Home Credit 固定规则的独立复核

Summary

Metadata

{"report_id":"050120ae-8498-4d2f-b5b8-c32c27da38c7","report_type":"rule","format_version":1,"tables":{"summary":{"rows":1,"grain":"one mining run (or benchmark summary); no feature-level metrics","schema_version":1,"fields":{"selected_count":{"dtype":"Int64","unit":"count","meaning":"最终 RuleSet 中入选规则数；no_rules 为 0","null":"unavailable; inspect status and report parameters"},"status":{"dtype":"String","unit":"state","meaning":"summary success/no_rules；候选 selected/rejected/candidate/deferred；评估 ok/unobserved/empty；结合 reason 和 rejection_stage","null":"unavailable; inspect status and report parameters"}},"state":"computed","feature_query":"unsupported"},"rules":{"rows":1,"grain":"final rule_id / selection rank; grades are rule membership, not metrics","schema_version":1,"fields":{"rule_id":{"dtype":"String","unit":"identifier","meaning":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次","null":"unavailable; inspect status and report parameters"},"expression":{"dtype":"String","unit":"definition","meaning":"既有 DSL 规范表达式；不执行报告文本","null":"unavailable; inspect status and report parameters"},"rank":{"dtype":"Int64","unit":"ordinal","meaning":"最终规则集的 1 起始顺序；cumulative 使用此前全部规则的并集","null":"unavailable; inspect status and report parameters"},"source":{"dtype":"String","unit":"category","meaning":"规则生成来源；不是业务特征来源","null":"unavailable; inspect status and report parameters"},"labels":{"dtype":"List(String)","unit":"unknown","meaning":"已保存的 labels；未登记业务含义保持 unknown","null":"unavailable; inspect status and report parameters"},"grades":{"dtype":"List(String)","unit":"category","meaning":"既有 RuleSet 等级成员列表","null":"unavailable; inspect status and report parameters"},"generation_round":{"dtype":"Int64","unit":"ordinal","meaning":"候选生成轮次；相同规则在不同轮次的审计不可合并","null":"unavailable; inspect status and report parameters"},"selection_round":{"dtype":"Int64","unit":"ordinal","meaning":"实际 cascade 入选轮次；未入选为 null","null":"unavailable; inspect status and report parameters"}},"state":"computed","feature_relation":{"table":"rule_features","key":"rule_id","feature":"feature","roles":{"rule_id":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次"}},"feature_query":"relation"},"evaluation":{"rows":3,"grain":"dataset / rule_id / target / slice / group (hit, miss, total)","schema_version":1,"fields":{"dataset":{"dtype":"String","unit":"dimension","meaning":"实际保存的 train/validation/in_sample 范围；不推断未保存的逐轮剩余样本统计","null":"unavailable; inspect status and report parameters"},"rule_id":{"dtype":"String","unit":"identifier","meaning":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次","null":"unavailable; inspect status and report parameters"},"target":{"dtype":"String","unit":"identifier","meaning":"真实标签列 ID；各 target 独立排除未表现标签","null":"unavailable; inspect status and report parameters"},"slice":{"dtype":"String","unit":"dimension","meaning":"__overall__ 或 evaluator 实际构造的时间／客群切片值","null":"unavailable; inspect status and report parameters"},"group":{"dtype":"String","unit":"dimension","meaning":"hit / miss / total；此列不是原始客群，客群见 slice","null":"unavailable; inspect status and report parameters"},"total_count":{"dtype":"Int64","unit":"count","meaning":"当前范围全部样本人数，包括当前 target 未观测的样本","null":"unavailable; inspect status and report parameters"},"observed_count":{"dtype":"Int64","unit":"count","meaning":"当前 target 已观测人数，与 sample_count 相同","null":"unavailable; inspect status and report parameters"},"unobserved_count":{"dtype":"Int64","unit":"count","meaning":"total_count 减 observed_count；不把未观测当好样本","null":"unavailable; inspect status and report parameters"},"distribution_coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前范围 total_count / 同 dataset、target、slice 的总体人数","null":"unavailable; inspect status and report parameters"},"status":{"dtype":"String","unit":"state","meaning":"summary success/no_rules；候选 selected/rejected/candidate/deferred；评估 ok/unobserved/empty；结合 reason 和 rejection_stage","null":"unavailable; inspect status and report parameters"},"sample_count":{"dtype":"Int64","unit":"count","meaning":"当前 target 已表现样本数；group/slice/dataset 定义统计范围","null":"unavailable; inspect status and report parameters"},"event_count":{"dtype":"Int64","unit":"count","meaning":"当前范围已表现 label=1 样本数","null":"unavailable; inspect status and report parameters"},"coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前范围 sample_count / 同 dataset、target、slice 的 base sample_count；只含已表现样本","null":"unavailable; inspect status and report parameters"},"event_rate":{"dtype":"Float64","unit":"ratio","meaning":"event_count / sample_count；零分母 null","null":"unavailable; inspect status and report parameters"},"lift":{"dtype":"Float64","unit":"dimensionless","meaning":"event_rate / 同 dataset、target、slice 的 base event_rate；零分母 null","null":"unavailable; inspect status and report parameters"},"amount_total":{"dtype":"Float64","unit":"unknown_business_unit","meaning":"当前范围已表现样本金金额和；amount_col 未提供则 null；币种见 business_context","null":"unavailable; inspect status and report parameters"},"event_amount":{"dtype":"Float64","unit":"unknown_business_unit","meaning":"当前范围已表现 label=1 金额和；amount_col 未提供则 null","null":"unavailable; inspect status and report parameters"},"amount_coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前 amount_total / 同范围 base amount_total","null":"unavailable; inspect status and report parameters"},"amount_event_rate":{"dtype":"Float64","unit":"ratio","meaning":"event_amount / amount_total；零分母 null","null":"unavailable; inspect status and report parameters"},"amount_lift":{"dtype":"Float64","unit":"dimensionless","meaning":"amount_event_rate / 同范围 base amount_event_rate","null":"unavailable; inspect status and report parameters"},"customer_count":{"dtype":"Null","unit":"count","meaning":"当前已表现范围去重客户数；customer_col 未提供则 null","null":"unavailable; inspect status and report parameters"},"event_customer_count":{"dtype":"Null","unit":"count","meaning":"当前已表现 label=1 范围去重客户数","null":"unavailable; inspect status and report parameters"},"customer_coverage":{"dtype":"Null","unit":"ratio","meaning":"customer_count / 同范围 base customer_count；不可跨规则相加","null":"unavailable; inspect status and report parameters"},"customer_event_rate":{"dtype":"Null","unit":"ratio","meaning":"event_customer_count / customer_count；不可跨规则相加","null":"unavailable; inspect status and report parameters"},"customer_lift":{"dtype":"Null","unit":"dimensionless","meaning":"customer_event_rate / 同范围 base customer_event_rate","null":"unavailable; inspect status and report parameters"},"event_rate_ci_lower":{"dtype":"Float64","unit":"ratio","meaning":"Wilson 事件率区间下界；confidence_level 见实际参数","null":"unavailable; inspect status and report parameters"},"event_rate_ci_upper":{"dtype":"Float64","unit":"ratio","meaning":"Wilson 事件率区间上界；confidence_level 见实际参数","null":"unavailable; inspect status and report parameters"},"lift_ci_lower":{"dtype":"Float64","unit":"dimensionless","meaning":"规则 Wilson 事件率下界 / 同范围总体事件率；保守 Lift 下界","null":"unavailable; inspect status and report parameters"},"lift_ci_upper":{"dtype":"Float64","unit":"dimensionless","meaning":"规则 Wilson 事件率上界 / 同范围总体事件率；保守 Lift 上界","null":"unavailable; inspect status and report parameters"},"p_value":{"dtype":"Float64","unit":"probability","meaning":"按 direction 的单侧事件富集检验 p 值；方法见 evaluator","null":"unavailable; inspect status and report parameters"},"q_value":{"dtype":"Float64","unit":"probability","meaning":"同数据范围、target、slice、group 候选检验族 BH 校正 q 值","null":"unavailable; inspect status and report parameters"}},"state":"computed","feature_relation":{"table":"rule_features","key":"rule_id","feature":"feature","roles":{"rule_id":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次"}},"feature_query":"relation"},"slices":{"rows":15,"grain":"dataset / rule_id / target / slice / group; slice encodes supplied time or segment","schema_version":1,"fields":{"dataset":{"dtype":"String","unit":"dimension","meaning":"实际保存的 train/validation/in_sample 范围；不推断未保存的逐轮剩余样本统计","null":"unavailable; inspect status and report parameters"},"rule_id":{"dtype":"String","unit":"identifier","meaning":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次","null":"unavailable; inspect status and report parameters"},"target":{"dtype":"String","unit":"identifier","meaning":"真实标签列 ID；各 target 独立排除未表现标签","null":"unavailable; inspect status and report parameters"},"slice":{"dtype":"String","unit":"dimension","meaning":"__overall__ 或 evaluator 实际构造的时间／客群切片值","null":"unavailable; inspect status and report parameters"},"group":{"dtype":"String","unit":"dimension","meaning":"hit / miss / total；此列不是原始客群，客群见 slice","null":"unavailable; inspect status and report parameters"},"total_count":{"dtype":"Int64","unit":"count","meaning":"当前范围全部样本人数，包括当前 target 未观测的样本","null":"unavailable; inspect status and report parameters"},"observed_count":{"dtype":"Int64","unit":"count","meaning":"当前 target 已观测人数，与 sample_count 相同","null":"unavailable; inspect status and report parameters"},"unobserved_count":{"dtype":"Int64","unit":"count","meaning":"total_count 减 observed_count；不把未观测当好样本","null":"unavailable; inspect status and report parameters"},"distribution_coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前范围 total_count / 同 dataset、target、slice 的总体人数","null":"unavailable; inspect status and report parameters"},"status":{"dtype":"String","unit":"state","meaning":"summary success/no_rules；候选 selected/rejected/candidate/deferred；评估 ok/unobserved/empty；结合 reason 和 rejection_stage","null":"unavailable; inspect status and report parameters"},"sample_count":{"dtype":"Int64","unit":"count","meaning":"当前 target 已表现样本数；group/slice/dataset 定义统计范围","null":"unavailable; inspect status and report parameters"},"event_count":{"dtype":"Int64","unit":"count","meaning":"当前范围已表现 label=1 样本数","null":"unavailable; inspect status and report parameters"},"coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前范围 sample_count / 同 dataset、target、slice 的 base sample_count；只含已表现样本","null":"unavailable; inspect status and report parameters"},"event_rate":{"dtype":"Float64","unit":"ratio","meaning":"event_count / sample_count；零分母 null","null":"unavailable; inspect status and report parameters"},"lift":{"dtype":"Float64","unit":"dimensionless","meaning":"event_rate / 同 dataset、target、slice 的 base event_rate；零分母 null","null":"unavailable; inspect status and report parameters"},"amount_total":{"dtype":"Float64","unit":"unknown_business_unit","meaning":"当前范围已表现样本金金额和；amount_col 未提供则 null；币种见 business_context","null":"unavailable; inspect status and report parameters"},"event_amount":{"dtype":"Float64","unit":"unknown_business_unit","meaning":"当前范围已表现 label=1 金额和；amount_col 未提供则 null","null":"unavailable; inspect status and report parameters"},"amount_coverage":{"dtype":"Float64","unit":"ratio","meaning":"当前 amount_total / 同范围 base amount_total","null":"unavailable; inspect status and report parameters"},"amount_event_rate":{"dtype":"Float64","unit":"ratio","meaning":"event_amount / amount_total；零分母 null","null":"unavailable; inspect status and report parameters"},"amount_lift":{"dtype":"Float64","unit":"dimensionless","meaning":"amount_event_rate / 同范围 base amount_event_rate","null":"unavailable; inspect status and report parameters"},"customer_count":{"dtype":"Null","unit":"count","meaning":"当前已表现范围去重客户数；customer_col 未提供则 null","null":"unavailable; inspect status and report parameters"},"event_customer_count":{"dtype":"Null","unit":"count","meaning":"当前已表现 label=1 范围去重客户数","null":"unavailable; inspect status and report parameters"},"customer_coverage":{"dtype":"Null","unit":"ratio","meaning":"customer_count / 同范围 base customer_count；不可跨规则相加","null":"unavailable; inspect status and report parameters"},"customer_event_rate":{"dtype":"Null","unit":"ratio","meaning":"event_customer_count / customer_count；不可跨规则相加","null":"unavailable; inspect status and report parameters"},"customer_lift":{"dtype":"Null","unit":"dimensionless","meaning":"customer_event_rate / 同范围 base customer_event_rate","null":"unavailable; inspect status and report parameters"},"event_rate_ci_lower":{"dtype":"Float64","unit":"ratio","meaning":"Wilson 事件率区间下界；confidence_level 见实际参数","null":"unavailable; inspect status and report parameters"},"event_rate_ci_upper":{"dtype":"Float64","unit":"ratio","meaning":"Wilson 事件率区间上界；confidence_level 见实际参数","null":"unavailable; inspect status and report parameters"},"lift_ci_lower":{"dtype":"Float64","unit":"dimensionless","meaning":"规则 Wilson 事件率下界 / 同范围总体事件率；保守 Lift 下界","null":"unavailable; inspect status and report parameters"},"lift_ci_upper":{"dtype":"Float64","unit":"dimensionless","meaning":"规则 Wilson 事件率上界 / 同范围总体事件率；保守 Lift 上界","null":"unavailable; inspect status and report parameters"},"p_value":{"dtype":"Float64","unit":"probability","meaning":"按 direction 的单侧事件富集检验 p 值；方法见 evaluator","null":"unavailable; inspect status and report parameters"},"q_value":{"dtype":"Float64","unit":"probability","meaning":"同数据范围、target、slice、group 候选检验族 BH 校正 q 值","null":"unavailable; inspect status and report parameters"}},"state":"computed","feature_relation":{"table":"rule_features","key":"rule_id","feature":"feature","roles":{"rule_id":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次"}},"feature_query":"relation"},"rule_features":{"rows":2,"grain":"distinct rule_id / English feature; includes rejected candidates","schema_version":1,"fields":{"rule_id":{"dtype":"String","unit":"identifier","meaning":"规范 DSL 的稳定哈希 ID；同 ID 可出现在不同 dataset、target 或轮次","null":"unavailable; inspect status and report parameters"},"feature":{"dtype":"String","unit":"identifier","meaning":"AST 提取的英文输入特征 ID；中文名不替代此身份","null":"unavailable; inspect status and report parameters"}},"feature_roles":{"feature":"feature"},"state":"computed","feature_query":"direct"}},"parameters":{"dataset":"fixed_all_partitions","primary_target":"TARGET","targets":["TARGET"],"group_col":"dataset_flag","time_col":null,"time_grain":null,"amount_col":"AMT_CREDIT","customer_col":null,"batch_size":100,"direction":"high_risk","confidence_level":0.95,"compute_statistics":true,"features":["EXT_SOURCE_2","EXT_SOURCE_3"],"selection_report_id":"408b6056-e5a0-4558-88d4-98cbbc56b0ea","fixed_rule_set":true},"feature_metadata":{"EXT_SOURCE_2":{"display_name":"外部来源分 2","description":"Normalized score from external data source","data_source":"Home Credit 2018 application_{train|test}.csv","unit":"归一化值（无量纲）"},"EXT_SOURCE_3":{"display_name":"外部来源分 3","description":"Normalized score from external data source","data_source":"Home Credit 2018 application_{train|test}.csv","unit":"归一化值（无量纲）"}},"business_context":{"dataset_id":"home-credit-default-risk-2018","sample_unit":"unknown","label_definition":"unknown","observation_window":"unknown","source":"https://www.kaggle.com/competitions/home-credit-default-risk/data","real_features":{"source":"application_train.csv/application_test.csv","simulated":false,"derived_features_added":false},"target":{"field":"TARGET","definition":"Target variable (1 - client with payment difficulties: he/she had late payment more than X days on at least one of the first Y installments of the loan in our sample, 0 - all other cases)","simulated":false,"available_in":"application_train.csv","unavailable_in":"application_test.csv","unobserved_value":null,"performance_window":"X/Y 未公开；不推断 bad30/bad60"},"date":{"field":"APPLY_DATE","simulated":true,"train_range":["2026-01-01","2026-06-30"],"test_range":["2026-07-01","2026-07-31"],"limitation":"功能演示日期，不支持真实跨期泛化、季节性或线上漂移结论。"},"partitions":{"field":"dataset_flag","simulated":true,"0train":"拟合、候选发现、开发筛选与 PSI 参考分布","1val":"验证或筛选；参与选择后不是最终独立评估","2test":"有标签独立评估，不等同于原始 application_test","3oot":"模拟时间外独立评估，不是真实业务 OOT","4_oot_undue":"原始比赛测试表，标签未公开；名称不代表真实贷款未到期"},"amount":{"field":"AMT_CREDIT","unit":"原始金额单位","currency":"未公开","meaning":"贷款金额，不等于损失或利润","included_as_feature":false},"weights":null,"sample_scope":"两个 application 文件的全部样本","preparation_version":"home-credit-2018-v1","special_values":{"DAYS_EMPLOYED":{"value":365243,"count":64648,"handling":"保留原值；该异常正值作为本字段的特殊值配置，不应用于其他字段。","meaning":"官方字段说明未解释此值，不推断实际就业时长。"}},"labels":{"TARGET":{"field":"TARGET","definition":"Target variable (1 - client with payment difficulties: he/she had late payment more than X days on at least one of the first Y installments of the loan in our sample, 0 - all other cases)","simulated":false,"available_in":"application_train.csv","unavailable_in":"application_test.csv","unobserved_value":null,"performance_window":"X/Y 未公开；不推断 bad30/bad60"}},"question":"同一外部来源分 2 等级内，来源分 3 的风险有何差异？固定组合规则能否通过独立复核？","score_directions":{"EXT_SOURCE_2":"lower_risk","EXT_SOURCE_3":"lower_risk"},"selection_scope":"0train discovery; 1val selection; 2test/3oot untouched until fixed evaluation","binning_origin":"0train only; fixed quantile cutpoints","evaluation_roles":"2test/3oot independent; 4_oot_undue distribution only"},"source":{"dataset_id":"home-credit-default-risk-2018","computation":"rule","producer":"mars.rule.MarsRuleReport","mars_version":"unknown","origin_project":"unknown"},"public_functions":[],"saved_operations":"public functions accept load_report ReportSnapshot; no source samples or fitted objects required","operations":["describe","get_table","query_page","get_feature","search_features","to_ai_context","save","show_table","write_excel","write_html"],"limitations":["Only saved aggregate evidence; no causal or deployment conclusion.","Null is unavailable, not numeric zero. See status and denominators.","Feature matching uses rule membership union; features AND sources may match different features of the same rule. sources is business data_source, not candidate generators.","summary and benchmark do not support features/sources queries. Advanced analysis scope is advanced_analysis; no dataset identity is inferred for supplied analysis samples.","Absent analyses are not computed; present zero-row tables are computed_empty unless explicitly not_computed. Null may be undefined or unconfigured; inspect parameters and denominators.","Snapshot is aggregate evidence, not RuleSet JSON, transform capability or deployment permission. No original samples or models are saved."],"maturity":"Experimental"}

Rules

Evaluation

Slices

Rule Features