1 00:00:02,130 --> 00:00:07,140 Once we've studied art ends, we can go even deeper into how deep learning can be used on text. 2 00:00:07,590 --> 00:00:13,320 That's the next course natural language processing with deep learning in Python or deep learning in 3 00:00:13,320 --> 00:00:14,490 Python Part six. 4 00:00:15,470 --> 00:00:19,070 In this course, we look at a very important concepts called the word embeddings. 5 00:00:19,640 --> 00:00:25,280 These allow us to turn words which are categorical variables into vectors, which are numbers that a 6 00:00:25,280 --> 00:00:26,390 neural network can read. 7 00:00:27,230 --> 00:00:31,990 This is why it's very important to have studied unsupervised learning before this, because finding 8 00:00:32,000 --> 00:00:34,640 word embeddings is actually an unsupervised task. 9 00:00:35,330 --> 00:00:39,950 Word embeddings also allow us to make use of pre-training, which was discussed in deep learning in 10 00:00:39,950 --> 00:00:41,000 Python Part four. 11 00:00:42,020 --> 00:00:47,480 In this course, we also look at a very advanced model for doing sentiment analysis called a recursive 12 00:00:47,480 --> 00:00:49,430 neural network or a tree neural network. 13 00:00:50,180 --> 00:00:55,550 This is an example of a dynamic neural network because it changes its structure based on what input 14 00:00:55,550 --> 00:00:56,060 you give it. 15 00:00:56,900 --> 00:01:01,820 Most deep learning libraries are not equipped to handle dynamic neural networks, and I demonstrate 16 00:01:02,240 --> 00:01:07,370 what happens if you try to build one the naive way you basically and the building a separate neural 17 00:01:07,370 --> 00:01:12,200 network for each of your training samples, and it's going to eat up all your RAM and make your computer 18 00:01:12,200 --> 00:01:13,340 slow down to a crawl. 19 00:01:14,180 --> 00:01:19,280 So you get a firsthand perspective of why working with dynamic neural networks is not easy. 20 00:01:20,030 --> 00:01:24,430 But luckily, we are able to make use of our knowledge of recurrent neural networks. 21 00:01:24,440 --> 00:01:30,830 And remember, that's the prerequisite to this course in order to convert a tree into a sequence which 22 00:01:30,830 --> 00:01:32,810 Noonan is capable of handling. 23 00:01:33,380 --> 00:01:36,320 So that's why CNN's is a prerequisite to this course. 24 00:01:38,280 --> 00:01:42,450 Now you'll notice that this course deep in OP, has no outgoing edges. 25 00:01:42,480 --> 00:01:45,750 This is because it's the most advanced course that I have on this path. 26 00:01:45,840 --> 00:01:49,740 For now, I certainly plan on expanding in this direction in the future. 27 00:01:50,490 --> 00:01:54,150 However, this is not the last deep learning course in the series. 28 00:01:54,810 --> 00:01:59,910 After this, we have deep reinforcement learning, which is deep learning in Python Part seven. 29 00:02:00,660 --> 00:02:05,400 With that said, I think now is a good time to go back and start exploring the reinforcement learning 30 00:02:05,400 --> 00:02:05,760 path. 31 00:02:06,990 --> 00:02:11,190 So you'll notice that I have Bayesian machine learning feeding into reinforcement learning. 32 00:02:12,610 --> 00:02:17,710 On the surface, these two courses might seem unrelated, but there is a very important concept you'll 33 00:02:17,710 --> 00:02:23,320 learn that applies to both called the Explore exploit dilemma in Bayesian machine learning. 34 00:02:23,330 --> 00:02:28,450 You learn this idea in the context of trying to optimize it, click through rate or a conversion rate, 35 00:02:28,990 --> 00:02:33,820 or, in other words, the number of times people buy things from your website versus the number of times 36 00:02:33,820 --> 00:02:35,170 people visit at your web site. 37 00:02:35,650 --> 00:02:36,850 Very practical concept. 38 00:02:36,850 --> 00:02:43,000 I think if you do anything related to e-commerce now in reinforcement learning, we look at the Explore 39 00:02:43,000 --> 00:02:44,370 exploit dilemma again. 40 00:02:44,380 --> 00:02:49,990 But in the context of playing games, reinforcement learning is like a third branch of machine learning, 41 00:02:49,990 --> 00:02:53,590 whereas the other two are supervised and unsupervised learning. 42 00:02:54,220 --> 00:02:59,980 The main difference is that supervised and unsupervised learning look at static data and reinforcement 43 00:02:59,980 --> 00:03:00,340 learning. 44 00:03:00,350 --> 00:03:04,270 The idea is more like you have a robot living in the real world. 45 00:03:04,540 --> 00:03:10,330 It can take the experiences it had today and based on them, behave more intelligently tomorrow. 46 00:03:10,840 --> 00:03:14,020 So the learning paradigm in reinforcement learning is sequential. 47 00:03:14,590 --> 00:03:20,260 This is opposed to supervised, unsupervised learning, where your dataset usually resides in some file 48 00:03:20,260 --> 00:03:21,190 on your hard drive. 49 00:03:22,370 --> 00:03:26,480 So in reinforcement learning, part one, we get all the basics, as you might expect. 50 00:03:27,020 --> 00:03:32,510 This is a prerequisite to deep reinforcement learning, since deep reinforcement learning applies those 51 00:03:32,510 --> 00:03:34,430 concepts to more difficult games. 52 00:03:35,210 --> 00:03:40,790 But you'll notice that this is not the only prerequisite to deep reinforcement learning, as the title 53 00:03:40,790 --> 00:03:41,400 suggests. 54 00:03:41,420 --> 00:03:47,690 This is also dependent on knowledge of deep learning, in particular convolutional neural networks. 55 00:03:49,550 --> 00:03:54,920 But as we know, in order to build a CNN, we have to know how to build a regular neural network, which 56 00:03:54,920 --> 00:03:58,370 means we have to know what a neural network is and why it's useful and so on. 57 00:03:58,970 --> 00:04:02,300 So in deep reinforcement learning, these two paths converge. 58 00:04:02,750 --> 00:04:07,580 You combine your knowledge of both reinforcement learning and deep learning in this course. 59 00:04:08,920 --> 00:04:16,540 Now, the reason it depends on CNN's and not say aren't ends, which is over here is because we'll be 60 00:04:16,540 --> 00:04:18,230 learning to play visual games. 61 00:04:18,250 --> 00:04:24,130 So for example, we can learn how to play a video game like Pong, a breakout which are classic Atari 62 00:04:24,130 --> 00:04:25,300 games from the old days. 63 00:04:25,690 --> 00:04:29,830 And so those are images because they're basically screenshots from the screen. 64 00:04:30,790 --> 00:04:36,550 In the future, we might end up applying Ardennes, in which case Ardennes will become a prerequisite 65 00:04:36,550 --> 00:04:37,270 to that course. 66 00:04:39,080 --> 00:04:41,900 So that's the end of the reinforcement learning path for now. 67 00:04:42,140 --> 00:04:45,860 I'm very excited to bring you more updates in this area in the future. 68 00:04:48,310 --> 00:04:50,850 Let's now jump back to logistic regression. 69 00:04:52,020 --> 00:04:56,070 Where we can see another outgoing age to supervised machine learning. 70 00:04:57,270 --> 00:04:58,620 So why is this edge here? 71 00:04:59,490 --> 00:05:04,590 Well, you may recall that linear regression and logistic regression are both linear models that do 72 00:05:04,590 --> 00:05:06,870 regression and classification, respectively. 73 00:05:07,530 --> 00:05:09,660 These are both supervised learning tasks. 74 00:05:11,760 --> 00:05:17,100 And so it makes sense that now that you know, one model for regression and one model for classification, 75 00:05:17,580 --> 00:05:20,070 it's time to dig deeper into supervised learning. 76 00:05:20,910 --> 00:05:25,700 The thing with linear regression and logistic regression is that they aren't really different models. 77 00:05:25,710 --> 00:05:29,100 They are both just the line because they do different tasks. 78 00:05:29,100 --> 00:05:32,040 The techniques and interpretations are slightly different, though. 79 00:05:33,150 --> 00:05:37,380 And there are, of course, different models that are not linear models that can do these tasks, and 80 00:05:37,380 --> 00:05:39,360 that's what this course is all about. 81 00:05:40,140 --> 00:05:45,720 In this course, we look at classic supervised machine learning techniques like K Nearest Neighbor Decision 82 00:05:45,720 --> 00:05:48,540 Trees, the perceptron and the Bayes classifier. 83 00:05:49,530 --> 00:05:53,730 Much like how logistic regression was the basic building block of the neural network. 84 00:05:54,210 --> 00:05:58,770 Classic models like decision trees are the basic building block of ensemble methods. 85 00:05:59,810 --> 00:06:04,250 So that's why this course is a prerequisite to ensemble machine learning. 86 00:06:04,940 --> 00:06:10,970 Again, we use the same logo with a different color to signify that these two courses are very closely 87 00:06:10,970 --> 00:06:17,060 related in ensemble machine learning, we learn how to combine multiple decision trees in different 88 00:06:17,060 --> 00:06:20,480 ways in order to make some very powerful classifiers. 89 00:06:21,260 --> 00:06:26,180 What's really remarkable about these methods is that they are very easy to plug and play on data. 90 00:06:26,660 --> 00:06:31,820 So if you're looking for a plug and play solution without having to learn a lot of theory, then deep 91 00:06:31,820 --> 00:06:33,800 learning is most likely not for you. 92 00:06:34,190 --> 00:06:36,110 But ensemble methods are a great fit. 93 00:06:37,060 --> 00:06:41,620 Deep learning is very dependent on hyper parameters, and if you choose incorrectly, your model will 94 00:06:41,620 --> 00:06:42,760 perform very poorly. 95 00:06:43,450 --> 00:06:47,350 Sometimes it requires immense computing power to find good hyper parameters. 96 00:06:47,890 --> 00:06:50,800 This is an active area of research it has not yet solved. 97 00:06:51,670 --> 00:06:55,180 This is why you can implement what you see in a deep learning paper. 98 00:06:55,240 --> 00:07:00,730 But suppose the author left out some seemingly insignificant detail, so you end up having to make an 99 00:07:00,730 --> 00:07:03,400 assumption and then your results end up totally different. 100 00:07:03,850 --> 00:07:08,050 So deep learning is fragile, but luckily ensemble methods are not. 101 00:07:08,770 --> 00:07:15,370 We focus on two very famous ensemble methods the random forests and Ada Boost, so that's everything 102 00:07:15,370 --> 00:07:17,770 on these supervised machine learning track for now. 103 00:07:19,060 --> 00:07:25,690 Next, we see an edge going from supervised machine learning to unsupervised machine learning, in particular 104 00:07:25,690 --> 00:07:26,830 cluster analysis. 105 00:07:27,700 --> 00:07:32,920 The reason we study supervised learning before unsupervised learning is because unsupervised learning 106 00:07:32,920 --> 00:07:34,300 is a little more abstract. 107 00:07:34,960 --> 00:07:40,270 It takes more effort on the student's part to realize why it's practical and what it can be used for. 108 00:07:41,290 --> 00:07:47,380 Cluster analysis shows us how to model data that does not come with targets, as you might guess, we 109 00:07:47,380 --> 00:07:49,120 do this in the form of clustering. 110 00:07:49,900 --> 00:07:52,060 The idea behind clustering is very simple. 111 00:07:52,600 --> 00:07:57,550 We want to know how many naturally occurring groups of data there are and what are the relationships 112 00:07:57,550 --> 00:07:59,350 between the data in these clusters. 113 00:07:59,800 --> 00:08:05,140 So for example, if you were clustering books, you might find a book about Steve Jobs and a book about 114 00:08:05,140 --> 00:08:06,790 Elon Musk in the same cluster. 115 00:08:07,360 --> 00:08:10,660 This cluster is probably about tech companies in Silicon Valley. 116 00:08:12,610 --> 00:08:17,590 But you don't need a label in your data to tell you that you can discover it yourself by looking at 117 00:08:17,920 --> 00:08:19,810 how the data naturally groups together. 118 00:08:21,220 --> 00:08:26,410 Now, as I mentioned earlier, I think once you learn about both supervised and unsupervised learning, 119 00:08:26,710 --> 00:08:29,260 you'll be ready to jump into reinforcement learning. 120 00:08:30,440 --> 00:08:35,750 I haven't made cluster analysis a prerequisite to reinforcement learning since none of the material 121 00:08:35,750 --> 00:08:40,820 depends on this course, but it's good to know about these techniques so that you have a more mature 122 00:08:40,820 --> 00:08:42,740 and experienced view on machine learning. 123 00:08:46,320 --> 00:08:50,640 What is sort of a sequel to cluster analysis is hidden Markov models. 124 00:08:51,880 --> 00:08:55,300 The reason might not be clearer first, so let me give you two reasons. 125 00:08:56,530 --> 00:08:57,220 Number one. 126 00:08:57,400 --> 00:09:03,130 They're both unsupervised machine learning models, just that H.M. Ms is harder, so it's natural to 127 00:09:03,130 --> 00:09:04,510 learn about clustering first. 128 00:09:05,020 --> 00:09:10,810 Clustering is also about static data, whereas humans are about sequences, so it's similar to the process 129 00:09:10,810 --> 00:09:11,830 we did in deep learning. 130 00:09:11,950 --> 00:09:16,150 We looked at static data like images, then sequential data like text. 131 00:09:17,760 --> 00:09:23,310 Reason number two, in cluster analysis, we learn about a technique called the Gaussian mixture model, 132 00:09:23,400 --> 00:09:25,800 which we make use of in the emem course. 133 00:09:26,550 --> 00:09:33,470 One key point is they both learn by using the expectation maximization algorithm, so it's good to first 134 00:09:33,480 --> 00:09:35,600 see the EMR algorithm on a simple model. 135 00:09:35,610 --> 00:09:41,610 And then when you see em again on a more complicated model like the hmm, it won't be as intimidating. 136 00:09:43,130 --> 00:09:48,920 One key concept you learn in your members is the mark of assumption that just means the current state 137 00:09:48,920 --> 00:09:52,700 depends only on the previous state, but not any states before it. 138 00:09:53,480 --> 00:09:57,380 This is a simplifying assumption that usually makes the math easier to work with. 139 00:09:58,310 --> 00:10:02,520 You will also notice that we encounter the mark of assumption in reinforcement learning. 140 00:10:02,540 --> 00:10:05,360 However, it's not too hard to learn it from scratch. 141 00:10:05,690 --> 00:10:11,000 And so for that reason, I do not consider humans to be a prerequisite to reinforcement learning. 142 00:10:11,690 --> 00:10:14,570 The mark of assumption is really the only thing they have in common. 143 00:10:15,560 --> 00:10:22,190 There is also a slight connection between cluster analysis and unsupervised deep learning, so I'm not 144 00:10:22,190 --> 00:10:27,800 going to draw the link right now, but I sometimes consider this to be unsupervised machine learning 145 00:10:27,800 --> 00:10:31,370 part one and this to be unsupervised machine learning part two. 146 00:10:33,390 --> 00:10:38,340 We also see that each imams feeds into the renowned course, which is about deep learning. 147 00:10:38,610 --> 00:10:39,660 So why might that be? 148 00:10:40,560 --> 00:10:46,110 This is, of course, because both these models are models that can learn about sequences in particular. 149 00:10:46,590 --> 00:10:49,350 In both these courses, we model text as sequences. 150 00:10:50,040 --> 00:10:56,190 But whereas the HMO makes use of the mark of assumption, the Arnon does not enhance the Arnon is a 151 00:10:56,190 --> 00:10:57,300 more powerful model. 152 00:10:58,300 --> 00:11:02,740 And so this just goes along with the main theme that we always go from simple, basic models to more 153 00:11:02,740 --> 00:11:03,730 complex models. 154 00:11:04,390 --> 00:11:06,730 This is also something you should do in your work as well. 155 00:11:07,330 --> 00:11:11,500 If you start with a simple model, you often find that it is faster and more robust. 156 00:11:11,890 --> 00:11:17,410 Complex models sometimes break down, but they are also more difficult to implement and might not even 157 00:11:17,410 --> 00:11:18,280 be fast enough. 158 00:11:18,700 --> 00:11:23,740 Of course, that's just a generalization, so you always want to analyze engineering tradeoffs individually 159 00:11:24,040 --> 00:11:25,260 for every problem you have. 160 00:11:27,410 --> 00:11:30,890 Now there is one last link in this part of the graph here that I want to explain. 161 00:11:31,310 --> 00:11:36,890 And that's the first, you know, because you can see that it depends on supervised machine learning 162 00:11:36,890 --> 00:11:38,840 and feeds into deep NLP. 163 00:11:40,220 --> 00:11:45,980 The main purpose of this basic NLP course is to apply basic machine learning models to text. 164 00:11:46,520 --> 00:11:51,410 So that's why supervised machine learning comes before it is because that this course was all about 165 00:11:51,680 --> 00:11:53,180 basic machine learning models. 166 00:11:54,110 --> 00:12:00,020 The important skills for NLP was not the implementation of those models, but rather a bigger picture 167 00:12:00,020 --> 00:12:02,450 perspective on how machine learning is used. 168 00:12:03,140 --> 00:12:05,660 What is the interface between the data in the model? 169 00:12:06,080 --> 00:12:07,190 What does the model do? 170 00:12:07,790 --> 00:12:09,380 What is its input and output? 171 00:12:09,560 --> 00:12:10,910 How is the output interpreted? 172 00:12:11,720 --> 00:12:17,540 And so we take those principles and we apply them to text, and this way we can see that text can be 173 00:12:17,540 --> 00:12:23,000 created in such a way that you don't have to think about it any differently than any other data. 174 00:12:24,310 --> 00:12:27,280 This reinforces the principle that all data is the same. 175 00:12:27,910 --> 00:12:30,730 The machine learning model doesn't care what your data is. 176 00:12:30,940 --> 00:12:32,770 All it sees is a table of numbers. 177 00:12:32,770 --> 00:12:36,730 It doesn't care if it's text or images or radio signals from space. 178 00:12:37,270 --> 00:12:41,140 The model just does what it was designed to do on the numbers that you give it. 179 00:12:41,800 --> 00:12:46,900 So this course gives you a high level systems perspective on working with machine learning models in 180 00:12:46,900 --> 00:12:47,440 text. 181 00:12:48,280 --> 00:12:54,400 This easy NLP course also feeds into deep NLP, which is, of course, not so easy because it depends 182 00:12:54,400 --> 00:12:56,110 on a lot of background in deep learning. 183 00:12:57,200 --> 00:13:03,230 One of the main questions I get in the NLP course is how do I improve the results of these basic models? 184 00:13:03,860 --> 00:13:08,300 And a lot of the time the answer to that is, well, you have to use a more complex model. 185 00:13:08,300 --> 00:13:12,590 But of course, that necessitates learning how that complex model works. 186 00:13:13,490 --> 00:13:19,430 And deep NLP is an example of that because we learn a state of the art method for sentiment analysis, 187 00:13:19,430 --> 00:13:22,670 whereas in Easy A.P., we used only a linear model. 188 00:13:23,360 --> 00:13:28,550 So it's important to understand that while yes, it's possible to improve the predictive ability on 189 00:13:28,550 --> 00:13:33,140 simple, basic models, as you can see, it's not always an easy path to get there. 190 00:13:33,620 --> 00:13:35,510 So you have to make sure you're prepared. 191 00:13:36,200 --> 00:13:40,370 Case in point, just look at all the time spent just to get to deep in LP. 192 00:13:40,550 --> 00:13:41,780 It's not an easy task. 193 00:13:43,930 --> 00:13:47,500 Let's now scroll over to ganze and variational auto encoders. 194 00:13:48,280 --> 00:13:54,040 Just like how deep reinforcement learning is not related to deep in LP, this isn't really related to 195 00:13:54,040 --> 00:13:55,510 deep reinforcement learning either. 196 00:13:56,140 --> 00:13:59,620 This is deep learning part eight by order of creation only. 197 00:14:00,190 --> 00:14:06,280 It is the spiritual sequel to Unsupervised Deep Learning, which was deep learning in Python Part four. 198 00:14:07,620 --> 00:14:12,960 OK, so just to reiterate, this is part six, part seven and part eight. 199 00:14:13,110 --> 00:14:19,140 By order only, but they are not related to each other conceptually, although it's always nice to know 200 00:14:19,140 --> 00:14:22,830 all these things because more context makes future things easier to learn. 201 00:14:24,230 --> 00:14:32,660 So the reason this is linked to this is because Gans and variational auto and coders are also unsupervised 202 00:14:32,660 --> 00:14:33,650 deep learning models. 203 00:14:34,520 --> 00:14:40,820 But whereas unsupervised deep learning was all about how to improve supervised learning, Gans and variational 204 00:14:40,820 --> 00:14:44,990 auto encoders don't have any direct benefit to supervised learning at all. 205 00:14:45,470 --> 00:14:51,350 Although we do make use of supervised learning within the course, in this course, the focus is on 206 00:14:51,350 --> 00:14:52,640 generating images. 207 00:14:53,420 --> 00:14:58,940 We've seen that Gans can create photorealistic images based on a dual neural network system. 208 00:14:59,630 --> 00:15:03,920 That's pretty cool because before Gans, we didn't have any kind of machine learning model that could 209 00:15:03,920 --> 00:15:05,930 generate real looking images. 210 00:15:06,680 --> 00:15:12,800 Nowadays, Gans are able to generate high resolution, high quality images of people that you can't 211 00:15:12,800 --> 00:15:14,330 even tell are not real people. 212 00:15:14,870 --> 00:15:18,110 It certainly makes the idea of the Matrix seem very possible. 213 00:15:19,830 --> 00:15:22,080 All right, so I hope you found this lecture helpful. 214 00:15:22,620 --> 00:15:27,000 We saw that these courses are related to each other in some pretty complicated ways. 215 00:15:27,840 --> 00:15:30,270 Learning machine learning is not exactly linear. 216 00:15:30,780 --> 00:15:34,110 Sometimes you have to take one course before you can take the next. 217 00:15:34,740 --> 00:15:38,760 Sometimes one course might just be related to another course by some key concepts. 218 00:15:39,120 --> 00:15:42,120 But maybe in one context, it's a lot easier to understand. 219 00:15:42,900 --> 00:15:48,570 So remember to keep in mind that these arrows did not all indicate strong prerequisites, but rather 220 00:15:48,570 --> 00:15:51,390 there's just a relationship between the two courses. 221 00:15:52,140 --> 00:15:57,030 I hope that this lecture explained any nuances between what is a prerequisite and what is not. 222 00:15:57,420 --> 00:16:02,130 And I hope I did a good job of answering which order should you take these courses in and why?