in sequence. If the sliding window is exceeded, the event counter is no longer incremented.
If multiple event chains match the specified condition, the longest event chain is returned.

Syntax

BIGINT window_funnel(BIGINT window, DATE|DATETIME time, INT mode, array[cond1, cond2, ..., condN])

Parameters

window: The length of the sliding window. The supported data type is BIGINT. The unit depends on the time parameter. If the data type of time is DATE, the unit is days. If the data type of time is DATETIME, the unit is seconds.
time: The column containing timestamps. DATE and DATETIME types are supported.
mode: The mode in which the event chain is filtered. The supported data type is INT. Value range: 0, 1, 2.
- 0 is the default value, which indicates common funnel calculation.
- 1 indicates the DEDUPLICATION mode, that is, the filtered event chain cannot have repeated events. Suppose the array parameter is [event_type = 'A', event_type = 'B', event_type = 'C', event_type = 'D'] and the original event chain is "A-B-C-B-D". Event B is repeated and the filtered event chain is "A-B-C".
- 2 indicates the FIXED mode, that is, the filtered event chain cannot have events that disrupt the specified sequence. Suppose the previous array parameter is used and the original event chain is "A-B-D-C". Event D interrupts the sequence and the filtered event chain is "A-B".
- 4 indicates the INCREASE mode, which means the filtered events must have strictly increasing timestamps. Duplicate timestamp disrupts the event chain. This mode is supported since version 2.5.
array: The defined event chain. It must be an array.

Return value

Returns a value of the BIGINT type.

Examples

Example 1: Calculate the maximum number of consecutive events based on uid. The sliding window is 1800s and the filtering mode is 0.

This example uses table action, in which data is sorted by uid.

mysql> select * from action;
+------+------------+---------------------+
| uid  | event_type | time                |
+------+------------+---------------------+
| 1    | Browse     | 2020-01-02 11:00:00 |
| 1    | Click      | 2020-01-02 11:10:00 |
| 1    | Order      | 2020-01-02 11:20:00 |
| 1    | Pay        | 2020-01-02 11:30:00 |
| 1    | Browse     | 2020-01-02 11:00:00 |
| 2    | Order      | 2020-01-02 11:00:00 |
| 2    | Pay        | 2020-01-02 11:10:00 |
| 3    | Browse     | 2020-01-02 11:20:00 |
| 3    | Click      | 2020-01-02 12:00:00 |
| 4    | Browse     | 2020-01-02 11:50:00 |
| 4    | Click      | 2020-01-02 12:00:00 |
| 5    | Browse     | 2020-01-02 11:50:00 |
| 5    | Click      | 2020-01-02 12:00:00 |
| 5    | Order      | 2020-01-02 11:10:00 |
| 6    | Browse     | 2020-01-02 11:50:00 |
| 6    | Click      | 2020-01-02 12:00:00 |
| 6    | Order      | 2020-01-02 12:10:00 |
+------+------------+---------------------+
17 rows in set (0.01 sec)

Execute the following statement:

select uid,
       window_funnel(1800,time,0,[event_type='Browse', event_type='Click', 
        event_type='Order', event_type='Pay']) AS level
from action
group by uid
order by uid; 
+------+-------+
| uid  | level |
+------+-------+
| 1    |     4 |
| 2    |     0 |
| 3    |     1 |
| 4    |     2 |
| 5    |     2 |
| 6    |     3 |
+------+-------+

Description of the results:

The matching event chain for uid = 1 is "Browse-Click-Order-Pay", and 4 is returned. The time of the last "Browse" event (2020-01-02 11:00:00) does not meet the condition and is not counted.
The event chain of uid = 2 does not start from the first event "Browse", and 0 is returned.
The matching event chain for uid = 3 is "Browse", and 1 is returned. The "Click" event exceeds the 1800s time window and is not counted.
The matching event chain for uid = 4 is "Browse-Click", and 2 is returned.
The matching event chain for uid = 5 is "Browse-Click", and 2 is returned. The "Order" event (2020-01-02 11:10:00) does not belong to the event chain and is not counted.
The matching event chain for uid = 6 is "Browse-Click-Order", and 3 is returned.

Example 2: Calculate the maximum number of consecutive events based on uid. The sliding window is 1800s, and filtering modes 0 and 1 are used.

This example uses table action1, in which data is sorted by time.

mysql> select * from action1 order by time;
+------+------------+---------------------+ 
| uid  | event_type | time                |     
+------+------------+---------------------+
| 1    | Browse     | 2020-01-02 11:00:00 |
| 2    | Browse     | 2020-01-02 11:00:01 |
| 1    | Click      | 2020-01-02 11:10:00 |
| 1    | Order      | 2020-01-02 11:29:00 |
| 1    | Click      | 2020-01-02 11:29:50 |
| 1    | Pay        | 2020-01-02 11:30:00 |
| 1    | Click      | 2020-01-02 11:40:00 |
+------+------------+---------------------+
7 rows in set (0.03 sec)

Execute the following statement:

select uid,
       window_funnel(1800,time,0,[event_type='Browse', 
        event_type='Click', event_type='Order', event_type='Pay']) AS level
from action1
group by uid
order by uid;
+------+-------+
| uid  | level |
+------+-------+
| 1    |     4 |
| 2    |     1 |
+------+-------+
2 rows in set (0.02 sec)

For uid = 1, the "Click" event (2020-01-02 11:29:50) is a repeated event but it is still counted because mode 0 is used. Therefore, 4 is returned.

Change mode to 1 and execute the statement again.

+------+-------+
| uid  | level |
+------+-------+
| 1    |     3 |
| 2    |     1 |
+------+-------+
2 rows in set (0.05 sec)

The longest event chain filtered after deduplication is "Browse-Click-Order", and 3 is returned.

Example 3: Calculate the maximum number of consecutive events based on uid. The sliding window is 1900s, and filter modes 0 and 2 are used.

This example uses table action2, in which data is sorted by time.

mysql> select * from action2 order by time;
+------+------------+---------------------+
| uid  | event_type | time                |
+------+------------+---------------------+
| 1    | Browse     | 2020-01-02 11:00:00 |
| 2    | Browse     | 2020-01-02 11:00:01 |
| 1    | Click      | 2020-01-02 11:10:00 |
| 1    | Pay        | 2020-01-02 11:30:00 |
| 1    | Order      | 2020-01-02 11:31:00 |
+------+------------+---------------------+
5 rows in set (0.01 sec)

Execute the following statement:

select uid,
       window_funnel(1900,time,0,[event_type='Browse', event_type='Click', 
        event_type='Order', event_type='Pay']) AS level
from action2
group by uid
order by uid;
+------+-------+
| uid  | level |
+------+-------+
| 1    |     3 |
| 2    |     1 |
+------+-------+
2 rows in set (0.02 sec)

3 is returned for uid = 1 because mode 0 is used and the "Pay" event (2020-01-02 11:30:00) does not disrupt the event chain.

Change mode to 2 and execute the statement again.

select uid,
       window_funnel(1900,time,2,[event_type='Browse', event_type='Click', 
        event_type='Order', event_type='Pay']) AS level
from action2
group by uid
order by uid;
+------+-------+
| uid  | level |
+------+-------+
| 1    |     2 |
| 2    |     1 |
+------+-------+
2 rows in set (0.06 sec)

2 is returned because the "Pay" event disrupts the event chain and the event counter stops. The filtered event chain is "Browse-Click".

Example 4: Calculate the maximum number of consecutive events based on uid. The sliding window is 1900s, and filter modes 0 and 4 are used.

This example uses table action3, in which data is sorted by time.

select * from action3 order by time;
+------+------------+---------------------+
| uid  | event_type | time                |
+------+------------+---------------------+
| 1    | Browse     | 2020-01-02 11:00:00 |
| 1    | Click      | 2020-01-02 11:00:01 |
| 2    | Browse     | 2020-01-02 11:00:03 |
| 1    | Order      | 2020-01-02 11:00:31 |
| 2    | Click      | 2020-01-02 11:00:03 |
| 2    | Order      | 2020-01-02 11:01:03 |
+------+------------+---------------------+
3 rows in set (0.02 sec)

Execute the following statement:

select uid,
       window_funnel(1900,time,0,[event_type='Browse', event_type='Click',
        event_type='Order']) AS level
from action3
group by uid
order by uid;
+------+-------+
| uid  | level |
+------+-------+
|    1 |     3 |
|    2 |     3 |
+------+-------+

3 is returned for uid = 1 and uid = 2.

Change mode to 4 and execute the statement again.

select uid,
       window_funnel(1900,time,4,[event_type='Browse', event_type='Click',
        event_type='Order']) AS level
from action3
group by uid
order by uid;
+------+-------+
| uid  | level |
+------+-------+
|    1 |     3 |
|    2 |     1 |
+------+-------+
1 row in set (0.02 sec)

1 is returned for uid = 2 because mode 4 (strictly increasing) is used. "Click" happens at the same second as "BROWSE". Therefore, "Click" and "Order" are not counted.

Keywords

window funnel, funnel, window_funnel

Edit

Feedback

In this article

window_funnel