File tree Expand file tree Collapse file tree 2 files changed +54
-0
lines changed
Expand file tree Collapse file tree 2 files changed +54
-0
lines changed Original file line number Diff line number Diff line change @@ -1294,6 +1294,23 @@ message InputLayer {
12941294- ` wide_output_dim ` wide模型每个特征的参数权重维度,一般设定为1
12951295- ` concat_seq_feature ` 是否需要把序列特征的embedding拼接在一起
12961296
1297+ 比如同一个group的所有特征 pooling 为一个特征的示例如下:
1298+
1299+ - 前提条件:这个group内的所有特征的在embedding_dim都相同
1300+
1301+ ``` protobuf
1302+ blocks {
1303+ name: 'feat_pooling'
1304+ inputs {
1305+ feature_group_name: 'feat_group'
1306+ }
1307+ input_layer {
1308+ only_output_3d_tensor: true
1309+ }
1310+ extra_input_fn: 'lambda x: tf.reduce_sum(x, axis=1)'
1311+ }
1312+ ```
1313+
12971314## 3. Lambda组件块
12981315
12991316` Lambda组件块 ` 可以配置一个lambda函数,执行一些较简单的操作。示例如下:
Original file line number Diff line number Diff line change @@ -73,6 +73,43 @@ blocks {
7373}
7474```
7575
76+ ## 对输入序列pooling的例子
77+
78+ ``` protobuf
79+ model_config: {
80+ model_name: 'SumPooling'
81+ model_class: 'RankModel
82+ ...
83+ feature_groups: {
84+ group_name: 'sequence'
85+ feature_names: "tag_category_list"
86+ feature_names: "tag_brand_list"
87+ wide_deep: DEEP
88+ }
89+ backbone {
90+
91+ blocks {
92+ name: 'seq_input'
93+ inputs {
94+ feature_group_name: 'sequence'
95+ }
96+ input_layer {
97+ output_seq_and_normal_feature: true
98+ }
99+ }
100+ blocks {
101+ name: 'sum_pooling'
102+ inputs {
103+ block_name: 'seq_input'
104+ input_slice: '[0]'
105+ }
106+ extra_input_fn: 'lambda x: tf.reduce_sum(x, axis=1)'
107+ }
108+ ...
109+ }
110+ }
111+ ```
112+
76113## 完整的例子
77114
78115- [ DIN] ( ../models/din.md )
You can’t perform that action at this time.
0 commit comments