1 00:00:11,030 --> 00:00:16,370 So in this lecture, we are going to discuss how you, as a beginner, should approach the coding lectures 2 00:00:16,370 --> 00:00:17,330 in this course. 3 00:00:18,230 --> 00:00:23,240 The reason I've created this lecture is that I've noticed some beginner students are a bit confused 4 00:00:23,240 --> 00:00:25,970 about the approach we take to the code lectures. 5 00:00:26,480 --> 00:00:31,580 Specifically, they might be wondering why I'm not coding live on the screen or why they can't code 6 00:00:31,580 --> 00:00:32,060 along. 7 00:00:32,750 --> 00:00:37,520 What I hope you'll see by the end of this lecture is that these questions are just misinformed. 8 00:00:37,940 --> 00:00:41,870 Any confusion you may have is based on incorrect assumptions. 9 00:00:42,800 --> 00:00:49,100 Specifically, this lecture is going to answer a few questions, namely no one should you code along. 10 00:00:49,400 --> 00:00:50,390 The answer is no. 11 00:00:51,050 --> 00:00:54,260 Number two, can you call along if you really want to? 12 00:00:54,650 --> 00:00:55,670 The answer is yes. 13 00:00:56,150 --> 00:00:59,540 Number three, how can you practice what you've learned in this course? 14 00:00:59,780 --> 00:01:02,060 And what are the exercises in this course? 15 00:01:02,660 --> 00:01:07,070 So if you had any or perhaps all of these questions, this lecture is for you. 16 00:01:11,920 --> 00:01:17,800 OK, so the first question we're going to tackle is, should you call along, as stated in my recommendation 17 00:01:17,800 --> 00:01:18,400 is no. 18 00:01:19,090 --> 00:01:24,190 This may surprise some of you, especially if you are a beginner or if you've taken other beginner courses 19 00:01:24,190 --> 00:01:24,850 online. 20 00:01:25,480 --> 00:01:29,950 In those courses, coding along is typically encouraged from my experience. 21 00:01:30,430 --> 00:01:32,590 In fact, I'm not immune to this myself. 22 00:01:33,070 --> 00:01:37,930 If you take some of my oldest courses, which came out approximately five years before this course, 23 00:01:38,320 --> 00:01:43,420 I also encourage students to code along because I assume that this was the normal thing to do. 24 00:01:44,140 --> 00:01:48,880 Of course, after five years of teaching, I've realized why this is not a good idea at all. 25 00:01:49,390 --> 00:01:54,760 In fact, this live coding approach is not good for either of us, neither you nor myself. 26 00:01:59,390 --> 00:02:02,300 So let me start with why coding along is bad for you. 27 00:02:03,840 --> 00:02:07,710 Firstly, it gives you an extremely false impression of how this code is written. 28 00:02:08,520 --> 00:02:13,920 This is not how code is written in the real world, especially complex code like the code in this course. 29 00:02:14,490 --> 00:02:19,350 Doing things by going along with the instructor makes sense in other areas like playing guitar or doing 30 00:02:19,350 --> 00:02:21,870 yoga, but it makes no sense in coding. 31 00:02:22,530 --> 00:02:25,230 Unlike guitar and yoga, coding is a mental thing. 32 00:02:25,380 --> 00:02:27,630 It's about what goes on inside your mind. 33 00:02:28,140 --> 00:02:30,150 Guitar and yoga are physical things. 34 00:02:30,150 --> 00:02:32,070 They are about what you can do with your body. 35 00:02:33,030 --> 00:02:38,840 A coding lecture where the instructor types everything from start to end might take maybe 10 or 20 minutes, 36 00:02:39,270 --> 00:02:42,630 but this is only because the code was prepared beforehand. 37 00:02:43,110 --> 00:02:48,840 In reality, this code required thinking in order to design the structure research in order to learn 38 00:02:48,840 --> 00:02:53,490 the syntax for any APIs that were used and debugging to fix mistakes. 39 00:02:55,010 --> 00:02:59,930 This does not teach you how to write the code from scratch because this code was not written from scratch 40 00:03:00,140 --> 00:03:01,700 as you were seeing it typed lot. 41 00:03:02,540 --> 00:03:07,550 And by the way, if anyone tells you they write code without making mistakes, they are not only lying 42 00:03:07,550 --> 00:03:11,840 but also too inexperienced to even know that what they are saying is impossible. 43 00:03:12,530 --> 00:03:17,300 In effect, all you are really doing is wasting time watching the instructor type in the screen. 44 00:03:22,030 --> 00:03:27,880 What you should be paying more attention to is the thinking process behind the coding, that is to say 45 00:03:27,880 --> 00:03:30,850 coding is more about thinking, not about typing. 46 00:03:31,510 --> 00:03:35,290 Coding complex problems does not happen linearly from start to end. 47 00:03:35,950 --> 00:03:41,860 One example of that is test driven development, also known as TDD, which is a common process used 48 00:03:41,860 --> 00:03:45,550 in modern software engineering companies such as Google and Facebook. 49 00:03:46,570 --> 00:03:52,720 This process is used with the expectation that you will make mistakes, and thus it is much more realistic. 50 00:03:53,680 --> 00:03:57,190 So let's reiterate this coding lie from start to end. 51 00:03:57,220 --> 00:04:00,730 Not realistic, test driven development. 52 00:04:01,120 --> 00:04:05,530 This is a real coding process that is used by professional companies around the world. 53 00:04:06,130 --> 00:04:10,990 So instead of mindlessly watching someone type on the screen and believing that anyone actually codes 54 00:04:10,990 --> 00:04:14,410 anything this way, let's review what you should be doing instead. 55 00:04:15,040 --> 00:04:18,490 Firstly, you should be thinking about the overall structure of the code. 56 00:04:18,820 --> 00:04:23,620 As I describe it to you, and as I describe the reasons why it is structured the way it is. 57 00:04:24,310 --> 00:04:27,190 Secondly, you should be paying attention to the syntax. 58 00:04:27,730 --> 00:04:32,770 And of course, without knowing the syntax, you can write any of this code yourself, which is presumably 59 00:04:32,770 --> 00:04:34,210 what you want to do in the future. 60 00:04:38,810 --> 00:04:43,700 So now let's enter the other side of this question, which is why coding live is bad for me. 61 00:04:44,240 --> 00:04:46,700 We've just established why it's not going to help you. 62 00:04:48,050 --> 00:04:49,280 Well, it's pretty simple. 63 00:04:49,670 --> 00:04:54,110 If I'm trying to type and talk at the same time, then I'm trying to do two things at once. 64 00:04:54,590 --> 00:04:57,980 But if I'm only talking, then I can focus on that one thing. 65 00:04:58,250 --> 00:05:03,260 And therefore, I can go into much greater detail and be much more precise in what I say. 66 00:05:03,830 --> 00:05:08,210 This ends up being good for you because it means you're getting better information transfer. 67 00:05:08,540 --> 00:05:13,250 And what I'm saying will be much smoother than if I tried to come up with it while typing at the same 68 00:05:13,250 --> 00:05:13,790 time. 69 00:05:14,660 --> 00:05:20,420 So not only is typing pointless as we previously established, it's also better to not type because 70 00:05:20,420 --> 00:05:22,730 it helps me communicate better with you. 71 00:05:27,380 --> 00:05:32,540 OK, so now let's move on to question number two, which is, can you quote along if you really want 72 00:05:32,540 --> 00:05:34,550 to disregard my previous advice? 73 00:05:35,060 --> 00:05:36,770 The answer to this question is yes. 74 00:05:37,430 --> 00:05:41,360 Now, before some student comes along and asks, Why do you even have to say this? 75 00:05:41,360 --> 00:05:42,500 Isn't this obvious? 76 00:05:42,950 --> 00:05:47,990 Let me just shut that down right now and tell you for some students, it's not obvious, given that 77 00:05:47,990 --> 00:05:49,940 I've been asked this question a few times. 78 00:05:50,480 --> 00:05:55,790 You see, I think these students may have been confused, perhaps assuming that because I wasn't typing 79 00:05:55,790 --> 00:05:57,440 that they also could not type. 80 00:05:57,950 --> 00:06:01,910 Of course, now that I'm saying this out loud, you should realize this is complete nonsense. 81 00:06:02,300 --> 00:06:04,670 For example, I can show you this piece of code. 82 00:06:05,060 --> 00:06:07,460 You can type this out right now if you like. 83 00:06:07,820 --> 00:06:09,980 I don't have to type it for you to type it. 84 00:06:10,550 --> 00:06:15,200 There is no requirement that I'm actually typing on the keyboard for you to start typing. 85 00:06:15,890 --> 00:06:18,170 Again, this is not like yoga for yoga. 86 00:06:18,200 --> 00:06:20,870 Yes, you want to see your teacher doing the moves. 87 00:06:21,410 --> 00:06:26,390 Now again, I know this is obvious, but some students really have asked me this question, so please 88 00:06:26,390 --> 00:06:27,080 bear with me. 89 00:06:31,670 --> 00:06:35,780 OK, so now let's move on to question number three, which I've also discussed elsewhere, but I'll 90 00:06:35,780 --> 00:06:37,370 answer again here for good measure. 91 00:06:38,060 --> 00:06:43,220 And this is how can you practice what you've learned in this course, given that I've now deprived you 92 00:06:43,220 --> 00:06:45,860 of this interactive exercise of coding a law? 93 00:06:46,670 --> 00:06:50,780 Well, let's remind ourselves that coding along is not actually a good exercise. 94 00:06:51,290 --> 00:06:55,730 Remember, you were not learning anything about the coding process when you do that, because that's 95 00:06:55,730 --> 00:06:57,980 not even how the coding process was done. 96 00:06:58,580 --> 00:07:04,400 If the coding lecture was 10 minutes, the coding process probably took maybe 60 minutes, maybe even 97 00:07:04,400 --> 00:07:05,120 days. 98 00:07:05,630 --> 00:07:10,790 So that should be the time commitment you're looking at when you do an exercise, at least at this level 99 00:07:10,790 --> 00:07:11,720 of machine learning. 100 00:07:12,590 --> 00:07:17,510 Now, to really answer this question, we have to go a bit meta, and that's to discuss the two kinds 101 00:07:17,510 --> 00:07:19,820 of courses that can exist in machine learning. 102 00:07:24,570 --> 00:07:29,370 The first kind, of course, is the theory course, where I teach you, say, the naive Bayes algorithm 103 00:07:29,790 --> 00:07:35,880 and then you go and implement that algorithm in code in that kind, of course, implementing the algorithm 104 00:07:35,880 --> 00:07:39,900 is the exercise, and that's a really powerful skill to have. 105 00:07:40,380 --> 00:07:46,500 That means you can take an idea and make that idea practical by turning it into a working computer program. 106 00:07:47,160 --> 00:07:51,660 And of course, anyone who invents a new machine learning models must have this skill. 107 00:07:53,040 --> 00:07:58,290 By definition, if you invent a new machine learning algorithm, then there is no existing code for 108 00:07:58,290 --> 00:07:58,650 it. 109 00:07:59,040 --> 00:08:02,040 You will be the one who makes the first working example. 110 00:08:02,460 --> 00:08:05,180 And if you can't code, then it will never be made. 111 00:08:06,360 --> 00:08:09,960 So in this case, the concept of exercise is very straightforward. 112 00:08:10,380 --> 00:08:15,750 You're given an idea, an algorithm, and your exercise is to put that into working code. 113 00:08:16,350 --> 00:08:21,750 And obviously, you should do this before looking at my code, which is the solution to the exercise 114 00:08:22,170 --> 00:08:27,510 that is, don't look at the solution to the exercise before doing the exercise yourself. 115 00:08:32,140 --> 00:08:36,340 Now, that doesn't really apply to this course, because this is not a theory course. 116 00:08:36,760 --> 00:08:38,650 This course is an API course. 117 00:08:38,860 --> 00:08:44,140 I'm teaching you how to use an existing library where your goal is mostly to learn about the syntax 118 00:08:44,410 --> 00:08:45,790 rather than the concepts. 119 00:08:46,240 --> 00:08:51,250 Although we do still touch on concepts in this class, we're not going to do something like derive back 120 00:08:51,250 --> 00:08:55,180 propagation because that's irrelevant to the library we're learning about. 121 00:08:55,750 --> 00:09:01,210 So in this type of class, it would be impossible for you to write out any code before I show it to 122 00:09:01,210 --> 00:09:07,270 you, because obviously you simply won't know the syntax in this type of class. 123 00:09:07,540 --> 00:09:10,210 There are two kinds of exercises and projects. 124 00:09:10,840 --> 00:09:13,930 The first kind is the kind I'm going to explicitly give you. 125 00:09:14,410 --> 00:09:17,740 That is, you'll hear me say in the lecture, Please do this exercise. 126 00:09:18,100 --> 00:09:23,440 And after hearing there, it will be your choice, whether or not you want to go ahead and do that exercise 127 00:09:24,070 --> 00:09:24,940 as a side note. 128 00:09:24,970 --> 00:09:30,730 Please note that for this class, the explicit exercise prompts and data sets are included only in the 129 00:09:30,730 --> 00:09:31,720 VIP version. 130 00:09:32,620 --> 00:09:37,960 If you don't know what that means, please contact me using the contact form on my website, which is 131 00:09:37,960 --> 00:09:39,760 that lazy programmer taught me. 132 00:09:44,450 --> 00:09:50,510 Now, the second kind of exercise is implicit, that is, although I'm not going to explicitly tell 133 00:09:50,510 --> 00:09:54,110 you to do something, it should be obvious that you would do it anyway. 134 00:09:55,130 --> 00:09:58,430 My assumption is that we're all responsible, mature adults. 135 00:09:58,850 --> 00:10:02,900 You know why you're in this course and you know why you want to learn about machine learning. 136 00:10:03,350 --> 00:10:07,490 And that's probably because you want to apply it to some data set you have. 137 00:10:08,060 --> 00:10:14,210 For example, if you work in manufacturing, perhaps you're interested in using computer vision to identify 138 00:10:14,210 --> 00:10:15,380 defective parts. 139 00:10:16,280 --> 00:10:21,680 Well, after taking this course, you should be able to apply the code I've shown you to your own data 140 00:10:21,680 --> 00:10:22,250 sets. 141 00:10:23,150 --> 00:10:25,430 Now, here's the really neat thing about machine learning. 142 00:10:26,030 --> 00:10:28,730 This process does not require any new code. 143 00:10:29,240 --> 00:10:32,840 Therefore, you're going to simply be applying the code you've already learned. 144 00:10:33,320 --> 00:10:36,050 This will give you practice in writing that code yourself. 145 00:10:36,740 --> 00:10:40,550 But again, this code is more or less the same as what you've learned in this course. 146 00:10:41,150 --> 00:10:42,800 I encapsulate this with my rule. 147 00:10:42,800 --> 00:10:44,090 All data is the same. 148 00:10:44,810 --> 00:10:48,710 What this means is that the same code will work no matter the data sets. 149 00:10:53,290 --> 00:10:55,210 For example, take a neural network. 150 00:10:55,630 --> 00:10:58,810 We know that neural networks can work for something like fraud detection. 151 00:10:59,560 --> 00:11:04,360 Well, now suppose that I want to use a neural network for a new task like disease prediction. 152 00:11:05,080 --> 00:11:08,260 Surprisingly, this does not require any new code at all. 153 00:11:08,770 --> 00:11:12,010 All that's left to do is to simply plug in your data set. 154 00:11:12,490 --> 00:11:17,680 And this is why we can have libraries like Circuit Learn, which is used by researchers in many fields 155 00:11:17,770 --> 00:11:20,680 like climate prediction, biology and finance. 156 00:11:21,310 --> 00:11:25,180 The data in biology is completely different from the data in finance. 157 00:11:25,510 --> 00:11:29,920 And yet the machine learning code to process this data is exactly the same. 158 00:11:30,580 --> 00:11:35,770 And this is what we mean when we say all data is the same in the eyes of your machine learning model, 159 00:11:36,520 --> 00:11:38,980 and that is how you will practice what you've learned. 160 00:11:39,520 --> 00:11:44,770 Although you can't come up with the syntax for an API out of thin air, what you can do is internalize 161 00:11:44,770 --> 00:11:50,650 it by rewriting it again and again on new datasets, which are not just the generic data sets from this 162 00:11:50,650 --> 00:11:54,040 course, but data sets you really care about in your life. 163 00:11:54,970 --> 00:11:59,410 Furthermore, you'll need to test different hyper parameters on your own, which will force you to think 164 00:11:59,410 --> 00:12:01,480 about things like choosing the best learning rate. 165 00:12:01,720 --> 00:12:04,030 Finding the right number of layers and so forth. 166 00:12:05,230 --> 00:12:10,660 Now, of course, not all of you already have jobs where you have access to data sets that you want 167 00:12:10,660 --> 00:12:11,350 to use. 168 00:12:11,890 --> 00:12:17,080 In this case, you can still apply this approach by choosing any dataset you want to use. 169 00:12:17,680 --> 00:12:21,130 The beauty of this is you get to choose what you care about. 170 00:12:21,610 --> 00:12:25,630 I can't tell you what that is, but I'll trust that you can figure this out on your own. 171 00:12:26,440 --> 00:12:28,480 And so let me just make this explicit. 172 00:12:28,960 --> 00:12:31,660 You shouldn't depend on me to tell you to do this. 173 00:12:31,990 --> 00:12:38,020 This is just the obvious thing to do if you know why you're here and you're a productive and proactive 174 00:12:38,020 --> 00:12:43,210 student who is serious about learning the subject, then this is simply the obvious course of action. 175 00:12:47,840 --> 00:12:53,600 Now I want to end this lecture with a question for you, the students of this course, over the years, 176 00:12:53,600 --> 00:12:58,460 I've seen students get triggered by this lecture, meaning that they got really angry or offended by 177 00:12:58,460 --> 00:12:59,420 the things I've said. 178 00:12:59,930 --> 00:13:02,900 Now, personally, I don't understand why that's the case. 179 00:13:03,290 --> 00:13:08,570 So if that was the case for you or if you disagreed with anything I said in this lecture, please let 180 00:13:08,570 --> 00:13:10,250 me know about it on the Q&A. 181 00:13:10,730 --> 00:13:15,620 I'm always available to listen to your concerns, so the sooner you can let me know, the sooner they 182 00:13:15,620 --> 00:13:16,430 will be dealt with.