How do you handle dates in your data? Do you convert them into time stamp or you discard it? I am working on one regression problem where one attribute is having dates and it is an important feature. but I am not sure what is the best way to encode it. Can you share some ways to handle this kind of situation? Thanks a lot!
As per my experience, this has to come from business side. For example, if you are trying to detect Fraud in transactions then yes, you have to consider the time stamp or stock prices yes time is important as well, but say you are looking into Real estate sales prices, then you can ignore the time, and instead even aggregate to week level. Because in real estates, the price changes slowly or such attribute can also be called slowly changing attributes. Hope this helps.
![[Data Science YVR] How to (almost) win at Kaggle - Kiri Nichol](http://i.ytimg.com/vi/JyEm3m7AzkE/mqdefault.jpg)
Very informative...
Are the slides available for download?
speakerdeck.com/datasciencela/jeong-yoon-lee-winning-data-science-competitions-data-science-meetup-oct-2015 I found them here
How do you handle dates in your data? Do you convert them into time stamp or you discard it?
I am working on one regression problem where one attribute is having dates and it is an important feature. but I am not sure what is the best way to encode it. Can you share some ways to handle this kind of situation? Thanks a lot!
As per my experience, this has to come from business side. For example, if you are trying to detect Fraud in transactions then yes, you have to consider the time stamp or stock prices yes time is important as well, but say you are looking into Real estate sales prices, then you can ignore the time, and instead even aggregate to week level. Because in real estates, the price changes slowly or such attribute can also be called slowly changing attributes. Hope this helps.
convert it into seconds. LOL.I've seen some people convert it to number of days past xx/xx/xx or number of weeks since xx/xx/xx