Machine Learning/AI инженер - ПРОФЕССИЯ БУДУЩЕГО!

I propose we start the recording. Hello everyone. Today we have a wonderful evening. We will be exploring the IT profession of Machine Learning as part of our IT Professions Marathon. Today with us is Denis, a Machine Learning specialist, and Evgeny, a manager at PAZV school and a career consultant for professions. We will explain in detail what the profession entails. We will talk about the school, and we will try to fit it into about an hour, maybe a little more. It depends on the number of your questions and your activity. Denis, let's see what's next on the slides. Accordingly, your humble servant Evgeny will be guiding you today on how to feel comfortable in the IT sphere. Next, a little about the school. We were founded in 2016, starting in California. Now all states know about us. Teachers work in the best companies in the USA, Canada, and Europe. This means we teach according to international standards. That is, the programs are developed taking into account the requirements and processes of IT corporations, startups, and various IT-related fields. So, you will undergo training and then be able to apply your knowledge in practice anywhere on the planet. Students undergo internships on real US projects. This means you will have a background, references, and an understanding of how to act within a company, within activities, what to do, and how to behave. It will be a full-fledged internship. I will tell you a little more about it later. We help prepare for interviews and achieve employment. Therefore, in the bottom right corner, we indicate 100% readiness for employment. I will also talk about this in more detail after we introduce you to the field of machine learning. Next. You can find reviews from our graduates on our YouTube channel and also on our website. I will send the information to the chat a little later so you can copy the links, follow them, and familiarize yourself. If you are doubting today whether the IT sphere is for you or not, look at the stories of these people, our graduates. They are truly motivating, with very diverse cases. Some already had IT experience, some did not. Some knew English, some did not. Somewhere you will recognize yourself, somewhere you will think that you are even in a better position than some of our graduates at the start of their IT journey. Therefore, don't be afraid, watch the videos, and visit our website. On the website, you can also sign up for a career consultation. Leave your applications, and a manager will contact you, tell you more about how to get into IT, and how we can be useful to you. We will answer all your questions. Let's go further. Since 2016, we have met not only online but also offline. Our website also has a schedule of workshops. We held several meetings in the first half of the year. Now we are planning the second half. Details will be on our website later, and this requires some preparation. Accordingly, why are offline meetings needed? To get to know each other personally, to create a community of Russian-speaking people in each state, to unite people as much as possible, to understand that you are not alone, that many are interested in the IT sphere, there are various paths, there is an understanding of how to act. Everyone has their own way of getting into IT, their own experience. Accordingly, we explain and show what this field is like. It's clearer to get acquainted in person, to trust, to learn some nuances. And in general, online today, everyone is probably a little tired, so there is an opportunity for live meetings to network, participate in the community, make friends, acquaintances, and continue to be in this field. Graduates work here. These are not just famous global brands, these are truly success stories. You can see this in our reviews. All company names, I think, are familiar to you. There is an opportunity to also get employed in them after completing training at our school and successful internships. Next, Denis will tell you about Machine Learning today. He is a lecturer of our course at the PAZVS school. I will now hand over the word to Denis. Denis, please, tell us about yourself and continue. Hello everyone. I'm glad to see everyone. Yes, my name is Denis, and I am the main lecturer of our Machine Learning course at the school. And I really want to show you, to open up the world of ML, neural networks, and working with Python. I've been doing this for a very long time. And I would like to do this not only for companies, although I still continue to work in this field, of course. But also for you, for students, to tell you how you can start using all of this. In our course, we go from the very basics of the language to modern neural networks, large language models, and so on. So, our course is designed for people who want to start from scratch, or, accordingly, from some intermediate position where you might not have been very involved or understood this at all. To start dealing with this. In my experience, I have mainly worked with natural language, so text generation, creating chatbots, search engines, processing a large number of articles, products, user messages. But, of course, there is experience in other areas of machine learning as well. And I've been teaching our students here for over a year now. We've had many groups graduate. The fifth is currently running, even the sixteenth, if I remember correctly. As for today's webinar, we wanted to fit it into about an hour. First, I will present a more theoretical part for those who, for example, haven't been with us before, who are not familiar with it at all, so that you can at least start to understand what it is and why it is needed. And then I will try to conduct a small practical lesson where I will show with a not very complex Python code example what can be achieved quickly with Python. I will note that you can write any questions in the chat, and at the end of the lesson, we will try to answer them all. So, [music] feel free to write at any time. However, the answers will be... Right now, everyone is talking about artificial intelligence, data science, a whole bunch of buzzwords and so on. And we would like to understand a little better what it all is, and how they differ from each other. So, there is a certain hierarchical dependence. So, there is AI in general. So, this is some kind of computer science that allows computers to solve tasks at a human level, or at least try to do so. So, generally, high-level things are included here, like self-driving robots, self-creating films, programs, and so on. So, everything that machines do themselves and learn themselves, that's all about AI. So, this includes voice assistants, and modern navigation programs for small and large robots, chatbots, and our favorite GPT, ChatGPT, and all that. So, everything that is quite advanced and automated, so to speak. Whereas ML is something, let's say, it's a sub-part of AI, on which all of this essentially works. So, in AI, you might have more abstract questions, like should robots be given rights or what is that? ML is all about specific models, specific functions that we can train and that will solve our specific target task. So, in machine learning, we would like to show you what it means to train a machine, what it is in general. So, these can be more applied, more business tasks that usually go into this area. So, of course, everyone uses machine learning, but machine learning itself is some specific algorithm for recognition, classification, clustering, scoring. So, fraud detection, credit scoring, chatbots, and so on. So, what directly allows you to solve a task, but not with classic programming where a person manually writes the entire algorithm, but so that a part of this algorithm, the main key part, learns, so to speak, independently. Based on the data we show it. We showed it that these are the numbers we have, or this is the distribution of emails. It remembered it and based on that it makes a decision. Deep learning is also a sub-division of ML. And this is what is responsible for neural networks. So, initially, although neural networks are just one of many directions, one of the algorithms that exist in ML, there are many algorithms in ML. Deep learning is specifically neural networks, certain types of these algorithms that exist in the general ML framework. But, it so happened that neural networks are the most popular, the coolest thing invented so far. They solve the most advanced tasks, thanks to their ability to learn from a large amount of data, unlike many other algorithms. And we are getting more and more data on the internet. This has led to the development of neural networks. So, you have all the advanced tasks here for text generation, images, translation of anything into anything, images to text, text to images, video to sound, anything you want, you can implement with ML and make something quite advanced out of it. This can be an application, a website, a service, or just your applied script for, for example, classifying images on your computer. So, neural networks can be both local and cloud-based, and we will talk about them. Data science is a bit different. So. Because initially, data science is more about calculating statistics, gaining insights from data, visualizing them, preparing them, extracting trends. So, initially, data science is more of a mathematical history, building mathematical models and calculating them. But over time, it has merged a bit with ML, because for many data science algorithms, it is much easier to create a special model that will calculate these trends. So, in the end, it's something like that, it's not entirely pure ML, it uses ML, let's say, it's a direction, and the role of a company, so to speak. But there is a combination with the more classic world. We would just like, as future potential specialists in these fields, to understand the difference a little deeper, not just to hear these loud words, "we've added AI to our program." But to understand what should usually be behind it, at least. So, DS is still a more mathematical history with a blend of computer automated calculations. So, this is with the help of algorithms, for example. So, DS is better, you should understand what domain you have, what data you have, what peculiarities there might be. So, for example, in regular ML, you might not understand much about marketing, and you can solve some marketing tasks. Of course, it's recommended, but it might not be as mandatory. In data science, you usually need to understand the specifics of the data and the field you are working in much more deeply. So, AI, usually there isn't such a direct profession, like an AI developer, for example. So, usually it's like a programmer. Now there are no just programmers either, there are back-end developers, front-end developers, testers, and so on. It's the same with artificial intelligence. So, there is, let's say, an abstract AI specialist. But in a specific company, they will be involved in research, or building models, or creating chatbots or search systems, or solving some more specific tasks. Moreover, an AI specialist can even be in a more humanitarian field, for example, regarding ethics, security issues. What we are striving for here. What does AI give us, who should it replace, who shouldn't it replace? So, this is also a kind of history. So, it's not always purely about programming. Unlike ML and DL, of course. Here, in fact, specialists in these fields are primarily programmers who can write code that will solve a specific applied business task. Maybe it's not always necessary to train a large model from scratch. You can take some ready-made implementation, fine-tune it, or use cloud solutions, for example. Nevertheless, to be a machine learning or deep learning specialist, you first need to be a good programmer. I will note that our course is aimed at turning you into a Machine Learning Engineer. So, this is a programmer who is well-versed in programming, primarily in Python, because it is our key language for machine learning. They know how to build models, analyze data, and accordingly, create ready-made services from it. So, a person who creates a business solution in a company. So, a pure deep learning specialist is a rarer case, because although we often use neural networks, it's not the only applicable solution everywhere. Sometimes it's necessary to do something more classic, and it still requires a lot of data. For example, data science is always present here as well. I will note that serious mathematical apparatus in code is not directly required. It is desirable to know, especially for interviews, strangely enough, not for the work itself. But we will talk a little more about mathematics later. Regarding the slight difference between ML and DL. The idea is primarily that in pure machine learning, we typically need to work more with data, engage in feature extraction, so to speak, characteristic extraction. Deep learning, neural networks, in this regard, they have more power, more, so to speak, abstract representation of our data. They can solve a larger number of tasks independently in this regard, without your help, so to speak. So, this is solving a task like image classification. So, artificial intelligence itself can mean many different things, but it is usually considered a field of computer science that tries to solve tasks that are at the level of human intelligence. And indeed, in the last few years, this has seemed to be true. If earlier ML, and AI as its consequence, was more about solving very specialized tasks. So, there is a specific dataset, it has specific features, and you just need to predict the probability, say, of whether a store built with certain characteristics, with a certain number of floors, shelves, departments, and so on, will be successful. So, very specialized tasks. Now, as we see, AI can solve very, very good tasks that previously could only be requested from some cool specialists. But despite all this, AI is often built on the same principles as regular ML, and accordingly, regular neural networks. It's just that everything is constantly moving to new levels of development, thanks to the feature of neural networks that they are much easier to scale, develop, add new layers to. And the more data, the cooler tasks they can solve in this regard. The question often arises, what is the difference between programming and machine learning? Because we program in both, and in both cases, we have input data and output data. So, I don't know, a user comes, writes something or sends their characteristics, and we give them a credit score, for example. And in fact, a very large part of regular programming is similar. Both require writing quite new code that will solve, will do some wrapping. But the main algorithm on which everything will be built in ML is built differently. In regular programming, you need to manually write everything, so to speak, everything you need. So, if some characteristic comes from a person, how much they earn. You need to write exactly what to do with it yourself, what to multiply by, what to add, what coefficient to calculate, and so on. But in machine learning, our main algorithm, so to speak, the business essence of the process you are doing. We can use some function, so to speak, a mathematical one, from a programming perspective, it's just some code where we can send our input data, and it will automatically, based on matching input and output data, select the most effective coefficients that will build and calculate a credit score or the probability that a person will have a certain disease, or that in a certain area of an image there is an anomalous object. And this is the main essence of what you are doing, it is much more automated. You can use some ready-made models, or train them from scratch, or fine-tune them for your task. In any case, the process is accelerated. So, where there is data, you can use it. So, but it's important that in regular programming, you might not have any data, and you've done everything, written it. But in machine learning, it's very important that you have datasets. The more data, the better your model, the better your service result. If you have no data at all, there's nothing to try. So, in the case of a typical example, we have the task of image classification. So, when you try to determine what is happening in an image based on the distribution of pixels, areas in the image, their specific characteristics, and to which class, type, or category this image belongs. So, the idea is that we need to show, so to speak, our algorithm some images, and so that it remembers them and distributes them into groups. So, you can say that it remembers some statistical characteristics of your data. So, for example, for such examples, there are more black pixels here, or in this table, for this feature, there is a smaller value, for example. It remembers it, and then for data that it has never seen before, they are processed by the same algorithm and give you some result. And in this way, although the principle is very similar everywhere, to show examples, build some statistics from them. In reality, very cool things happen, and cars drive themselves, and now you can generate any images in seconds. So, in some sense, it's all just calculations of statistics, multiplication of a huge number of numbers. So, you just have certain matrices that store some coefficients of these functions that predict something. You get input data, for example, an image. So, you convert this image into numbers. The numbers are now multiplied by these internal coefficients. You get some vector of numbers, you convert it, for example, into text. And you have a model that converts images into text, their description. Or, conversely, from text, you convert everything into images. So, the main idea here is very similar everywhere. But there are nuances, of course, everywhere. So, every service, every large company is trying to integrate this algorithm. Some do it better, some do it worse, but it is indeed a very popular direction now, with many vacancies, many sub-types. So, as I said, there used to be a regular programmer, and now there are a huge number of sub-types. So, now there isn't just an abstract artificial intelligence specialist. There are specific sub-types, which we will get to know a little more about later. These algorithms themselves can solve various tasks, from writing your code to responding to cyberattacks. And mainly, all of this has become popular thanks to neural networks. So, earlier, it was, let's say, a direction that has existed for a long time. So, all of this, as a mathematical apparatus, appeared in the middle of the last century, but there wasn't enough popularity and need for it, because there wasn't much data. And as I said, for ML and the entire sphere, the most important thing is data. So, it's clear that everyone is constantly competing for it. Secondly, there weren't enough computing resources for it. Now there are both. And thanks to this, it turned out that neural networks can indeed be very useful in this regard and solve a lot of different tasks. This is all thanks to the same principle, roughly speaking, as in the human brain. We have a large number of very simple computational centers. So, let's say, a linear combination of your numbers is taken, passed on. Then, maybe a simple non-linear formula is applied to it. But if we increase the number of such neurons, then these models suddenly start to process complex hierarchical data, features are automatically identified. So, like a person, if they have one neuron, or 10, or 100 neurons, they don't do much useful for us. But when we start creating billions of neurons, then a large number of potential combinations, their joint work, allows us to solve the most complex tasks. So, these can be the same tasks that were solved in regular machine learning. But, accordingly, with a large amount of data, neural networks perform much better. So, if you don't have much data, then, as a rule, there's no point in even deploying it. It will be difficult to train, it will be more difficult to use, and the quality will likely be about the same, maybe even worse. But when it comes to the modern era, big data, millions of users, a huge array of data, then neural networks...

They shine and allow you to solve tasks that previously, ah, yes, were very difficult to solve, even manually, but here they are solved automatically. Here. And as for typical code, it looks something like this, for example, some simple Python code for working with numbers. Here, we will do something similar ourselves, right? So, we have rules for importing some libraries, then creating, for example, our data, then creating, for example, some model, and, accordingly, using it for prediction. So, let's assume that we have the simplest possible scenario, right? We need to predict, for example, the probability of loan approval based on, for example, salary. And we have just a dozen, for example, examples, that a person with such and such a salary was approved for a loan, with such and such, for example, no, right? So, yes, it's understandable, in real systems, you use data. There are, possibly, dozens of different sources, right, a huge number of features, but here we have everything super simple, right, we know the salary, we know the result. And, accordingly, our story is to train a model so that based on salary, right, it makes some credit scoring, gives some percentage, ah, whether it's worth issuing or not worth, for example, issuing a loan. And we have some new clients, right, about whom we know the salary, ah, accordingly, we don't know the results about them and we want to get them. And we can, here, accordingly, on the graph, right, build it. So, the crosses are, accordingly, the predictions for new clients, right, so, ah, conditions, the lower, right, the further to the left this point is, the lower the person's salary, the higher this point, the higher the probability of loan approval. And, roughly speaking, we can, in principle, observe here that with, for example, a conditional salary above 50,000, right, we, accordingly, start issuing loans. Or, at least, give a scoring above 50%. So, well, of course, the threshold value itself, right, can be something different, that we, ah, should expect that the model will give at least 80%. Here. But nevertheless, we have built the model and can use it, right. Here is a very simple example. Let's talk a little about Python, right? So, Python is, accordingly, the key language, right, we teach it in the course, accordingly, and, ah, from scratch, and up to advanced concepts. So, unlike, which, well, you can't just take it all and learn it in an adequate amount of time, right, so in the time it takes us to learn it all, right, another 10 different models will come out and there will be a bunch of new areas, right, with which, yes, we need to work. Therefore, Python as a base, we work through it very deeply, right, the first half of the course, right, there is a block on programming. Well, not only Python, but also, for example, other related things in this regard. Here. And the second half of our course is ML. And Python is good because, right, it is, accordingly, quite easy to read, right, it is, accordingly, quite abstract. This means that we write more human-readable commands here, right, and do not refer, for example, to processor registers or manual memory allocation. in RAM, right? So, roughly speaking, we just need to write the command print h here, right, to, accordingly, get, right, our first such, one might say, program in Python, right, we run it, right, here we get world, right, so that's it, so in many other programming languages we would have to create some functions, import some, maybe, libraries. Yes, here, in general, connect some kind of output, right, it's understandable that there are a large number of languages that are, well, let's say, at the Python level in terms of their complexity. Here. But it so happened that Python is the key language for machine learning. Here. Well, and in general, one of, yes, the most popular in the world right now. Here, yes, it should definitely be in the top three. Here. And therefore, knowledge of Python, right, in principle, opens up the world not only of machine learning, but also of some other areas. So, you can create backend services, and even simple frontends, and solve various tasks of interaction with some systems, and write, accordingly, some code snippets, for example, some modern methods, not methods, rather, automation systems. Here. So, thanks to its, let's say, simplicity, right, you can do quite complex things here. Here. Well, let's take a simple calculation example, I don't know, 2 + 6 minus 3 to the fourth power, right? Well, here is an example of mathematical output. Here. Well, in general, the main idea of Python, of course, is not just that we use some standard functions here and so on. Although, of course, a lot can be done with that too. Here, the main thing is that we can import some ready-made libraries, modules, take code from them, right, and, accordingly, you don't need to write, for example, some complex function, some model, right, you can take a ready-made prototype, for example, and adjust it for yourself or run it with your parameters. So, for example, we take the random module, right, import the random integer generation function Rent in from it. Here. And simply refer to it, specify the range from minus 100 to 100, right, here we get some random number. We run it again. Here we get another, right, integer, right, and, accordingly, we don't need to think about how to create some kind of random number generator, how this sequence should work, no, we just take some ready-made function and use it. Here, of course, there are tens of such standard and external modules from the community. Here. But the advantage is that we don't need to write all this ourselves, right. Rather, we need to correctly combine already ready-made modules and libraries into a single entity, right, into that very dream program that you wanted to write, right, maybe to launch with asphalt, right, this is the philosophy, so to speak, that's why it's displayed here. Here. Well, let's now take and import, for example, some set of various functions, right, so let's just import all this here, and then we can use it. So, we don't need to, accordingly, write implementations for all of this, right, we can just take and use ready-made ones. So, for example, if we want to work with some dataset. For example, there are very useful functions for loading datasets from the internet, right? This is, for example, all sorts of load datet, right, in which we specify the dataset name. Well, let's do something more fun, for example, here and save it all into some variable. Here. And now we have some variable. Ah, DF, dataframe, right, as it's abbreviated, which stores information about, accordingly, penguins, right, we know, they have some species of penguin, lives on some island, it has some characteristics, its beak length, flipper length, for example, weight, sex, right? Here, and so there are 344 such penguins. You can always, for example, take and look not only at the first and last, but just take a sample of 10, and this will give us 10 random penguins. Here. So, this allows you to simply load data from the internet, right, or, for example, consider the stories that we want. So, what's the story now? We want to try to use Python, right, to quickly visualize some data, that is, essentially, to solve a task, right, so to speak, data science. Here. And then to solve the task of building some model that will, based on, accordingly, characteristics, well, for example, right, of a penguin, decide which species it will belong to, for example. Here. And we can always look at the dataset, for example, for some characteristics. precisely numerical, right? So, here we have, for example, we are interested in BMS, right, how much penguins weigh. Well, accordingly, here we see the line min. Well, so the smallest weighs 2 kg 700 g, and the largest 6 kg 300 g, right? Here. Ah, and on average, for example, it weighs 4 kg, right? So, ah, it allows us to quickly learn something about our data. Here. Or, for example, we can, we are interested in, what characteristics are here, right, of our dataset, right? So, ah, here you are sent, accordingly, some data, you want to know, what is the distribution here, right, so what kind of data have you been sent, right? So, ah, this is, accordingly, a typical situation where you are sent data, you need to review it, and based on it, accordingly, build some models, right, and, accordingly, solve some business task, I don't know, in this case, we wanted, probably, classification, right, so that, ah, you are sent other penguins that have never been seen before, right, that they live on such and such an island, they have such and such a beak, right, there weight, and so on. Ah, well, this is, of course, Chinstrap or, of course, Adelie. Here. So, we can only imagine that this is now a task not about penguins, right, but about, for example, real people, right, or, for example, about some cars, right, there helicopters, and, in general, anything, right, that your company will have, right? So, the important thing is that the principles that we are looking at here, they, in principle, do not differ much depending on what specific data and in what specific field you are working, right? So, ah, the principle of machine learning is that we can take data, right, transform it in a certain way, and based on it, accordingly, train some model, right, accordingly, machine learning, right, and get, right, the final, accordingly, predictive version. Here. And, for example, we want for our, for example, these columns, right, accordingly, for the columns. Ah, we are interested in the species column, we are interested in the island column, and let's say the sex column, right, so for all these columns, which have some categorical data. We want, for example, to calculate, what are their distributions, right? So, we just take, accordingly, a column from our dataset, and run the data calculation method on it, well, value counts, ah, well, and let's make some, ah, print separator, right, so that it's all in a more human-readable form. So, ah, yes, and value, right. Here. And we get, accordingly, this report, right? So, now I just want to show that you can, in a very simple way, right, so if you know the name of some function, right, or at least just know where to look, right, the name of this function, you can access your data, your programs, right, which you will write, and very quickly, right, produce, accordingly, some ready-made results that can be useful for analysts, clients, your bosses, right, and so on, right? So, here, for example, it shows how many species of penguins, right, by, accordingly, each of the options, right, or, on each island, how many live there, right, or what is your distribution, for example, of penguins by sex, right, so you don't need to sit and manually calculate in your dataset, I don't know, in some Excel, right, you can just run one function and get the result immediately. Here. And here, by the way, there are, ah, sometimes, ah, penguins with some missing values, right, so those missing values are not very good for our model, so let's, ah, remove them altogether, right, for this, we just specify, accordingly, the drop function call, right, and that's it. Here, we have removed all of this, accordingly, the missing values. Here, now let's do something more beautiful, right? Now we have been making some reports, right? So, look, we are now writing, ah, code that uses some, yes, quite popular libraries, right, so, ah, it's understandable that this doesn't mean that you just sit down with Python and immediately start writing like this. This is what we, accordingly, go through in the course. And then you will be able to easily write it from memory, right, and solve, accordingly, the task, for example, in this case, data analysis, right, and its study, right, showing, accordingly, what dependencies and their characteristics are there, right, well, and, accordingly, building some automated model, for example, classification. For example, we want to build now, for example, a task based on characteristics, for example, of penguins, for example, the ratio of their beak length, right, and, for example, width, right, accordingly, it is enough for us to specify these characteristics here. And, for example, ah, yes, we get, accordingly, immediately such an interactive graph, right, you see, it's interactive, right, so all our 340 penguins, they are all presented here. Ah, and we can hover over each one, look at, accordingly, now the data that is presented here. Here, we can, for example, color them, right, in our species colors, right? Here, now blue is one species, right, green is another, and red is the next. Here. Ah, what can be interesting for us here, right, is that again, we took, wrote one line, right, and, accordingly, immediately got such a distribution graph, right, which we looked at. Here. And it's immediately clear, right, what characteristics the blue penguins have, what the red ones have, what the green ones have. Here. Ah, as our current dataset, let me remind you again, we are just working with some data, right, let's imagine that this is just data that, ah, has just arrived, and you want to analyze it somehow. Here, right, so you, understandably, will have your own data for your specific company, for your specific project. Here. But, I say, ah, in principle, the processing of this data does not change at all, right. Here. Or, for example, we want to build some joint work with our graphs, right? Let's just look at a few more types of graphs, right? For example, where we build such a distribution graph, right, where each of these species, right, we build such an area graph, right, of the most frequent values, how they intersect with each other, for example, right, so, or, for example, you can even take and build this graph in 3D, for example. We have built-in visualization capabilities here. You see, we can rotate all these points, zoom in, right, examine each of them, right, and, accordingly, depending on how long, for example, the text is, how long the flipper is, right, accordingly, a different color, right, so, ah, what's important to understand, right, is not that you will even go through this in the first or second month, right, so, roughly speaking, we start, of course, with much simpler things, right? So, we, yes, we first turn to the most basic things, right, there we write some functions, classes, right, such simple things, we open some files, right, so, for example, to read something from there, we can write some, I don't know, simple scripts, for example, right, for data processing on your computer, for example, search for duplicate files. Here. Or, for example, ah, we want to look at, for example, some other examples, right? Here. Ah, yes, in any case, you can quite quickly, accordingly, go through the path from doing, for example, a very simple dataset to, accordingly, our data, which is more complex. Here. Well, let's quickly build, for example, some model, right, for this, here. Then, ah, accordingly, after dividing all this into training and testing sets, we will build, accordingly, our current training model. For example, we use here, for example, some kind of random forest classifier. This is just one of the many classic models. Here. And let's see how it works for us. Well, you see, our RF is quite an advanced model. It worked for us at 100% in this case. This means that for all test data, that is, for the data that the model has never even seen, it still managed to absolutely correctly predict one of our three classes. Yes, we are trying to classify our penguin species here, right? So, we can also build a matrix, right, where we can observe that each of the species has been correctly transformed and analyzed. Here, let's, while we haven't strayed too far from this, answer questions. Yes. So, ah, yes, Denis, let me help you with that. Olga asks us, please tell me, where do we get the initial data from? Will we search for it ourselves in official sources like statistics, or will clients, bosses, someone from our environment at work provide it to us? How does it happen? Yes, of course, our clients provide us with data, right, so, well, for example, ah, you might have an internal project, you might have an external project, so, for example, you are solving a task as a contractor, I don't know, finding clients, for example, for your bank. Well, then you are sent, accordingly, a dataset in which there is information about, for example, some people, and you need to calculate, what credit products are worth offering them, what cards, what other options, accordingly. Here. And I say, it all looks, in general, exactly the same. So, you also have a table, only here you will have not species, for example, I don't know, city, right, not island, but, for example, age, right, there, and so, accordingly, there are also dataframes with data, just from the client for a specific task, right? Here. So. Ah, yes, so, ah, what else do we have here? Yes, so, yes, so, the data sources are, first and foremost, of course, for the task, right, the corresponding ones, right, so depending on what product you are making, your data can be tables, images, text, voice, video, anything, right? Here. But, ah, the principles themselves, they are actually very often, yes, quite close and similar. And the main idea is that if you know, let's say, the basics of Python, right, so you understand how to write code in it, right, so, ah, yes, write some of your own function, right, you can take, for example, a function to calculate the cube of a number, right, so we get x, ah, we return, ah, x³, right, well, and accordingly, in the future, when we need to simply calculate numbers, we call it by name, pass the number here, for example, five, right? Here. And we get, accordingly, the result, right? And these kinds of mini-stories. All of this, ah, you will be able to understand and figure out how to use it quite calmly within the first few classes. So, ah, there, in fact, the syntax of the language, all its basics, right, we go through it in maybe a month or a month and a half, right, so this is everything that you, so to speak, need to know in ordinary Python. But the main thing, right, the whole trick, right, is that it has a huge number of, yes, accordingly, all sorts of external modules, libraries, right, which we can import and then simply use, right, take, build, distribute, take, build such a graph, take, build such a graph, take, build a model, accordingly, for our task, take, create, ah, a backend, take, create a frontend. So, all of this, ah, can often be written with quite short commands, right, simply by using already ready-made modules and libraries. Here. This, of course, does not mean that you will have the full documentation of all possible functions and methods that exist, right? So, I say, the main thing is that you just know the syntax, what you go through, right, the first, say, month and a half. Here. And then you move on and, ah, for a specific task, right, we analyze a specific library, right, which allows us to, accordingly, quickly solve it. Here. So. Ah, well, let's return to our presentation, perhaps. We have a little bit left here, right, I'll note that modern neural networks, right, as I say, they are all based on the same principles, right, but, accordingly, of course, the development of, precisely, language models, right, what is called, yes, it has, of course, brought artificial intelligence to the people, right, it has become much more popular thanks to this. Here. So, precisely thanks to them, right, we can now, let's say, simply communicate with the computer in natural language, right, so you don't have to write, as before, some of these commands, as we did now, right, and even these are still, well, quite high-level commands, right, without any punch cards or direct access to the processor. Here. Ah, yes, here, accordingly, what is interesting for us is that we can, yes, create a large number of such assistants independently. But, of course, you can also use ready-made ones, right, I generally advise you to try, if you haven't worked with GPT, or GM, or Copilot before, or now, right, so you can, in principle, start with any of them. Almost all of them, yes, among all of them, yes, there is always some free model, right, which allows you, accordingly, to communicate with it. Here, you can try a paid subscription later, right, for example, some more advanced, complex models, right, which, ah, think longer, for example, about the answer, double-check it, search the internet, better, for example, generate some pictures or recognize your, for example, images. Here. So, in this regard, I highly recommend trying, right, for solving almost any daily tasks, from, I don't know, choosing, yes, where to go, what to watch, what to cook, accordingly, right, to some, accordingly, psychological tasks for planning the future, decisions, right, you can use it as a psychologist, and as anyone, in principle, right, it just depends on what is called the prompt, right, what text hint you ask for from it. Here, I'll note that there are a large number of professions, right, that are similar in name to ML engineer, but they are a bit different. Here we are looking for, ah, yes, often, yes, precisely a prompt engineer, right, because this is precisely the main, so to speak, indicative option, right, how to solve a business task using code, right, analysts, this is still more about mathematics and working with data. Data engineers, I mean, are more about connecting, for example, databases, right, researchers are simply about theory, right, and DevOps is generally about working with, yes, some servers, services, not the code itself. Here. And the path, right, as we have here, is quite typical, right, first of all, we need to understand programming, right, a little bit of mathematics. Here. And then, accordingly, based on programming, right. Ah, first analyze classical models, then neural networks, then start creating your own services from this, right? So, this is a brief overview of what we are currently going through, ah, yes, what I will tell you within the framework of, accordingly, our course. Here. Then we start to delve deeper and, ah, it will be possible to create, based on all of this, accordingly, a portfolio, accordingly, some websites, services, right? And here, accordingly, below is a large number of libraries, right, which, in this regard, will help us, with which we will learn to work. Here. Yes, there is always a question about mathematics. I'll note again that mathematics, in general, is, of course, important, right, but not very much. You can actually be a highly skilled engineer without knowing mathematics, essentially. Well, if you read the basics of statistics, yes, so that you can, for example, build and analyze some graphs, so, ah, yes, we, of course, go through mathematics, right, so we teach you a little bit, ah, yes, so to speak, higher mathematics that may be required for Data Science. Here. But this is precisely just as one of the auxiliary moments, right, it is not key to working in this field. Here. And as for salaries, then, of course, on average, salaries for working as a machine engineer and all other similar, so to speak, directions are very high. It can be called a machine learner, or some, ah, AI analyst, and, accordingly, just a net developer. Here. So, it can be called quite differently. Here. But, as it were, the target profession title is machine learner, right, so, an engineer specialist who knows how to program and create their own services based on machines for specific tasks. Here. And yes, salaries in this regard, of course, are very pleasant in America, right? Well, accordingly, I will hand over the floor further. Yes, thank you very much. Please send, turn on the video so I can be seen. Thank you very much to the participants of today's lesson for asking your questions. I see that you were a little scared. No need to be afraid, honestly, because we teach from scratch. If you have no experience, but you want to figure it out and are not afraid, then, honestly. Our profession start, learning machine learning to machine learning engineer, is from June 16th. Accordingly, you will be able to create and implement algorithms that allow systems to learn from data and, accordingly, make predictions without explicit programming of logic. So, the Python programming language will help you with this. Salary expectations are very pleasant, from $120,000 starting and above. Accordingly, the training lasts 8 months, it's 3-4 classes per week. The schedule filling by the number of classes depends on the module you are taking. So, at the beginning, in Python, there are some recorded lessons, but there are always live meetings every week with the teacher so that you can ask all your questions to a specialist, understand the answers, and move forward. Don't be afraid, you will figure it out during the course and training, where the penguins come from, where the databases of cats, dogs, and everything else come from. Ah, Denis, here, for the future, for example, they already say that they know everything about penguins, you can use someone else next time, for example. Well, this is just for variation. You can go to the next slide, I'll tell you more about how our training takes place and why you shouldn't be afraid of it. Our start is June 16th. The course duration is 8 months, classes are three times a week on average. Accordingly, they take place in the evenings and on weekends. So, it's convenient to combine with work, with personal life, right, with studies. And what's important, you will have time to sleep. You need to restore strength so that the brain works and perceives all this information. All lessons are with the teacher and partially recorded. If, by chance, you cannot attend the lessons in person with the teacher, then they are all recorded. Accordingly, if you missed a lesson, you can watch it the next day in your personal account. All recordings will be stored. We do not take away access to these recordings. In our educational system, you can return after a year, after 2 years, to rewatch something, if you forgot, you can return. Accordingly, if it so happens that you could not complete the course or there are some life circumstances, it's not scary, don't be afraid, you can retake the course, restart at any time. So, it means that we do not take away access, you pay once, you study. You can take breaks, you can return for another cohort. All dates are formed on the website, and you will have a coordinator in Slack who will help you navigate. And you will have a manager who will also help you navigate what to do if something doesn't work out. Therefore, feel free to come, we will deal with all the complex nuances. All training is conducted in Russian. Complex things are discussed in Russian. English will be useful because the terms are in English, and then the internship is in English. So, it means that you are preparing for interviews with our English-speaking mentor. You will work on your resume and you will work with LinkedIn. We confirm your experience and provide background and reference checks. So, technical interview training in English, self-presentation practice, help with creating a proper resume, all of this is in English so that you feel comfortable in further communication with the mentor, with other users, and with the systems that post vacancies where you apply. So, you will have 100% readiness for job searching. Accordingly, the project is the US territory, it takes place online. Our training is online, you don't need to go anywhere specifically. You will have a project with a team, with a team lead. You will be a machine learning specialist on the project. And you will also have developers, testers, and UX designers in your team. So, a full-fledged project, as if you were at a real job. So, you will know in the future where to go, what questions to ask, what to do, who to contact, and how to structure your workday. Let's move on. Accordingly, you will learn the Python programming language, you will be able to solve complex tasks, more complex than Denis showed today. There will be more data, more volumes, more complex integrations. Accordingly, we will teach you not to be afraid of this and to approach it with a cool head. You will understand how to do complex tasks. Accordingly, various machine learning algorithms, what Denis talked about, you will create prediction models. Denis gave quite a few examples. Accordingly, $7,399 for a one-time payment is 8 months of training plus 3 months of internship. The internship is included in the course cost. If you wish to pay in installments, the cost is $7,999. All details will be provided by your manager. I will send you my details now. If, by chance, you don't know your manager, you can write to me. But most likely, you have already been called this week, or last week, or recently, or have been texted an SMS message: "Check your email." We also send messages there if we couldn't reach you. You can even go to the next slide. Accordingly, today there is a cherry on top. The Early Bird price is valid for 24 hours. Instead of $7,399 for the course, there is a discount of $1,400. The amount payable is $5,999. Copy the QR code. If you can't copy it, take a screenshot, you can take a photo. If you can't proceed right now, want to ask more questions, connect, then just save it, and you can calmly use it after today's broadcast. What does the QR code give? The English IT video course. These are 12 recorded lessons to help you get oriented with technical English, so you will learn some new phrases that will be useful in your future work. And it's free for your spouse. So, you pay for one course for $5,999, and you are like two full students, so you have your own program, well, rather, it's the same for you, then you dive into the internship together, and accordingly, you have two separate certificates, so there is an opportunity to become two specialists in the family at once. If you wish to pay for the course in installments, you can discuss all the details with your manager. I will also send you our YouTube channel now. It is run by our school founder, Viktor Bogotsky. Watch, subscribe, like, and ring the bell, and our website so that you can also see reviews from our graduates or sign up for a career consultation. You can also see our future activities on the website. We are currently running a marathon. Tomorrow there will be another meeting. Come. It will be at 9:00 AM California time. That's 12:00 PM New York time. We will also talk about fears, what not to be afraid of, and how to act in the first steps. On the website, you can also see the modularity of the machine learning course, you can see what the entire training program consists of. If something is unclear, then contact us, we will explain and show. Ah, Denis, let's see what else we have regarding questions. I see that everything has been answered in general. Uh-huh. So, yes, I just want to emphasize, right, that the code we wrote today, it is, of course, not immediately understandable to you, right, because we have never gone through it before. Here. But at the same time, you will be able to write such code from memory without problems, right, after the course, and much cooler, right, which will do a whole bunch of different tasks for you, right, which you will accordingly receive from your boss, management, right, and, accordingly, you will be able to automate a whole layer of work in the company, right, so that again, not a person manually does it, but your algorithm that you wrote. I suggest, then, yes, to ask any questions, perhaps, someone has, to ask them verbally.