Platform For AI:Registro de datos y configuración de campos

Data registration

Para registrar los datos, active MaxCompute, vincule un proyecto y registre una tabla de MaxCompute.. This registered table serves as an intermediate table for performance statistics, called the experiment report source table. It uses a composite primary key of user_id, exp_id, and dimension fields to aggregate statistics. Each row represents metrics such as exposures and clicks for a specific user in a specific experiment. You can add dimensions such as country to build a more granular intermediate statistics table keyed on user_id, country, and exp_id. All basic and derived metrics depend on this intermediate table. The overall experiment architecture is as follows:

PAI-Rec engine is a recommendation service that integrates an SDK from the configuration center to retrieve experiment information. The response includes the experiment ID (exp_id) and a request ID (reqid).

When a user interacts with your application, front-end event tracking collects behavior logs and sends them to MaxCompute. An SQL task processes these logs to generate the experiment report source table. After you configure metrics in the configuration center, the system dynamically assembles SQL queries to compute statistics from this source table. The results are stored in Hologres, and you can view them on the experiment report page.

I. Offline experiment reports

Offline experiment report source table code

CREATE TABLE IF NOT EXISTS experiment_report (
	user_id STRING COMMENT 'User ID',
        exp_id STRING COMMENT 'Experiment ID',
	exposure_count BIGINT COMMENT 'Exposure count',
	click_count BIGINT COMMENT 'Click count',
	like_comment_collect_count BIGINT COMMENT 'Like, comment, and collect count'
)
PARTITIONED  by (dt STRING COMMENT 'Date')
STORED AS ALIORC;
INSERT OVERWRITE TABLE experiment_report PARTITION (dt = '${bdp.system.bizdate}')
SELECT  user_id
        ,exp_id
        ,SUM(IF(event == 'expose',1,0))
        ,SUM(IF(event == 'click',1,0))
        ,SUM(IF(event IN ('like','comment','collect'),1,0))
FROM    rec_sln_demo_behavior_table_v1
WHERE   ds = '${bdp.system.bizdate}'
GROUP BY user_id
         ,exp_id

Offline experiment report source table schema

You can generate the source table daily for offline reports or hourly for near-real-time reports, depending on your business needs. For example, use DataWorks to schedule a job that runs every hour to process logs from the previous hour. For offline generation, user_id, exp_id, and dt are required. For hourly reports, user_id, exp_id, dt, and hh are required; the mm partition is optional.

Dimension fields are optional but enable dimensional analysis of your metrics. Common dimension fields include gender and os (operating system, such as iOS or Android), which let you view report data segmented by gender or operating system.

Código de la tabla de origen del reporte de significancia sin conexión

CREATE TABLE IF NOT EXISTS experiment_report_v3 (
    user_id STRING COMMENT 'User ID',
    exp_id STRING COMMENT 'Experiment ID',
    item_id STRING COMMENT 'Item ID',
    request_id STRING COMMENT 'Request ID',
    click_uv BIGINT COMMENT 'Click UV',
    expr_pv BIGINT COMMENT 'Exposure PV'
)
PARTITIONED  by (dt STRING COMMENT 'Date')
STORED AS ALIORC;
INSERT OVERWRITE TABLE experiment_report_v3 PARTITION (dt = '${bdp.system.bizdate}')
SELECT
    user_id,
    exp_id,
    item_id,
    request_id,
    MAX(IF(event = 'click', 1, 0)) AS click_uv,
    SUM(IF(event = 'expr', 1, 0)) AS expr_pv
FROM rec_sln_demo_behavior_table_v1
WHERE   ds = '${bdp.system.bizdate}'
GROUP BY
    user_id,
    exp_id,
    item_id,
    request_id

Offline significance report source table schema

II. Real-time experiment reports

Real-time experiment report source table code

CREATE TABLE IF NOT EXISTS experiment_report_real (
	user_id STRING COMMENT 'User ID',
        exp_id STRING COMMENT 'Experiment ID',
	exposure_count BIGINT COMMENT 'Exposure count',
	click_count BIGINT COMMENT 'Click count',
	like_comment_collect_count BIGINT COMMENT 'Like, comment, and collect count'
)
PARTITIONED  by (
    dt string
    ,hh string
)
STORED AS ALIORC;
INSERT OVERWRITE TABLE experiment_report_real PARTITION (dt = '${bdp.system.bizdate}',hh = '${hour}')
SELECT  user_id
        ,exp_id
        ,SUM(IF(event == 'expose',1,0))
        ,SUM(IF(event == 'click',1,0))
        ,SUM(IF(event IN ('like','comment','collect'),1,0))
FROM    rec_sln_demo_behavior_table_v1
WHERE   ds = '${bdp.system.bizdate}'
AND hh = hour(now()) -1
GROUP BY user_id
         ,exp_id;

Real-time experiment report source table schema

Offline metrics and real-time metrics cannot use the same result table. A result table for offline metrics requires only the dt date partition, whereas a real-time table must include at least the hh hour partition.

After you create the experiment report source table, go to Metric Management > Data Registration in the console to add the new MaxCompute data table.

III. Register a MaxCompute table

Click Create Data Table, select your bound MaxCompute project, choose the MaxCompute data table, provide a custom name, and then click Start Import to import the table data.

Field configuration

After registration, the data table appears in the list. Click View Fields to review and edit the fields. The user_id, exp_id, and dt fields are required. Together, these fields track behavior totals such as exposures, clicks, and likes for a specific user (user_id) in a specific experiment (exp_id) on a given day (dt). Custom metrics are defined and calculated based on this table.

Dimension fields

Puede agregar uno o dos campos de dimensión a la tabla. Por ejemplo, si establece city como dimensión, el reporte del experimento muestra una comparación de los resultados por cada ciudad. Si configura tanto city como gender como campos de dimensión, el reporte presenta los resultados agrupados según la combinación de ambos campos.