firebase - 在实际流失发生后记录 Firebase 流失预测

Question

我正在尝试查看 Firebase 预测的准确度。我已将 firebase 预测和分析事件导出到 BigQuery。为了测试预测的性能，我编写了一个查询来收集churn事件的预测，然后将其与app_remove分析中的事件结合起来。问题是对于很多很多用户来说，app_remove事件是在预测日期之前记录的！好像 google 等待用户流失，然后将流失的预测放在表中。我错过了什么吗？

这是我的代码：

SELECT
    b.day AS prediction_day,
    b.user_id,
    c.day AS churn_day
FROM
(
    SELECT 
        MIN(day) AS day,
        user_id
    FROM
    (
        SELECT
            CAST(u.prediction_time AS DATE) AS day,
            u.user_id,
            p.id,
            p.probability,
            p.observed_label
        FROM
            `Predictions_table` AS u,
            u.predictions AS p
    )a
    WHERE
        a.probability > 0.2
        AND a.id = 'churn'
        AND NOT (a.user_id IS NULL OR a.user_id = '')
    GROUP BY
        user_id
)b
LEFT JOIN
(
    SELECT
        MAX(day) AS day,
        user_id
    FROM
    (
        SELECT 
            CAST(TIMESTAMP_MICROS(event_timestamp) AS DATE) AS day,
            user_id
        FROM
            `Analytics_table`
        WHERE
            event_name = 'app_remove'
            AND NOT (user_id IS NULL OR user_id = '')
    )e
    GROUP BY
        user_id

)c
ON b.user_id = c.user_id

标签： firebasegoogle-cloud-platformgoogle-bigqueryfirebase-predictions