1
00:00:00,360 --> 00:00:01,470
Welcome back.

2
00:00:01,470 --> 00:00:05,810
We're in a totally new section now which is very very exciting.

3
00:00:05,910 --> 00:00:10,050
And once again we'll start off reminding ourselves with the framework we're using for machine learning

4
00:00:10,050 --> 00:00:10,980
projects.

5
00:00:10,980 --> 00:00:12,810
We begin with problem definition.

6
00:00:12,810 --> 00:00:14,450
We look at what data we have.

7
00:00:14,520 --> 00:00:17,370
We define an evaluation metric for ourselves.

8
00:00:17,430 --> 00:00:20,210
We check out the features which one should we model.

9
00:00:20,340 --> 00:00:23,190
We figure out what kind of machine learning model we use.

10
00:00:23,240 --> 00:00:27,550
Now we run some experiments on seeing if we can improve things or try something else.

11
00:00:27,630 --> 00:00:33,940
And once again all of this is an iterative process and we've seen a few tools now that we can use for

12
00:00:33,940 --> 00:00:38,700
our framework which is is that pot lib num pie.

13
00:00:38,770 --> 00:00:44,110
But now I'm bursting at the seams here with excitement for this section why because we're gonna start

14
00:00:44,140 --> 00:00:50,380
getting into the machine learning of things writing machine learning code and to do so we're going to

15
00:00:50,380 --> 00:00:52,390
be using psychic loan.

16
00:00:52,540 --> 00:00:56,540
Now you might have a question what is psychic loan.

17
00:00:56,650 --> 00:01:03,550
Now I've put in brackets SCA loan because you'll often hear psychic loan and SBA loan used interchangeably

18
00:01:03,850 --> 00:01:07,290
and it's SBA loan is just the shortened version of psychic loan.

19
00:01:07,630 --> 00:01:15,280
So as a loan you say yes Kay loan or psychic loan you can think of them talking about the same thing.

20
00:01:15,310 --> 00:01:23,470
Well fundamentally psychic loan is a python machine learning library which means if we have data so

21
00:01:23,480 --> 00:01:29,770
I get loan helps us build machine learning models to make predictions or loan patterns within that data

22
00:01:29,860 --> 00:01:31,460
and then make predictions.

23
00:01:31,510 --> 00:01:38,660
And it also implements tools to help us evaluate those predictions are they good or are they bad.

24
00:01:38,680 --> 00:01:39,970
Are they worth holding onto.

25
00:01:39,970 --> 00:01:44,640
Could we put them into our application and actually use them now.

26
00:01:44,690 --> 00:01:46,230
Why psychic loan.

27
00:01:46,340 --> 00:01:50,810
Well it's built on num pie and map plot lib a couple of tools that we've seen.

28
00:01:50,890 --> 00:01:54,320
Oh and Python of course which is our language of choice.

29
00:01:54,320 --> 00:01:58,550
It has many built in or inbuilt machine learning models as we'll see.

30
00:01:58,550 --> 00:02:00,290
We'll get hands on with a few of them.

31
00:02:00,320 --> 00:02:04,970
And now this is important right because a lot of the time people get confused thinking they have to

32
00:02:05,240 --> 00:02:10,190
learn machine learning the math inside and out before they can start to use machine learning models

33
00:02:10,460 --> 00:02:15,740
but psychic line sort of abstracts away from that and lets us apply the benefits of machine learning

34
00:02:15,980 --> 00:02:23,270
to our problem straight away by leveraging the machine learning models it has inbuilt in the library.

35
00:02:23,720 --> 00:02:29,720
And it also has methods to evaluate such machine learning models so say we build one as we talked about

36
00:02:29,720 --> 00:02:32,970
before on our data for some sort of problem that we're working on.

37
00:02:33,080 --> 00:02:38,030
How do we know if that model has actually learned anything or if we did put it into production could

38
00:02:38,030 --> 00:02:39,780
be trusted in some way or another.

39
00:02:40,040 --> 00:02:45,620
And as a little bit of an aside it is a very well-designed API and API.

40
00:02:45,650 --> 00:02:48,620
Another word for just programming library it's like it.

41
00:02:48,620 --> 00:02:49,620
Once you get into it.

42
00:02:49,700 --> 00:02:54,320
Once you get hands on with it there's a lot there but you'll start to realize oh wow the people who

43
00:02:54,320 --> 00:03:01,930
have made this have put a lot of thought into it and made it beautiful So what are we going to cover.

44
00:03:01,930 --> 00:03:06,090
Well we devised a psychic learn workflow.

45
00:03:06,100 --> 00:03:07,300
What's this look like.

46
00:03:07,300 --> 00:03:10,650
Well that first step is to get our data ready.

47
00:03:10,660 --> 00:03:13,560
And the next is to pick a model to suit your problem.

48
00:03:13,570 --> 00:03:19,420
Now if you're wondering what is going on here with this web of circles and blobs and different colors

49
00:03:19,420 --> 00:03:20,240
and different shapes.

50
00:03:20,240 --> 00:03:22,520
Well don't worry we're going to see what that is.

51
00:03:22,520 --> 00:03:24,010
Just hold tight hold tight.

52
00:03:24,010 --> 00:03:25,360
I'm excited too.

53
00:03:25,420 --> 00:03:28,180
And then step three is we're going to fit the model.

54
00:03:28,180 --> 00:03:34,500
So once we've chosen a model using this web here we're going to fit that to our data and fit in psychic

55
00:03:34,500 --> 00:03:36,880
line is another word for a model.

56
00:03:36,910 --> 00:03:38,920
Find patterns within our data.

57
00:03:38,920 --> 00:03:44,320
And then once we've finished once it's found patterns within the data we'll use that trained model which

58
00:03:44,320 --> 00:03:50,230
is a no word for a machine learning model who has found patterns in data to make a prediction then we'll

59
00:03:50,230 --> 00:03:54,070
evaluate those predictions see if they're worth holding on to or maybe not.

60
00:03:54,080 --> 00:03:59,770
Maybe we need to improve through experimentation that might involve turning the knobs on some of our

61
00:03:59,770 --> 00:04:01,740
models or doing experiments.

62
00:04:01,870 --> 00:04:08,350
And then finally we'll see how to save and reload a trained model because sometimes machine learning

63
00:04:08,350 --> 00:04:11,730
or training machine learning models it can take a while for them to learn.

64
00:04:11,800 --> 00:04:17,320
And so it'd be helpful if we could save a model and then reload it so we could share it with a colleague

65
00:04:17,560 --> 00:04:23,770
or share our results online or put our model into an application of sorts and we wouldn't have to learn

66
00:04:23,770 --> 00:04:26,350
the patterns in data every single time that you used it.

67
00:04:26,770 --> 00:04:28,630
So this is what we'll be working through.

68
00:04:29,680 --> 00:04:33,470
And then more specifically we're going to do an end to end psychic loan workflow.

69
00:04:33,610 --> 00:04:39,090
And then these steps here are really diving deeper into the steps in that workflow that we just saw.

70
00:04:39,640 --> 00:04:44,890
So we're gonna start off with this we're going to first and foremost just go straight through this to

71
00:04:44,890 --> 00:04:46,980
begin with and then we're gonna break it down.

72
00:04:47,140 --> 00:04:51,970
So we'll have an end to end psychic learn workflow and we'll look at the step one getting data ready

73
00:04:52,270 --> 00:04:56,740
prepare it to be used with our machine learning models we'll see how to choose a machine learning model

74
00:04:56,740 --> 00:05:02,050
for what problem we're working on we'll learn how to fit a model to the data a.k.a. learning patterns

75
00:05:02,380 --> 00:05:06,910
we'll learn how to make predictions with a machine learning model okay using those patterns that it's

76
00:05:06,910 --> 00:05:10,600
learned and then we'll learn how to evaluate those predictions.

77
00:05:10,600 --> 00:05:15,270
So evaluating whether a machine learning model is learning patterns that are worthwhile.

78
00:05:15,310 --> 00:05:21,460
We'll see how to improve said predictions and then we'll finally finish up with saving and reloading

79
00:05:21,520 --> 00:05:26,790
models another important question where can you get help.

80
00:05:27,540 --> 00:05:32,670
Well as always follow along as best you can with the code we'll be going hands on in a Jupiter notebook.

81
00:05:32,670 --> 00:05:36,420
So if you can write the code if in doubt run the code.

82
00:05:36,420 --> 00:05:37,410
This is part of this step.

83
00:05:37,410 --> 00:05:38,730
Try it for yourself.

84
00:05:38,730 --> 00:05:43,740
Remember our motto if in doubt run the code if you want to find out more about what a particular function

85
00:05:43,740 --> 00:05:44,510
is doing.

86
00:05:44,520 --> 00:05:47,720
Don't forget to press shift in tab to read the doctoring.

87
00:05:48,000 --> 00:05:51,750
So when you're in the middle of function we'll see this in play when we're going through the notebook.

88
00:05:51,750 --> 00:05:52,830
You can see the doctoring.

89
00:05:52,830 --> 00:05:56,640
This will take you straight to the documentation if you're still stuck.

90
00:05:56,640 --> 00:05:57,510
You can search for it.

91
00:05:57,510 --> 00:06:03,300
And when you do search for a psychic line function such as how to evaluate a psychic line classification

92
00:06:03,300 --> 00:06:08,530
model you'll probably come along resources such a stack overflow or the user guide.

93
00:06:08,550 --> 00:06:11,280
And remember these resources the library is vast.

94
00:06:11,310 --> 00:06:13,740
So just be patient with searching for it.

95
00:06:13,740 --> 00:06:16,130
It does take me a while to figure things out too.

96
00:06:16,200 --> 00:06:17,250
I'm not perfect.

97
00:06:17,250 --> 00:06:20,930
It is okay to look something up and and try and figure it out for yourself.

98
00:06:21,090 --> 00:06:22,460
And then once you have search for it.

99
00:06:22,470 --> 00:06:23,610
If you do find something.

100
00:06:23,700 --> 00:06:24,600
Try again.

101
00:06:24,600 --> 00:06:27,710
And if you're still stuck finally it's okay to ask.

102
00:06:27,720 --> 00:06:30,260
Never be afraid to ask a question.

103
00:06:30,420 --> 00:06:32,580
So you exile as I am.

104
00:06:32,580 --> 00:06:37,560
Well if you are let's get into it let's get hands on let's build some machine learning models and let's

105
00:06:37,560 --> 00:06:40,590
get our cells familiar with psychic learn.
