Got a YouTube account?

New: enable viewer-created translations and captions on your YouTube channel!

English subtitles

← Getting Rid of Features - Intro to Machine Learning

Get Embed Code
4 Languages

Showing Revision 3 created 05/25/2016 by Udacity Robot.

  1. So far we've talked about engineering new features that you
  2. might want to add to a data set.
  3. But it's just as important to know how to get rid of features that you
  4. might not want.
  5. Now it might not be clear right now why you would ever want to get
  6. rid of a feature.
  7. You might be like well, but
  8. Katie that's throwing out information, that's throwing out data.
  9. Why would I ever want to do that?
  10. And I would propose to you that there are many very good reasons why you
  11. might want to get rid of features.
  12. We'll be talking about some of those in the rest of this lesson.
  13. But I want to make you think about it on your own first.
  14. So here are a few reasons that you might want to ignore a feature.
  15. And I want to you to place a check next to all the ones that you think
  16. actually sound like good reasons to ignore a feature to you.
  17. So perhaps one problem with a feature is that it's noisy,
  18. that it's hard to distinguish whether it's,
  19. it's reliably measuring what you want it to be measuring.
  20. Another thing that could potentially be happening is that a feature maybe it
  21. is causing your model to over-fit for some reason.
  22. Maybe you think that the feature is strongly related or
  23. what we call highly correlated with a feature that's already present.
  24. So it's giving you information but it's just repeating information
  25. that's already present in the form of another feature.
  26. And then the last possibility is that additional features maybe they
  27. slow down the training or the testing process.
  28. And so in order to keep things moving along quickly, you want to be working with
  29. the bare minimum number of features that are required to get good performance.
  30. So you tell me what you think.
  31. Which of these are good reasons to ignore a feature?