2019-09-04工作进展

昨天工作：

1. mv-dssm效果改善：mv-dssm重新设计商品图片和title信息融合获取融合特征的attention结构，当前新的网络的训练效果：acc = 0.767 auc = 0.767 precision = 0.859。
2. inference数据生成：图片特征提取完成。今天可以得到inference结果。

今天计划：

1. 对mv-dssm网络进行inference并进行结果分析。

2. 改善mv-dssm网络效果。

3. inference数据

drop table hs_tmp_363;
yes
create table graph_embedding.hs_tmp_363 as
select title, graph_embedding:return_concat_pic_feature_4(pic_ws) as pic_emb from graph_embedding.hs_tmp_pic_inf_1 group by title;

create table graph_embedding.hs_tmp_362 as select b.item_id, a.title, a.pic_emb from graph_embedding.hs_tmp_363 a join graph_embedding.hs_dssm_dic_title_inf_10 b on a.title=b.title;

create table hs_tmp_364 as select a.*, b.pic_emb as pic_ws from hs_tmp_267 a join hs_tmp_362 b on a.video_id = b.item_id;

2. inference

pai -name tensorflow140 -Dscript="file:///home/hengsong/origin_deep_cluster_odps_8.tar.gz" -DentryFile="inference_mv_dssm_v3.py" -Dcluster='{"worker":{"count":1, "cpu":200, "memory":4000}, "ps":{"count":1, "cpu":200, "memory":5000}}' -Dtables="odps://graph_embedding/tables/hs_train_data_dssm_v2_8,odps://graph_embedding/tables/hs_test_data_dssm_v2_8,odps://graph_embedding/tables/hs_tmp_364" -Doutputs="odps://graph_embedding/tables/hs_dssm_result_5" -DcheckpointDir="oss://bucket-automl/hengsong/?role_arn=acs:ram::1293303983251548:role/graph2018&host=cn-hangzhou.oss-internal.aliyun-inc.com" -DuserDefinedParameters="--learning_rate=3e-4 --batch_size=1024 --is_save_model=True --attention_type=1 --num_epochs=1 --ckpt=hs_ugc_video_2e_4.ckpt-2" -DuseSparseClusterSchema=True;

3. train

pai -name tensorflow140 -Dscript="file:///home/hengsong/origin_deep_cluster_odps_8.tar.gz" -DentryFile="train_mv_dssm_v3.py" -Dcluster='{"worker":{"count":50, "cpu":200, "memory":4000}, "ps":{"count":10, "cpu":200, "memory":5000}}' -Dtables="odps://graph_embedding/tables/hs_train_data_dssm_v2_8,odps://graph_embedding/tables/hs_test_data_dssm_v2_8,odps://graph_embedding/tables/hs_tmp_364" -Doutputs="odps://graph_embedding/tables/hs_dssm_result_5" -DcheckpointDir="oss://bucket-automl/hengsong/?role_arn=acs:ram::1293303983251548:role/graph2018&host=cn-hangzhou.oss-internal.aliyun-inc.com" -DuserDefinedParameters="--learning_rate=3e-4 --batch_size=1024 --is_save_model=True --attention_type=1 --num_epochs=1 --ckpt=hs_ugc_video_2e_4.ckpt" -DuseSparseClusterSchema=True;

graph_embedding : 20190904043433512ghjd9x69

3. 添加multi_inf_attention测试

pai -name tensorflow140 -Dscript="file:///home/hengsong/origin_deep_cluster_odps_8.tar.gz" -DentryFile="train_mv_dssm_v4.py" -Dcluster='{"worker":{"count":50, "cpu":200, "memory":4000}, "ps":{"count":10, "cpu":200, "memory":5000}}' -Dtables="odps://graph_embedding/tables/hs_train_data_dssm_v2_8,odps://graph_embedding/tables/hs_test_data_dssm_v2_8,odps://graph_embedding/tables/hs_tmp_364" -Doutputs="odps://graph_embedding/tables/hs_dssm_result_5" -DcheckpointDir="oss://bucket-automl/hengsong/?role_arn=acs:ram::1293303983251548:role/graph2018&host=cn-hangzhou.oss-internal.aliyun-inc.com" -DuserDefinedParameters="--learning_rate=3e-4 --batch_size=1024 --is_save_model=True --attention_type=1 --num_epochs=1 --ckpt=hs_ugc_video_2e_4.ckpt" -DuseSparseClusterSchema=True;