求大牛指教 “How do you deal with missing data?”# DataSciences - 数据科学
a*y
1 楼
不sure 怎么回答这个题，求大牛指教。
要不要把missing value拿median/average补上？
十分感谢！
要不要把missing value拿median/average补上？
十分感谢！
s*h
2 楼
My answers for my onsite interview:
1. Do as you did
2. Assuming some distributions hidden in your missing, then filling the
missing value accordingly
3. If we can justify the missing at random (or complete at random, but it is
too optimistic), try several imputation methods like multiple imputation
methods. Basic idea is to assume some dependency relationship with other
predictors, using them to predict the missings.
For information, see Rubin's paper about Multiple imputation.
1. Do as you did
2. Assuming some distributions hidden in your missing, then filling the
missing value accordingly
3. If we can justify the missing at random (or complete at random, but it is
too optimistic), try several imputation methods like multiple imputation
methods. Basic idea is to assume some dependency relationship with other
predictors, using them to predict the missings.
For information, see Rubin's paper about Multiple imputation.
s*a
3 楼
不可以用avg补。除非是random。但是你一般又不知道他是不是random的。
a*g
4 楼
em算法是个选择吧
相关阅读
跪求一篇文献大家都怎么做kaggle 上面的题？非csee的engineer做machine learning engineer，办身份会有问题吗华为在Seattle开分部：诚招Cloud Computing Chief Architect & (转载)感觉已经把market上的data scientist投遍了Big Data参考书汇总一道time series的预测题Re: MapR Technologies continue hiring a lot of positions (转载)T家在线题2道用10-fold cross-validation 之后怎么挑Model？像 kappa-test 这种，什么书上会讲？ (转载)在R里merge两个dataframe太慢了找DS的工作 帮忙分析下用R画图如何设置X-Y轴的比例big data一般用到的硬件会是什么架构coursera上ml的课，需要交作业吗？Chief Data Scientist at NYCSalary Negotiation Question (转载)如何学习Hadoop?请教做信号处理的老帮菜怎么转DS?