頻繁模式挖掘函數通過分析多維資料,提取顯著差異的屬性群組合并量化其影響,支援多種參數配置以最佳化挖掘結果。
get_patterns
get_patterns 是頻繁集挖掘的運算元,除了挖掘頻繁項之外,get_patterns還會對挖掘到的頻繁項做合并和去重。專指提取表格式資料的模板(頻繁集)。
文法
get_patterns($TABLE, $HEADER, $PARAM)參數說明
參數 | 資料類型 | 是否必選 | 說明 |
$TABLE | row<array<T>, array<E>, ..., array<F>> | 是 | 待挖掘頻繁項的輸入資料的表格,每一列是一個待挖掘的維度列。 |
$HEADER | array<varchar> | 是 | 列名,和$TABLE對應,header名字數和table列的數量一致。 |
$PARAM | varchar | 否 | 參見param參數說明。 |
param參數說明
參數名 | 參數解釋 | 參數類型 | 是否必填 | 預設值 | 取值範圍 |
minimum_support_fraction | 輸出的pattern在測試組的最低的支援度。比如某個pattern在測試組出現的頻率是0.1,那麼我們說這個pattern的支援度是0.1。這個參數可以通過控制pattern的敏感度控制pattern的數量。 | double | 否 | 0.05 | (0, 1) |
樣本
查詢分析:
關於"set session enable_remote_functions=true ", 目前功能處於公測階段,需手動添加該 flag。後續版本將移除此要求,實現自動化支援。
(*)| set session enable_remote_functions=true ; with t0 as (select JSON_EXTRACT_SCALAR(entity, '$.platform') AS platform, JSON_EXTRACT_SCALAR(entity, '$.region') AS region, cast(value as double) as value, if((value > 100), 'true', 'false') as anomaly_label from log), t1 as ( select array_agg(platform) as platform, array_agg(region) as region, array_agg(anomaly_label) as anomaly_label, array_agg(value) as value from t0), t2 as (select row(platform, region) as table_row from t1), t3 as (select get_patterns(table_row, ARRAY['platform', 'region']) as ret from t2) select * from t3輸出結果:
[["platform=eBay","platform=edX","platform=Amazon","platform=Skillshare","platform=Shopify","platform=Khan Academy","platform=Coursera","platform=Udemy","platform=Alibaba","platform=Taobao","platform=Snapchat","platform=Amazon Prime","platform=YouTube","platform=Hulu","platform=Peloton","platform=Twitter","platform=Fitbit","platform=Nike Training","platform=LinkedIn","platform=Instagram","platform=Disney+","platform=Strava","platform=MyFitnessPal","platform=Facebook","platform=Netflix","platform=Console","platform=Samsung SmartThings","platform=Apple HomeKit","platform=Mobile","platform=PC","platform=Google Home","platform=VR"],[156960,149760,148320,148320,146880,145440,139680,136800,133920,133920,96480,95040,92160,90720,90720,89280,89280,87840,84960,83520,83520,82080,82080,77760,70560,46080,41760,41760,34560,33120,31680,30240],null,null]
返回結果說明
參數 | 類型 | 說明 | 對應樣本 |
$RET.patterns | array<varchar> | 表格模板,頻繁集。每個varchar是一個用 AND 串連的運算式,類似 |
|
$RET.test_supports | array<bigint> | 每一個模板出現的頻次。 |
|
$RET.labels | array<bigint> | 預留的傳回值位置,便於後續自動將資料歸類。目前始終為 |
|
$RET.error_msg | array<varchar>/null | 報錯資訊。如果沒有報錯,則傳回值為 |
|