From the course: Executive Guide to Predictive Modeling Strategy at Scale
Unlock the full course today
Join today to access over 22,600 courses taught by industry experts or purchase this course individually.
How to sample properly
From the course: Executive Guide to Predictive Modeling Strategy at Scale
How to sample properly
- [Instructor] Now we're going to talk about sampling. I find that a lot of folks are afraid to sample and there's really no need to be. First, in most projects that I do these days, I don't have to worry about sampling because the models will run in just a few minutes, even if I have five or 10 million rows. You get up into tens of millions, it might take 15 or 30 minutes but you'd be surprised how quickly these models will run. Secondly, folks are afraid it's going to introduce some kind of bias. It doesn't tend to and with these techniques I'm about to mention, you can eliminate that concern. First, there is a mistake that some folks make that is easy to avoid. Let me explain. If you're looking at my credit card transactions, it's okay that I'm not included in the sample of customers that you chose to look at but if I add 81 credit card transactions last year, you don't want to sample the transactions because then…
Contents
-
-
-
-
-
-
(Locked)
Understanding the modeling process2m 37s
-
(Locked)
Slow algorithms: Brute force1m 59s
-
(Locked)
Slow algorithms: More calculations1m 30s
-
(Locked)
Slow algorithms: More models2m 24s
-
(Locked)
How to sample properly2m 36s
-
(Locked)
Modeling with missing data3m 37s
-
(Locked)
Looking ahead to deployment and scoring in production2m 26s
-
(Locked)
-