1
00:00:00,370 --> 00:00:04,020
Machine learning approaches like
classifiers, regressions, and

2
00:00:04,020 --> 00:00:09,680
cluster analysis can help you navigate raw
data sets that need customized solutions.

3
00:00:09,680 --> 00:00:10,780
In computer science and

4
00:00:10,780 --> 00:00:15,915
software engineering, doing custom work
like this is often called low level

5
00:00:15,915 --> 00:00:19,055
because you're working with the most
basic pieces of the problem.

6
00:00:20,255 --> 00:00:24,475
But most approaches to machine learning
can be described using statistical

7
00:00:24,475 --> 00:00:26,425
analysis and math.

8
00:00:26,425 --> 00:00:31,065
Which means that some of the components
for a spam filter that classifies

9
00:00:31,065 --> 00:00:37,440
something as spam or not spam can be
reapplied to another similar problem, like

10
00:00:37,440 --> 00:00:41,400
classifying different types of zoo animals
based on their physical appearance.

11
00:00:42,700 --> 00:00:47,220
For this reason, there are many high
level machine learning frameworks

12
00:00:47,220 --> 00:00:51,230
that already have all
the low level code written so

13
00:00:51,230 --> 00:00:56,870
you can just plug your code and
your data into pre-existing models.

14
00:00:56,870 --> 00:01:01,660
Later in this course we're going to
use a free one called scikit-learn.

15
00:01:01,660 --> 00:01:05,860
Let's hop into a web browser and
take a quick look at the big

16
00:01:05,860 --> 00:01:08,370
machine learning frameworks
that are already out there.

17
00:01:09,960 --> 00:01:12,271
First up is scikit-learn.

18
00:01:12,271 --> 00:01:17,090
Scikit-learn is free to use and
it's written in Python.

19
00:01:17,090 --> 00:01:21,820
It allows you to utilize some of
the models we've already discussed

20
00:01:21,820 --> 00:01:24,110
in just a few lines of code.

21
00:01:24,110 --> 00:01:28,210
Right on the home page,
you can see it features tools for

22
00:01:28,210 --> 00:01:33,050
classification, regression,
clustering and more.

23
00:01:34,200 --> 00:01:37,210
This is a good choice for
many different types of projects,

24
00:01:37,210 --> 00:01:42,315
especially on very customized problem
domains, because it's all open-source.

25
00:01:43,710 --> 00:01:45,960
Next is TensorFlow.

26
00:01:45,960 --> 00:01:50,680
TensorFlow is another free
open-source Python project with

27
00:01:50,680 --> 00:01:55,500
a different approach to machine learning
that uses what are called Tensors,

28
00:01:55,500 --> 00:01:59,270
which can be thought of as
multi-dimensional arrays.

29
00:01:59,270 --> 00:02:02,710
Then operations can be
performed on Tensors, and

30
00:02:02,710 --> 00:02:05,830
the results can be visualized in graphs.

31
00:02:05,830 --> 00:02:09,190
This is somewhat of a departure from the
machine learning approaches I mentioned

32
00:02:09,190 --> 00:02:10,350
previously.

33
00:02:10,350 --> 00:02:13,150
But it's worth exploring the website and

34
00:02:13,150 --> 00:02:15,410
understanding what
TensorFlow has to offer.

35
00:02:16,880 --> 00:02:21,370
Next, is Amazon Web Services and
their machine learning products,

36
00:02:21,370 --> 00:02:25,630
which is more of a cloud based
platform than a framework.

37
00:02:25,630 --> 00:02:30,915
If we scroll down a bit there's
a section that lists some

38
00:02:30,915 --> 00:02:35,863
API services and
that includes image recognition,

39
00:02:35,863 --> 00:02:41,609
language translation,
tools to create chat bots and more.

40
00:02:41,609 --> 00:02:47,264
This can be a good choice for many
different problems, but especially if your

41
00:02:47,264 --> 00:02:52,840
problem domain can be neatly categorized
into one of these high level APIs.

42
00:02:54,340 --> 00:02:59,700
IBM Watson offers many high level products
and services for machine learning.

43
00:02:59,700 --> 00:03:04,590
They include common problem domains
like chat bots and imagine recognition.

44
00:03:04,590 --> 00:03:07,780
But also includes some others

45
00:03:07,780 --> 00:03:12,680
that focus on natural language
processing and emotional intent.

46
00:03:12,680 --> 00:03:15,540
Similar to the others,
this is a great choice for

47
00:03:15,540 --> 00:03:17,590
many different types of problems.

48
00:03:17,590 --> 00:03:21,570
And it can give you a big
head start when you need to

49
00:03:21,570 --> 00:03:26,210
utilize their higher level features that
would be difficult to create on your own.

50
00:03:27,480 --> 00:03:30,380
There are many more
frameworks out there so

51
00:03:30,380 --> 00:03:34,200
check the notes associated with
this video for more resources.

52
00:03:34,200 --> 00:03:38,348
As you've probably gathered, not all
machine learning frameworks are the same.

53
00:03:38,348 --> 00:03:43,540
And in fact, you could argue that not
all of these are really frameworks,

54
00:03:43,540 --> 00:03:48,140
but also software libraries or
cloud based platforms.

55
00:03:48,140 --> 00:03:53,550
The semantics and scopes of these tools
varies but as you gain more experience and

56
00:03:53,550 --> 00:03:56,510
encounter different types of
machine learning problems,

57
00:03:56,510 --> 00:04:00,430
it's important to know that there
are many resources available to you.