[Script Info]
Title:
[Events]
Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text
Dialogue: 0,0:00:00.31,0:00:02.62,Default,,0000,0000,0000,,So the way that I got the result that Chris,
Dialogue: 0,0:00:02.62,0:00:05.77,Default,,0000,0000,0000,,should be wearing the same size tee shirt as Sarah rather than Cameron,
Dialogue: 0,0:00:05.77,0:00:08.89,Default,,0000,0000,0000,,is that I compared these numbers that I computed for each of these people.
Dialogue: 0,0:00:08.89,0:00:13.02,Default,,0000,0000,0000,,And I said is Chris closer to Cameron's number or to Sarah's number, and as it
Dialogue: 0,0:00:13.02,0:00:17.98,Default,,0000,0000,0000,,turns out, he's about 26 away from Sarah and he's about 35 away from Cameron.
Dialogue: 0,0:00:17.98,0:00:19.37,Default,,0000,0000,0000,,Closer to Sarah.
Dialogue: 0,0:00:19.37,0:00:23.00,Default,,0000,0000,0000,,Now what went wrong here is that this metric of height plus weight
Dialogue: 0,0:00:23.00,0:00:27.03,Default,,0000,0000,0000,,has two very imbalanced features in it, height and weight.
Dialogue: 0,0:00:27.03,0:00:30.05,Default,,0000,0000,0000,,So here's what I mean by that, the height is going to
Dialogue: 0,0:00:30.05,0:00:34.28,Default,,0000,0000,0000,,be a number that generally goes between let's say, the numbers of five and
Dialogue: 0,0:00:34.28,0:00:38.62,Default,,0000,0000,0000,,seven, the weight, on the other hand, takes on much larger values.
Dialogue: 0,0:00:38.62,0:00:42.25,Default,,0000,0000,0000,,Between 115 and 175 pounds in this example.
Dialogue: 0,0:00:42.25,0:00:45.70,Default,,0000,0000,0000,,So what ends up happening when you compute the sum of the two of them, is that
Dialogue: 0,0:00:45.70,0:00:49.53,Default,,0000,0000,0000,,the weight almost always will completely dominate the answer that you get.
Dialogue: 0,0:00:49.53,0:00:52.47,Default,,0000,0000,0000,,And height ends up being effectively a rounding error.
Dialogue: 0,0:00:52.47,0:00:56.75,Default,,0000,0000,0000,,Whereas what you probably want is something where the two features are equally
Dialogue: 0,0:00:56.75,0:00:59.43,Default,,0000,0000,0000,,weighted in, in the sum when you add them together.
Dialogue: 0,0:00:59.43,0:01:01.49,Default,,0000,0000,0000,,And this is what feature scaling does.
Dialogue: 0,0:01:01.49,0:01:04.18,Default,,0000,0000,0000,,It's a method for re-scaling features like these ones,
Dialogue: 0,0:01:04.18,0:01:09.98,Default,,0000,0000,0000,,so that they always span comparable ranges, usually between zero and one.
Dialogue: 0,0:01:09.98,0:01:13.60,Default,,0000,0000,0000,,So then, the numbers that you get from height will be between zero and one,
Dialogue: 0,0:01:13.60,0:01:16.04,Default,,0000,0000,0000,,they'll still contain the same information.
Dialogue: 0,0:01:16.04,0:01:18.30,Default,,0000,0000,0000,,But just expressed in different units.
Dialogue: 0,0:01:18.30,0:01:21.63,Default,,0000,0000,0000,,And the weight will also be expressed between zero and one.
Dialogue: 0,0:01:21.63,0:01:24.97,Default,,0000,0000,0000,,Again, you'll still have the information there, that Cameron raised the most and
Dialogue: 0,0:01:24.97,0:01:29.01,Default,,0000,0000,0000,,Sarah raised the least, but it'll be expressed over this much smaller range.
Dialogue: 0,0:01:29.01,0:01:30.48,Default,,0000,0000,0000,,Then when you add them together,
Dialogue: 0,0:01:30.48,0:01:33.42,Default,,0000,0000,0000,,weight won't completely dominate the equation anymore.
Dialogue: 0,0:01:33.42,0:01:36.39,Default,,0000,0000,0000,,And when that happens, you should get a much more sensible result for
Dialogue: 0,0:01:36.39,0:01:37.88,Default,,0000,0000,0000,,Chris's t-shirt size.
Dialogue: 0,0:01:37.88,0:01:41.12,Default,,0000,0000,0000,,Because even though he's a little bit closer to Sarah in weight,
Dialogue: 0,0:01:41.12,0:01:43.52,Default,,0000,0000,0000,,he's a lot closer to Cameron in height and so
Dialogue: 0,0:01:43.52,0:01:45.76,Default,,0000,0000,0000,,he'll probably end up getting grouped with Cameron.
Dialogue: 0,0:01:45.76,0:01:48.34,Default,,0000,0000,0000,,In the next video I'll show you the equation for feature scaling.