Kinsight
kim用户群 doc用户文档
基础信息
任务名称: exp8_1125-2025-11-25 20:20:49
负责人: 陈子轩
创建时间: 2026-04-02 10:39
APP对象: 主站、极速版
平台: iOS、安卓
样本量: 10000000
模型目标: 未来2分钟的是否点赞
目标类型: 分类
实时特征:
i 展开查看 (5)
user_active_days play_ratio_1h like_cnt_10m comment_cnt_10m follow_cnt_10m
离线特征:
i 展开查看 (3)
user_age user_gender device_type
模型:
i 展开查看 (6)
NN_small
NN_large
Dragonnet
TLearner_xgboost
XLearner
CausalForest
任务执行情况
预处理
开始时间: 2026-04-02 10:39:29
结束时间: 2026-04-02 12:04:00
总计耗时: 1小时24分钟
»
训练
开始时间: 2026-04-02 12:08:01
结束时间: 2026-04-02 14:26:00
任务总数: 12
成功: 12 失败: 0 运行中: 0 等待中: 0
总计耗时: 2小时17分钟
»
评估
开始时间: 2026-04-02 12:32:01
结束时间: 2026-04-02 15:08:00
任务总数: 12
成功: 12 失败: 0 运行中: 0 等待中: 0
总计耗时: 2小时35分钟
»
打包
总计耗时: -
子任务详情
特征探查报告
评估报告
打包产物
查看日志
序号 子任务名称 状态 任务耗时 操作
1 总消费vv数-CausalForest 预处理 训练 评估 5小时9分钟 报告重试任务
2 APP使用时长-CausalForest 预处理训练评估 4小时29分钟 报告重试任务
3 活跃天数/LT-CausalForest 预处理训练评估 4小时17分钟 报告重试任务
4 总消费vv数-XLearner 预处理训练评估 5小时9分钟 报告重试任务
5 APP使用时长-XLearner 预处理训练评估 3小时31分钟 报告重试任务
6 活跃天数/LT-XLearner 预处理训练评估 3小时35分钟 报告重试任务
7 总消费vv数-TLearner_xgboost 预处理训练评估 5小时9分钟 报告重试任务
8 APP使用时长-TLearner_xgboost 预处理训练评估 2小时29分钟 报告重试任务
9 活跃天数/LT-TLearner_xgboost 预处理训练评估 2小时35分钟 报告重试任务
10 总消费vv数-Dragonnet 预处理训练评估 5小时9分钟 报告重试任务
1 2 共 12 条
数据量
9,960,625
条记录
特征数量
163
个特征
风险特征数
8
i 展开查看
fre_country_region(高偏态)
device_active_degree(异常尖峰)
is_lowactive_period(类别倾斜)
gender(采样偏差)
age_range(长尾离散)
fre_province(低频噪声)
network_type(缺失偏高)
app_open_cnt(离群值)
device_active_degree
is_lowactive_period
gender
age_range
fre_country_region
score_prob
play_duration
consume_depth
[score_prob] 特征分布
模型:
关键指标
问题类型 i 该选项仅用于前端开发区分分类/回归两类展示逻辑,线上实际页面不展示此控件。 /
模型 准确率 召回率 F1 score CrossEntropy Loss 混淆矩阵 AUC(分数/曲线) loss曲线 特征重要度 特征探查详情
Dragonnet0.850.900.431.24查看0.87 / 曲线查看详情
TLearner_xgboost0.840.890.421.29查看0.85 / 曲线查看详情
XLearner0.830.880.411.33查看0.84 / 曲线查看详情
CausalForest0.820.860.401.37查看0.82 / 曲线查看详情
NN_small0.860.910.441.18查看0.89 / 曲线查看查看详情
NN_large0.850.900.431.21查看0.88 / 曲线查看查看详情
模型 MAE MSE RMSE MAPE R2(决定系数) loss曲线 特征重要度 特征探查详情
Dragonnet0.2180.0970.3119.8%0.76查看详情
TLearner_xgboost0.2260.1030.32110.3%0.74查看详情
XLearner0.2310.1080.32910.8%0.73查看详情
CausalForest0.2390.1130.33611.4%0.71查看详情
NN_small0.2070.0910.3019.1%0.79查看查看详情
NN_large0.2140.0940.3069.4%0.78查看查看详情
loss 曲线
特征重要度报告
Features
00.10.20.30.40.50.6
Feature Importance Score
模型 打包时间 状态 负责人 操作
Dragonnet 2026-04-07 13:21 已打包 陈子轩 产物 proj_1024/task_8899/model_dragonnet
NN_small 2026-04-07 13:28 已打包 陈子轩 产物 proj_1024/task_8899/model_nn_small
XLearner 2026-04-07 13:32 已发布 陈子轩 产物 proj_1024/task_8899/model_xlearner 发布
CausalForest 2026-04-07 13:36 已打包 陈子轩 产物 proj_1024/task_8899/model_causalforest
TLearner_xgboost 2026-04-07 13:41 已发布 陈子轩 产物 proj_1024/task_8899/model_tlearner_xgboost 发布
NN_large 2026-04-07 13:47 已打包 陈子轩 产物 proj_1024/task_8899/model_nn_large
Dragonnet_recall 2026-04-07 13:55 已打包 陈子轩 产物 proj_1024/task_8901/model_dragonnet_recall
XLearner_ctr 2026-04-07 14:02 已发布 陈子轩 产物 proj_1024/task_8901/model_xlearner_ctr 发布
任务预处理
  • 大卡负反馈
模型训练
  • 活跃天数/LT-Dragonnet
  • APP使用时长-Dragonnet
  • 总消费vv数-Dragonnet
  • 活跃天数/LT-TLearner_xgboost
  • APP使用时长-TLearner_xgboost
  • 总消费vv数-TLearner_xgboost
  • 活跃天数/LT-XLearner
  • APP使用时长-XLearner
  • 总消费vv数-XLearner
  • 活跃天数/LT-CausalForest
  • APP使用时长-CausalForest
  • 总消费vv数-CausalForest
模型评估
任务预处理 / 大卡负反馈 负责人:lijiayun03
https://kml.corp.kuaishou.com/v2/#/project/517/training/tasks/631771?tab=records&selectedRecordId=15009679
2026-04-02T04:16:49.165E078Z stdout Looking in indexes: https://pypi.corp.kuaishou.com/kuaishou/prod/+simple/
2026-04-02T04:20:16.502E5456Z stdout Collecting git+https://oauth2:***@git.corp.kuaishou.com/intelligent-technology-algorithm/cluster-utils.git
2026-04-02T04:20:16.508E6614Z stderr Running command git clone --filter=blob:none --quiet https://oauth2:***@git.corp.kuaishou.com/intelligent-technology-algorithm/cluster-utils.git
2026-04-02T04:20:18.151E1653Z stdout Resolved https://oauth2:***@git.corp.kuaishou.com/intelligent-technology-algorithm/cluster-utils.git
2026-04-02T04:20:18.158E4445Z stdout Preparing metadata (setup.py): started
2026-04-02T04:20:18.527E5271Z stdout Preparing metadata (setup.py): finished with status 'done'
2026-04-02T04:20:18.769E2611Z stdout Collecting datasketches==4.1.0
2026-04-02T04:20:18.784E7515Z stdout Downloading https://pypi.corp.kuaishou.com/kuaishou/prod/datasketches/4.1.0/datasketches-4.1.0-cp37.whl (841.2 kB)
2026-04-02T04:20:19.009E8752Z stdout Collecting json5==0.9.6
2026-04-02T04:20:19.267E8407Z stdout Requirement already satisfied: numpy in /root/miniconda3/envs/dev/lib/python3.7/site-packages
2026-04-02T04:20:19.030E7131Z stdout Building wheels for collected packages: cluster-utils
2026-04-02T04:20:22.791E6193Z stderr [2026-04-02 10:40:22,791] infra.doc_init WARNING: 如有问题,请先参考通用FAQ: https://halo.corp.kuaishou.com
2026-04-02T04:20:22.792E0096Z stderr [2026-04-02 10:40:22,792] infra.kafka WARNING: 不存在环境变量 KWS_SERVICE_REGION, 使用默认值: HB1
2026-04-02T04:20:22.834E9706Z stderr [2026-04-02 10:40:22,834] apscheduler.scheduler INFO: Adding job tentatively -- it will be properly scheduled when the scheduler starts
2026-04-02T04:20:22.842E5061Z stderr [2026-04-02 10:40:22,842] __main__ INFO: 运行环境: prod
2026-04-02T04:20:22.849E0767Z stderr [2026-04-02 10:40:22,849] __main__ INFO: 将按处理配置 feature_process_config.json 执行任务