1 00:00:01,480 --> 00:00:04,600 Hey, everyone, and welcome to Time series Analysis. 2 00:00:04,630 --> 00:00:10,960 This is the best and most in-depth online resource for learning about how to use Python to analyze time 3 00:00:10,960 --> 00:00:11,830 series data. 4 00:00:12,460 --> 00:00:17,050 In this video, I'm going to give you a brief outline for what you will learn in this course. 5 00:00:17,380 --> 00:00:22,000 What's interesting about this course is that pretty much every machine learning technique I have ever 6 00:00:22,000 --> 00:00:25,270 taught can be applied to time series analysis. 7 00:00:25,480 --> 00:00:31,150 Not only will we get a chance to study classic time series techniques, but we will also see how modern 8 00:00:31,150 --> 00:00:33,640 machine learning and deep learning can be applied. 9 00:00:35,950 --> 00:00:39,340 So let's start with a very high level overview of the course. 10 00:00:40,060 --> 00:00:42,640 This course can be broken up into three parts. 11 00:00:42,910 --> 00:00:46,720 Part one covers classic methods like Exponential Smoothing and Arima. 12 00:00:46,990 --> 00:00:50,260 Part two covers modern machine learning and deep Learning. 13 00:00:50,800 --> 00:00:56,590 Part three covers VIP content, which will only be available in the VIP version of the course. 14 00:00:57,190 --> 00:01:01,960 Note that each of these parts can really be done in any order you please, so if you know what you're 15 00:01:01,960 --> 00:01:05,620 doing, don't hesitate to skip to the lectures you care about most. 16 00:01:07,060 --> 00:01:12,650 So let's go into a little more detail before we even go into time series modeling. 17 00:01:12,670 --> 00:01:16,660 We're going to have an entire section devoted to time series basics. 18 00:01:16,930 --> 00:01:20,770 So if you've never worked with a time series before, this is stuff you need to know. 19 00:01:21,340 --> 00:01:24,930 For example, what kind of data can be considered a time series? 20 00:01:24,940 --> 00:01:28,540 What kind of data transformations are useful for time series? 21 00:01:28,570 --> 00:01:30,560 How do we evaluate a forecast? 22 00:01:30,580 --> 00:01:32,590 What kind of metrics do we use? 23 00:01:32,920 --> 00:01:38,200 And because I know a lot of you are interested in Financial Time series, that will be a major focus 24 00:01:38,200 --> 00:01:39,220 of this course. 25 00:01:39,460 --> 00:01:44,290 We'll go through a Financial Time series Primer, where you will be introduced to important concepts 26 00:01:44,290 --> 00:01:48,520 such as log prices, log returns, and the random hypothesis. 27 00:01:50,700 --> 00:01:56,160 The first classical time series method we will discuss is called Exponential Smoothing or ETS. 28 00:01:56,460 --> 00:02:01,890 This kind of model is ideal for time series, which are made up of error, trend and seasonal components. 29 00:02:02,190 --> 00:02:06,030 We'll apply this method to both sales data and financial time series. 30 00:02:07,740 --> 00:02:11,100 The second classical method we will discuss is called ARIMA. 31 00:02:11,460 --> 00:02:15,480 This kind of model more closely resembles what you would see in a machine learning. 32 00:02:16,290 --> 00:02:21,840 Now, although a lot of theoretical analysis is possible, this course focuses on the practical aspects 33 00:02:21,840 --> 00:02:22,950 of this technique. 34 00:02:23,100 --> 00:02:26,520 For example, how to choose the arena orders, P, D, and Q. 35 00:02:27,420 --> 00:02:32,910 We'll learn about tools such as the ACF, F and Auto Arima to help us do this. 36 00:02:33,240 --> 00:02:38,160 We'll also look at vector models and see whether or not the interactions between multiple variables 37 00:02:38,160 --> 00:02:40,920 in a time series can help us predict the future. 38 00:02:42,700 --> 00:02:48,010 So after we learn about the classic Time series methods, we will then move on to machine learning. 39 00:02:48,490 --> 00:02:52,840 In this section, we'll see how machine learning models like logistic regression, the support vector 40 00:02:52,840 --> 00:02:56,890 machine and the random forest can be used for time series analysis. 41 00:02:57,340 --> 00:02:59,740 Thanks to the flexibility of machine learning. 42 00:02:59,740 --> 00:03:03,730 Most of these models will let us do both forecasting and classification. 43 00:03:04,450 --> 00:03:09,850 As an example, suppose you want to predict what a user is doing based on readings from their smartphone 44 00:03:09,850 --> 00:03:10,780 accelerometer. 45 00:03:13,030 --> 00:03:18,340 So after discussing machine learning, we'll discuss the latest and greatest machine learning approach, 46 00:03:18,340 --> 00:03:23,800 deep neural networks and these sections we'll look at and CNNs CNN's and origins. 47 00:03:23,920 --> 00:03:28,420 And yes, because everyone loves LCMS, this will include LCMS. 48 00:03:29,620 --> 00:03:34,750 Although you may have seen me teach these models before for different kinds of data, these can all 49 00:03:34,750 --> 00:03:36,820 be used for time series analysis. 50 00:03:37,030 --> 00:03:40,930 And if you think you know everything there is to know, because you've taken my deep learning courses 51 00:03:40,930 --> 00:03:43,540 in the past, you will be surprised. 52 00:03:45,550 --> 00:03:50,560 So if you've signed up for the VIP version of this course, you're going to get access to some very 53 00:03:50,560 --> 00:03:51,670 exciting material. 54 00:03:52,540 --> 00:03:56,380 So one of the VIP sections we'll look at of us forecast. 55 00:03:56,440 --> 00:04:01,840 This will let you use Amazon's state of the art forecasting methods with barely any code. 56 00:04:02,200 --> 00:04:07,000 In fact, you do not even have to understand time series analysis in order to use this. 57 00:04:07,330 --> 00:04:12,860 As long as you know how to use Amazon's APIs and you can get your data into CSV format. 58 00:04:12,880 --> 00:04:15,370 You can have access to this powerful technology. 59 00:04:16,270 --> 00:04:20,950 Amazon forecast also performs AutoML and Hyper parameter optimization. 60 00:04:20,950 --> 00:04:26,110 So all of the grunt work typically involved in machine learning is taken off your hands. 61 00:04:27,970 --> 00:04:29,260 Another VIP section. 62 00:04:29,260 --> 00:04:30,700 We'll look at the garden method. 63 00:04:30,970 --> 00:04:36,700 So this is a very important model for Financial Time series, because not only do we like to model the 64 00:04:36,700 --> 00:04:41,830 price over time, but also the volatility or variance, which is a measure of risk. 65 00:04:43,830 --> 00:04:48,570 Another VIP section will look at Prophet Facebook's famous time series tool. 66 00:04:48,900 --> 00:04:53,910 Nothing beats a free tool made by one of the largest tech companies in the world that specializes in 67 00:04:53,910 --> 00:04:54,840 machine learning. 68 00:04:55,500 --> 00:04:58,860 And yes, even more surprise VIP material is coming. 69 00:04:58,860 --> 00:05:00,570 So stay tuned for updates. 70 00:05:02,450 --> 00:05:08,870 Okay, So again, part one classical methods, Part two Machine learning and deep neural networks and 71 00:05:08,870 --> 00:05:10,970 part three VIP content. 72 00:05:11,330 --> 00:05:15,530 I hope you're excited for this course because I had such a fun time at putting it together. 73 00:05:15,830 --> 00:05:18,320 Thanks for listening and I'll see you in the next lecture.