feat: Refactor predictor

.git_bin_path

@@ -1,7 +1,7 @@
 {"leaf_name": "data/test", "leaf_file": ["data/test/batch_criteo_sample.tfrecord", "data/test/criteo_sample.tfrecord", "data/test/dwd_avazu_ctr_deepmodel_10w.csv", "data/test/embed_data.csv", "data/test/lookup_data.csv", "data/test/tag_kv_data.csv", "data/test/test.csv", "data/test/test_sample_weight.txt", "data/test/test_with_quote.csv"]}
 {"leaf_name": "data/test/export", "leaf_file": ["data/test/export/data.csv"]}
 {"leaf_name": "data/test/hpo_test/eval_val", "leaf_file": ["data/test/hpo_test/eval_val/events.out.tfevents.1597889819.j63d04245.sqa.eu95"]}
-{"leaf_name": "data/test/inference", "leaf_file": ["data/test/inference/lookup_data_test80.csv", "data/test/inference/taobao_infer_data.txt"]}
+{"leaf_name": "data/test/inference", "leaf_file": ["data/test/inference/lookup_data_test80.csv", "data/test/inference/taobao_infer_data.txt", "data/test/inference/taobao_infer_rtp_data.txt"]}
 {"leaf_name": "data/test/inference/fg_export_multi", "leaf_file": ["data/test/inference/fg_export_multi/saved_model.pb"]}
 {"leaf_name": "data/test/inference/fg_export_multi/assets", "leaf_file": ["data/test/inference/fg_export_multi/assets/pipeline.config"]}
 {"leaf_name": "data/test/inference/fg_export_multi/variables", "leaf_file": ["data/test/inference/fg_export_multi/variables/variables.data-00000-of-00001", "data/test/inference/fg_export_multi/variables/variables.index"]}
@@ -20,6 +20,9 @@
 {"leaf_name": "data/test/inference/tb_multitower_placeholder_rename_export", "leaf_file": ["data/test/inference/tb_multitower_placeholder_rename_export/saved_model.pb"]}
 {"leaf_name": "data/test/inference/tb_multitower_placeholder_rename_export/assets", "leaf_file": ["data/test/inference/tb_multitower_placeholder_rename_export/assets/pipeline.config"]}
 {"leaf_name": "data/test/inference/tb_multitower_placeholder_rename_export/variables", "leaf_file": ["data/test/inference/tb_multitower_placeholder_rename_export/variables/variables.data-00000-of-00001", "data/test/inference/tb_multitower_placeholder_rename_export/variables/variables.index"]}
+{"leaf_name": "data/test/inference/tb_multitower_rtp_export", "leaf_file": ["data/test/inference/tb_multitower_rtp_export/saved_model.pb"]}
+{"leaf_name": "data/test/inference/tb_multitower_rtp_export/assets", "leaf_file": ["data/test/inference/tb_multitower_rtp_export/assets/pipeline.config"]}
+{"leaf_name": "data/test/inference/tb_multitower_rtp_export/variables", "leaf_file": ["data/test/inference/tb_multitower_rtp_export/variables/variables.data-00000-of-00001", "data/test/inference/tb_multitower_rtp_export/variables/variables.index"]}
 {"leaf_name": "data/test/latest_ckpt_test", "leaf_file": ["data/test/latest_ckpt_test/model.ckpt-500.data-00000-of-00001", "data/test/latest_ckpt_test/model.ckpt-500.index", "data/test/latest_ckpt_test/model.ckpt-500.meta"]}
 {"leaf_name": "data/test/rtp", "leaf_file": ["data/test/rtp/taobao_fg_pred.out", "data/test/rtp/taobao_test_bucketize_feature.txt", "data/test/rtp/taobao_test_feature.txt", "data/test/rtp/taobao_test_input.txt", "data/test/rtp/taobao_train_bucketize_feature.txt", "data/test/rtp/taobao_train_feature.txt", "data/test/rtp/taobao_train_input.txt", "data/test/rtp/taobao_valid.csv", "data/test/rtp/taobao_valid_feature.txt"]}
 {"leaf_name": "data/test/tb_data", "leaf_file": ["data/test/tb_data/taobao_ad_feature_gl", "data/test/tb_data/taobao_clk_edge_gl", "data/test/tb_data/taobao_multi_seq_test_data", "data/test/tb_data/taobao_multi_seq_train_data", "data/test/tb_data/taobao_noclk_edge_gl", "data/test/tb_data/taobao_test_data", "data/test/tb_data/taobao_test_data_for_expr", "data/test/tb_data/taobao_test_data_kd", "data/test/tb_data/taobao_train_data", "data/test/tb_data/taobao_train_data_for_expr", "data/test/tb_data/taobao_train_data_kd", "data/test/tb_data/taobao_user_profile_gl"]}

.git_bin_url

@@ -1,7 +1,7 @@
 {"leaf_path": "data/test", "sig": "656d73b4e78d0d71e98120050bc51387", "remote_path": "data/git_oss_sample_data/data_test_656d73b4e78d0d71e98120050bc51387"}
 {"leaf_path": "data/test/export", "sig": "c2e5ad1e91edb55b215ea108b9f14537", "remote_path": "data/git_oss_sample_data/data_test_export_c2e5ad1e91edb55b215ea108b9f14537"}
 {"leaf_path": "data/test/hpo_test/eval_val", "sig": "fef5f6cd659c35b713c1b8bcb97c698f", "remote_path": "data/git_oss_sample_data/data_test_hpo_test_eval_val_fef5f6cd659c35b713c1b8bcb97c698f"}
-{"leaf_path": "data/test/inference", "sig": "e2c4b0f07ff8568eb7b8e1819326d296", "remote_path": "data/git_oss_sample_data/data_test_inference_e2c4b0f07ff8568eb7b8e1819326d296"}
+{"leaf_path": "data/test/inference", "sig": "9725274cad0f27baf561ebfaf7946846", "remote_path": "data/git_oss_sample_data/data_test_inference_9725274cad0f27baf561ebfaf7946846"}
 {"leaf_path": "data/test/inference/fg_export_multi", "sig": "c6690cef053aed9e2011bbef90ef33e7", "remote_path": "data/git_oss_sample_data/data_test_inference_fg_export_multi_c6690cef053aed9e2011bbef90ef33e7"}
 {"leaf_path": "data/test/inference/fg_export_multi/assets", "sig": "7fe7a4525f5d46cc763172f5200e96e0", "remote_path": "data/git_oss_sample_data/data_test_inference_fg_export_multi_assets_7fe7a4525f5d46cc763172f5200e96e0"}
 {"leaf_path": "data/test/inference/fg_export_multi/variables", "sig": "1f9aad9744382c6d5b5f152d556d9b30", "remote_path": "data/git_oss_sample_data/data_test_inference_fg_export_multi_variables_1f9aad9744382c6d5b5f152d556d9b30"}
@@ -20,6 +20,9 @@
 {"leaf_path": "data/test/inference/tb_multitower_placeholder_rename_export", "sig": "dc05357e52fd574cba48165bc67af906", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_placeholder_rename_export_dc05357e52fd574cba48165bc67af906"}
 {"leaf_path": "data/test/inference/tb_multitower_placeholder_rename_export/assets", "sig": "750925c4866bf1db8c3188f604271c72", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_placeholder_rename_export_assets_750925c4866bf1db8c3188f604271c72"}
 {"leaf_path": "data/test/inference/tb_multitower_placeholder_rename_export/variables", "sig": "56850b4506014ce1bd3ba9b6d60e2770", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_placeholder_rename_export_variables_56850b4506014ce1bd3ba9b6d60e2770"}
+{"leaf_path": "data/test/inference/tb_multitower_rtp_export", "sig": "f1bc6238cfab648812afca093da5dd6b", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_rtp_export_f1bc6238cfab648812afca093da5dd6b"}
+{"leaf_path": "data/test/inference/tb_multitower_rtp_export/assets", "sig": "ae1cc9ec956fb900e5df45c4ec255c4b", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_rtp_export_assets_ae1cc9ec956fb900e5df45c4ec255c4b"}
+{"leaf_path": "data/test/inference/tb_multitower_rtp_export/variables", "sig": "efe52ef308fd6452f3b67fd04cdd22bd", "remote_path": "data/git_oss_sample_data/data_test_inference_tb_multitower_rtp_export_variables_efe52ef308fd6452f3b67fd04cdd22bd"}
 {"leaf_path": "data/test/latest_ckpt_test", "sig": "d41d8cd98f00b204e9800998ecf8427e", "remote_path": "data/git_oss_sample_data/data_test_latest_ckpt_test_d41d8cd98f00b204e9800998ecf8427e"}
 {"leaf_path": "data/test/rtp", "sig": "76cda60582617ddbb7cd5a49eb68a4b9", "remote_path": "data/git_oss_sample_data/data_test_rtp_76cda60582617ddbb7cd5a49eb68a4b9"}
 {"leaf_path": "data/test/tb_data", "sig": "c8136915b6e5e9d96b18448cf2e21d3d", "remote_path": "data/git_oss_sample_data/data_test_tb_data_c8136915b6e5e9d96b18448cf2e21d3d"}

docs/source/models/mind.md

@@ -85,7 +85,7 @@ model_config:{
       # use the same numer of capsules for all users
       const_caps_num: true
     }
-    
+
     simi_pow: 20
     l2_regularization: 1e-6
     time_id_fea: "seq_ts_gap"
@@ -101,7 +101,7 @@ model_config:{
   - dnn:
     - hidden_units: dnn每一层的channel数
     - use_bn: 是否使用batch_norm, 默认是true
-- item_dnn: item侧的dnn参数, 配置同user_dnn 
+- item_dnn: item侧的dnn参数, 配置同user_dnn
   - note: item侧不能用batch_norm
 - pre_capsule_dnn: 进入capsule之前的dnn的配置
   - 可选, 配置同user_dnn和item_dnn
@@ -117,7 +117,7 @@ model_config:{
   - squash_pow: 对squash加的power, 防止squash之后的向量值变得太小
 - simi_pow: 对相似度做的倍数, 放大interests之间的差异
 - embedding_regularization: 对embedding部分加regularization，防止overfit
-- user_seq_combine: 
+- user_seq_combine:
   - CONCAT: 多个seq之间采取concat的方式融合
   - SUM: 多个seq之间采取sum的方式融合, default是SUM
 - time_id_fea: time_id feature的name, 对应feature_config里面定义的特征
@@ -128,6 +128,7 @@ model_config:{
 - 行为序列特征可以加上time_id, time_id经过1 dimension的embedding后, 在time维度进行softmax, 然后和其它sequence feature的embedding相乘
 
 - time_id取值的方式可参考:
+
   - 训练数据:  Math.round((2 * Math.log1p((labelTime - itemTime) / 60.) / Math.log(2.))) + 1
   - inference: Math.round((2 * Math.log1p((currentTime - itemTime) / 60.) / Math.log(2.))) + 1
   - 此处的时间(labelTime, itemTime, currentTime) 为seconds
@@ -136,17 +137,19 @@ model_config:{
 
 - 使用增量训练，增量训练可以防止负采样的穿越。
 
-- 使用HPO对squash_pow[0.1 - 1.0]和simi_pow[10 - 100]进行搜索调优。
+- 使用HPO对squash_pow\[0.1 - 1.0\]和simi_pow\[10 - 100\]进行搜索调优。
 
 - 要看的指标是召回率，准确率和兴趣损失，三个指标要一起看。
 
 - 使用全网的点击数据来生成训练样本，全网的行为会更加丰富，这有利于mind模型的训练。
 
 - 数据清洗:
+
   - 把那些行为太少的item直接在构造行为序列的时候就挖掉
   - 排除爬虫或者作弊用户
 
 - 数据采样:
+
   - mind模型的训练默认是以点击为目标
   - 如果业务指标是到交易，那么可以对交易的样本重采样
 
@@ -155,9 +158,11 @@ model_config:{
 [MIND_demo.config](https://easyrec.oss-cn-beijing.aliyuncs.com/config/mind_on_taobao_neg_sam.config)
 
 ### 效果评估
+
 离线的效果评估主要看在测试集上的hitrate. 可以参考文档[效果评估](https://easyrec.oss-cn-beijing.aliyuncs.com/docs/recall_eval.pdf)
 
 #### 评估sql
+
 ```sql
 pai -name tensorflow1120_cpu_ext
  -Dscript='oss://easyrec/deploy/easy_rec/python/tools/hitrate.py'
@@ -204,15 +209,18 @@ pai -name tensorflow1120_cpu_ext
   - 1: Inner Product similarity
 - emb_dim: user / item表征向量的维度
 - top_k: knn检索取top_k计算hitrate
-- recall_type: 
+- recall_type:
   - u2i: user to item retrieval
 
 #### 评估结果
+
 输出下面两张表
 
 - mind_hitrate_details:
+
   - 输出每一个user的hitrate = user_hits / user_recalls
   - 格式如下:
+
   ```text
      id         :  bigint
      topk_ids   :  string
@@ -221,10 +229,12 @@ pai -name tensorflow1120_cpu_ext
      bad_ids    :  string
      bad_dists  :  string
   ```
- 
+
 - mind_total_hitrate:
+
   - 输出平均hitrate = SUM(user_hits) / SUM(user_recalls)
   - 格式如下:
+
   ```text
      hitrate    :  double
   ```

docs/source/pre_check.md

@@ -3,30 +3,32 @@
 为解决用户常由于脏数据或配置错误的原因，导致训练失败，开发了预检查功能。
 在训练时打开检查模式，或是训练前执行pre_check脚本，即会检查data_config配置及train_config部分配置，筛查全部数据，遇到异常则抛出相关信息，并给出修改意见。
 
-
 ### 命令
 
 #### Local
 
 方式一: 执行pre_check脚本：
+
 ```bash
 PYTHONPATH=. python easy_rec/python/tools/pre_check.py --pipeline_config_path samples/model_config/din_on_taobao.config --data_input_path data/test/check_data/csv_data_for_check
 ```
 
 方式二: 训练时打开检查模式（默认关闭）：
 
 该方式会影响训练速度，线上例行训练时不建议开启检查模式。
+
 ```bash
 python -m easy_rec.python.train_eval --pipeline_config_path samples/model_config/din_on_taobao.config --check_mode
 ```
+
 - pipeline_config_path config文件路径
 - data_input_path 待检查的数据路径，不指定的话为pipeline_config_path中的train_input_path及eval_input_path
 - check_mode 默认False
 
-
 #### On PAI
 
 方式一: 执行pre_check脚本：
+
 ```sql
 pai -name easy_rec_ext -project algo_public
   -Dcmd='check'
@@ -42,6 +44,7 @@ pai -name easy_rec_ext -project algo_public
 方式二: 训练时打开检查模式（默认关闭）：
 
 该方式会影响训练速度，线上例行训练时不建议开启检查模式。
+
 ```sql
 pai -name easy_rec_ext  -project algo_public
   -Dcmd='train'