{"version":"2026-10-02.2","occupation":{"id":"DATA","code":"15-2051.00","name":"数据科学家","nameEn":"Data Scientists","mapping":"数据科学家是参照职业，涵盖建模与统计；不等同于所有数据分析岗位","sourceUrl":"https://www.onetonline.org/link/summary/15-2051.00","taskIds":["DATA-21823","DATA-21828","DATA-21837","DATA-21829","DATA-21836","DATA-21831","DATA-21826","DATA-21827","DATA-21833","DATA-21824","DATA-21832","DATA-21838","DATA-21835","DATA-21834","DATA-21825","DATA-21830"]},"task":{"id":"DATA-21837","occupationId":"DATA","sourceTaskId":"21837","title":"测试、验证并改进模型，确保对目标结果的预测准确","sourceStatement":"Test, validate, and reformulate models to ensure accurate prediction of outcomes of interest.","sourceVersion":"O*NET 31.0","sourceUrl":"https://www.onetonline.org/link/summary/15-2051.00#Tasks","sourceTaskType":"Core","importance":4.25,"scenario":"指定模型版本、数据划分及验收指标","coveredActions":["计算原模型的验证指标","定位预测误差分布","修改指定模型配置","重新验证改后模型","接受模型上线风险"],"notCovered":["该场景之外的组织规则、其他系统或数据、长期结果尚未验证"],"dependencyBasis":"本场景显式审阅的顺序依赖；不是O*NET或DWA规定的唯一工作流","decompositionBasis":"本站基于任务陈述及关联DWA设计的场景还原候选；DWA不是执行顺序或官方原子切分","relatedActivities":[{"id":"4.A.2.b.2.u.3","name":"Develop scientific or mathematical models."}],"atomIds":["DATA-21837-A1","DATA-21837-A2","DATA-21837-A3","DATA-21837-A4","DATA-21837-A5"],"caseIds":["DATA-04"],"completionRule":"全部必要原子产物与依赖通过，且需要的负责人确认、真实交互或现场动作有实际证据，才可验收本页限定场景；未覆盖的来源动作仍需补齐，不能据此宣告职业任务在所有场景完成","successRate":null,"fieldValidated":false},"atoms":[{"id":"DATA-21837-A1","taskId":"DATA-21837","title":"计算原模型的验证指标","scope":"指定模型版本、数据划分及验收指标","inputs":["获准的数据、字段字典、业务问题、SQL或统计环境与口径"],"output":"原模型验证指标","acceptance":["原模型验证指标注明数据版本、筛选规则、公式、单位及不确定假设","结果能独立复算；缺失值、边界或不适用情形明确保留，不自行填补"],"dependsOn":[],"mode":"ai_candidate","feasibility":"unverified_candidate","assessmentBasis":"输入与产物可界定，可提出独立验收；这是待测试的分工建议","evidence":[],"verifiedExamples":[],"successRate":null,"fieldValidated":false,"stopCondition":"输入或权限缺失、关键验收失败或存在未解决冲突时，保留产物并交回责任人；不能自动宣告父任务完成"},{"id":"DATA-21837-A2","taskId":"DATA-21837","title":"定位预测误差分布","scope":"指定模型版本、数据划分及验收指标","inputs":["获准的数据、字段字典、业务问题、SQL或统计环境与口径","前置产物：原模型验证指标"],"output":"误差分布表","acceptance":["误差分布表覆盖指定范围内的对象，事实与假设分开且事实有出处","缺失条件、冲突与未支持结论明确列出；表示方式可不同，不按固定措辞验收"],"dependsOn":["DATA-21837-A1"],"mode":"ai_candidate","feasibility":"unverified_candidate","assessmentBasis":"输入与产物可界定，可提出独立验收；这是待测试的分工建议","evidence":[],"verifiedExamples":[],"successRate":null,"fieldValidated":false,"stopCondition":"输入或权限缺失、关键验收失败或存在未解决冲突时，保留产物并交回责任人；不能自动宣告父任务完成"},{"id":"DATA-21837-A3","taskId":"DATA-21837","title":"修改指定模型配置","scope":"指定模型版本、数据划分及验收指标","inputs":["获准的数据、字段字典、业务问题、SQL或统计环境与口径","前置产物：误差分布表"],"output":"实际新配置及版本","acceptance":["保存实际新配置、模型版本、变更项和所用训练数据；配置说明不得代替变更产物","修改后的版本必须作为后续重新验证的输入，原版本指标不能验收新版本"],"dependsOn":["DATA-21837-A2"],"mode":"ai_candidate","feasibility":"unverified_candidate","assessmentBasis":"输入与产物可界定，可提出独立验收；这是待测试的分工建议","evidence":[],"verifiedExamples":[],"successRate":null,"fieldValidated":false,"stopCondition":"输入或权限缺失、关键验收失败或存在未解决冲突时，保留产物并交回责任人；不能自动宣告父任务完成"},{"id":"DATA-21837-A4","taskId":"DATA-21837","title":"重新验证改后模型","scope":"指定模型版本、数据划分及验收指标","inputs":["获准的数据、字段字典、业务问题、SQL或统计环境与口径","前置产物：实际新配置及版本"],"output":"新版本验证指标","acceptance":["在固定验证划分实际运行改后版本，保存新旧版本、指标、门槛和原始结果","对新版本独立判断是否达标；验证未运行或门槛未绑定，风险接收不得报已满足质量条件"],"dependsOn":["DATA-21837-A3"],"mode":"ai_candidate","feasibility":"unverified_candidate","assessmentBasis":"输入与产物可界定，可提出独立验收；这是待测试的分工建议","evidence":[],"verifiedExamples":[],"successRate":null,"fieldValidated":false,"stopCondition":"输入或权限缺失、关键验收失败或存在未解决冲突时，保留产物并交回责任人；不能自动宣告父任务完成"},{"id":"DATA-21837-A5","taskId":"DATA-21837","title":"接受模型上线风险","scope":"指定模型版本、数据划分及验收指标","inputs":["获准的数据、字段字典、业务问题、SQL或统计环境与口径","前置产物：新版本验证指标"],"output":"上线风险决定","acceptance":["上线风险决定写明决定人、授权范围、所依据的事实及仍未确认项","不能用AI草稿、模型成绩或工具成功状态替代负责人真实确认"],"dependsOn":["DATA-21837-A4"],"mode":"human_decision","feasibility":"requires_human","assessmentBasis":"当前场景含负责人的决定或现场动作","evidence":[],"verifiedExamples":[],"successRate":null,"fieldValidated":false,"stopCondition":"输入或权限缺失、关键验收失败或存在未解决冲突时，保留产物并交回责任人；不能自动宣告父任务完成"}],"completionRules":{"example":"只证明保存的AI产物在一个限定例中通过，不能外推子任务完成率","adjacent":"相近基准证明测试中存在相关能力，不证明本站原子项已完成","candidate":"可交给AI先尝试是编辑建议，完成性待验收","human":"人的决定或现场动作是当前场景的接管点，不等于AI不能提供辅助","rollup":"父任务须通过所有必要产物、依赖与人工门槛；禁止平均子任务数量或外部分数得到职业自动化率"}}