1
00:00:00,169 --> 00:00:04,623
[MUSIC]

2
00:00:04,623 --> 00:00:06,940
In this module we've
covered a lot of ground.

3
00:00:06,940 --> 00:00:10,330
To start with we motivated taking
a probabilistic model-based

4
00:00:10,330 --> 00:00:12,330
approach to clustering, and

5
00:00:12,330 --> 00:00:16,690
showed mixtures of Gaussians as
a special example of such an approach.

6
00:00:16,690 --> 00:00:21,750
Then we presented the Algorithm, which is
a very generally useful algorithm, and

7
00:00:21,750 --> 00:00:25,570
we specified it specifically for
mixtures of Gaussians.

8
00:00:25,570 --> 00:00:29,470
And throughout the module we've compared
and contrasted these model based

9
00:00:29,470 --> 00:00:34,160
approaches with the K-means algorithm
that we described in the last module.

10
00:00:34,160 --> 00:00:38,240
And what we saw in this module is
that you can actually view K-means

11
00:00:38,240 --> 00:00:41,770
as a special case of For
mixtures of Gaussians.

12
00:00:41,770 --> 00:00:44,909
So, thinking about
mixtures of Gaussians and

13
00:00:44,909 --> 00:00:49,070
In order to infer our cluster
parameters and our soft assignments,

14
00:00:49,070 --> 00:00:53,829
really is a generalization of the K-means
algorithm that we showed before.

15
00:00:53,829 --> 00:00:58,081
But there is a cost to it, because there
are a lot more parameters that we have

16
00:00:58,081 --> 00:01:00,085
to think about learning from data.

17
00:01:00,085 --> 00:01:04,673
And in addition, there's a computational
cost to doing Instead of K-means,

18
00:01:04,673 --> 00:01:07,666
both when we're computing
our responsibilities and

19
00:01:07,666 --> 00:01:10,710
when we're estimating
our model parameters.

20
00:01:10,710 --> 00:01:14,040
Each of these steps is more
intensive than the steps that you

21
00:01:14,040 --> 00:01:15,870
have to perform in K-means.

22
00:01:15,870 --> 00:01:19,050
So it is a trade-off in
terms of flexibility and

23
00:01:19,050 --> 00:01:20,554
the descriptive output you get.

24
00:01:20,554 --> 00:01:23,020
You get the soft assignments
capturing uncertainty and

25
00:01:23,020 --> 00:01:27,310
you can do different things with that,
but there is a cost to it.

26
00:01:28,450 --> 00:01:32,704
And in summary, I'll just leave you with a
list of things that you should be able to

27
00:01:32,704 --> 00:01:34,380
do having watched this module.

28
00:01:37,121 --> 00:01:41,029
[MUSIC]