1
00:00:00,000 --> 00:00:05,067
[MUSIC]

2
00:00:05,067 --> 00:00:08,854
Toward the end of these lessons,
we're going to Python and

3
00:00:08,854 --> 00:00:12,900
the scikit-learn project to
write our own classifier.

4
00:00:12,900 --> 00:00:17,590
But before we continue, we should formally
define some of the terms I've been using

5
00:00:17,590 --> 00:00:19,600
to describe machine learning and

6
00:00:19,600 --> 00:00:23,290
then break them down
further with more examples.

7
00:00:23,290 --> 00:00:29,840
Speaking of examples, an example
is a single element in a dataset.

8
00:00:29,840 --> 00:00:35,610
Sometimes you might hear an example
referred to as a sample,

9
00:00:35,610 --> 00:00:37,750
but it means the same thing.

10
00:00:37,750 --> 00:00:40,710
If your data is formatted in a table,

11
00:00:40,710 --> 00:00:45,020
an example might be
a single row in the table.

12
00:00:45,020 --> 00:00:48,930
A dataset is comprised on many examples.

13
00:00:48,930 --> 00:00:49,940
And in general,

14
00:00:49,940 --> 00:00:55,000
each example helps improve the confidence
of your model's predictions.

15
00:00:55,000 --> 00:01:00,150
Say for instance, you're running a movie
studio and you want to try an forecast

16
00:01:00,150 --> 00:01:04,492
how much money a movie might make,
so that you can set a budget.

17
00:01:04,492 --> 00:01:09,660
Your dataset would probably
be examples of older movies.

18
00:01:09,660 --> 00:01:13,510
So what about those older
movies might you include?

19
00:01:13,510 --> 00:01:17,920
Each part of an example
is called a feature.

20
00:01:17,920 --> 00:01:22,180
A feature is one
characteristic of an example.

21
00:01:22,180 --> 00:01:25,670
Again, if you formatted
your data in a table,

22
00:01:25,670 --> 00:01:29,200
each feature might be a single column.

23
00:01:29,200 --> 00:01:34,220
In the case of predicting a movie's box
office performance, your older examples of

24
00:01:34,220 --> 00:01:38,610
movies might include things like
their total box office sales.

25
00:01:38,610 --> 00:01:41,760
The budget, the genre, release date and

26
00:01:41,760 --> 00:01:46,510
maybe more advanced features,
like a star power calculation.

27
00:01:46,510 --> 00:01:50,760
Which could take all the actors in each
movie and calculate a weighted average of

28
00:01:50,760 --> 00:01:54,540
their typical box office performance
in other movies they've been in.

29
00:01:56,330 --> 00:02:00,340
A dataset might contain good and
bad features.

30
00:02:00,340 --> 00:02:04,070
And some features that are more
important than others.

31
00:02:04,070 --> 00:02:06,840
For example,
you might find that the genre and

32
00:02:06,840 --> 00:02:10,100
release date is more
important than the budget.

33
00:02:10,100 --> 00:02:14,290
So your model could weigh
those features more heavily.

34
00:02:14,290 --> 00:02:19,230
A feature that might be completely
irrelevant is the movie's title.

35
00:02:19,230 --> 00:02:23,830
Sure a movie needs a title and you might
be able to come up with a machine learning

36
00:02:23,830 --> 00:02:28,000
model that can determine what
makes a good and bad movie title.

37
00:02:28,000 --> 00:02:31,900
But in most cases,
its probably too subjective and

38
00:02:31,900 --> 00:02:36,349
inconsequential to weigh it against
other more quantifiable features.

39
00:02:37,550 --> 00:02:42,360
Something like the box office performance
of a movie is very difficult to predict.

40
00:02:42,360 --> 00:02:45,740
And it includes a huge number
of factors that are nearly

41
00:02:45,740 --> 00:02:48,570
impossible to simulate perfectly.

42
00:02:48,570 --> 00:02:51,830
But that's why a model is
nothing more than that.

43
00:02:51,830 --> 00:02:55,710
A model or
a simplification of the problem.

44
00:02:55,710 --> 00:03:00,590
It's just one tool that can be used
in combination with other approaches

45
00:03:00,590 --> 00:03:02,130
to arrive at a solution.