[doc]refactor documents #417
Conversation
|
|
|
Code Style Test Passed |
| | need_prefix | 可选项，true表示会拼上feature_name作为前缀，false表示不拼，默认为true，通常在shared_embedding的场景会用false | | ||
|
|
||
| ## 示例: | ||
|
|
There was a problem hiding this comment.
下面以item侧的特征is_main作为案例来说明在不同类型下的原始值和输出值：
| - needDiscrete=true，结果为: brand_hit_50006842_30068_19 | ||
| - needDiscrete=false，结果为: 19.0 | ||
|
|
||
| 如果只使用一层匹配，则需要将上面配置里的 category 的值改为 ALL。这种情况，用户也可以考虑使用lookup_feature。 假设各字段的值如下 |
There was a problem hiding this comment.
如果只使用一层匹配，则需要将上面配置里的 category 的值改为 ALL，即"category": "ALL”。这种情况，用户也可以考虑使用lookup_feature来实现，用lookup_feature的时候user_brand_tags_hit 里面的值的格式更简单，即”16788816:40,10122:40,29889:20,30068:20”。 假设两个字段的值如下：
| @@ -17,13 +17,13 @@ | |||
| "expression":"user:feat0" | |||
There was a problem hiding this comment.
上面序列长度改成50，正常item维度应该改成”user维度“？
例如我们需要对⽤户的点击序列进⾏ fg，序列⻓度为 50，每个序列提取 nid 和 price, seq_context 特征。正常 user 维度有⼀个 feat0 特征。配置如下：
| user_features { | ||
| key: "click__ts" | ||
| value { | ||
| string_feature: "23;113;401363;401369;401375;401405;486678;486803;486922;486969" |
There was a problem hiding this comment.
string_feature: "1694592001;1694592002;1694592003;1694592004;1694592005;1694592006;1694592006;1694592007;1694592007;1694592008”
| | 字段名 | 含义 | | ||
| | --------------- | -------------------------------- | | ||
| | feature_name | 必选项, 特征名 | | ||
| | expression | 必选项，expression描述该feature所依赖的字段来源 | |
There was a problem hiding this comment.
| expression | 必选项，expression描述该feature所依赖的字段来源，来源必须是user、item、context中的一种 |
|
already |
|
Code Style Test Passed |
github-actions
bot
added
code_style_test_passed
and removed
code_style_test_passed
labels
|
Code Style Test Passed |
|
Code Style Test Passed |
refactor fg documents