Getting Started with Machine Learning: Supervised and Unsupervised Learning + Scikit-Learn

Key Takeaways

The fun might be pretty simple. AI is artificial intelligence, right? And uh whenever you want to say AI, so AI comes with intelligence. So first of all, let's define intelligence, right? What is intelligence? So uh Lee is saying it's a subset of machine learning. That is correct. That is perfectly correct. AI is the subset of sorry machine learning is the subset of AI. Okay. So if the question is anything which make any machine behave like humans. Okay. Swabandi is saying anything that make the machine look like behave like humans. Okay. So is it behaving like humans or behaving like intelligent human beings? Okay. Because not all the human beings I came to here. I think there's again speak in English please. I understood comm. It's okay. Set up. I think sir network is not properly by s network Maybe network problem connecting is Hello. Hello. Hello. Hello. Hello. 1 million. 2 million. Hello. Hello. So really sorry for the delay guys. There's uh a storm here at my place. So there's some internet and electricity issue. So I hope the screen is visible to you again. And uh let's get back from where we have started. Okay. So the fun is uh we started with introduction to machine learning, right? So what is machine learning in machine or or basically we are starting with artificial intelligence. Someone has mentioned that AI is nothing but uh algorithm that help the algorithms uh basically algorithm that act as the human dos right so but it's not like that because not all human beings are intelligent right so the fun is uh the algorithm that can act intelligently that can make intelligent decisions we call them artificial intelligence algorithms okay it's not like these are some new algorithm We also have AI algorithms like probably 50 60 years ago as well. We have those uh such kind of algorithm. But the fun is these algorithms are very specific for a specific use case. For example, you might have heard about the algorithm called as breath for search, depth for search. These are some algorithms that we have. They also comes under AI machine learning. We have like alpha beta pruning and these kind of algorithms that comes under AI. uh but they are not very much uh they are very specific algorithm basically and they are not datadriven algorithms okay now in AI we have a lot of algorithm that basically act as an intelligent human being that can make decision on their own okay sir I think you are not audible am I not audible I think I am uh do let me know guys am I right okay yeah sir Okay. Okay. So now if I coming back to the same point, AI is basically AI has the algorithm that can act intelligently. Right. Now how comes a different aspect of it which is called as machine learning. If in AI we have algorithm that can act intelligently. Why we have a new term that's called as machine learning. What is machine learning then? And how machine learning is different from deep learning. Yes, someone has mentioned that machine learning is a subset of deep learning. This part is correct. This part is correct that machine learning is a subset of deep learning. Oh, sorry. Machine learning is a subset of AI. Now what is machine learning? There must be something in machine learning that does not comes under AI, right? So there must be some specific part. And this particular session like not today's one but all the session from today and the day after tomorrow and tomorrow's as well we are going to totally specifically basically discussion on machine learning algorithms but this session will give brief about how it is different from deep learning and generative AI agent what are the different things we have okay so someone has mentioned it's a subpart yes it's a subp part obviously what else do we have so uh kashif is saying training machine learning to make decisions in ML okay we have the concept of training here in machine learning that's also correct please stop annotation yes I have done it machine learning allow machine learn from given set of data to make its own prediction that's correct so that is the correct definition so in machine learning we have some set of algorithms that are trained on data in machine learning we have datadriven algorithms in which I will give some data And once the data is given the algorithms will go through data. It will find some patterns. It will do the pattern analysis or feature engineering or feature selection part. And once that part is done, deep learning al means then so in machine learning specifically we have datadriven algorithms. In machine learning as well we are going to discuss about supervised learning unsupervised learning and reinforcement learning. But let's go a level deeper just to give you a a level one overview of it. Right? So once we have it, we also have something called as some of might have guessed as well and their guess is correct and the guess is how it is different from deep learning. Then we also have the concept of deep learning. Now what is deep learning? Whenever I'm writing DL that means deep learning. Okay. So yes, bankrites show this is the correct part. We also have a subset called as deep learning. Now what is deep learning? How deep learning is different? Okay. It has ENN. Yes, that's that's a correct part. It has ENN. Now what is ENN? EN is artificial neural networks, right? So it has neural network that's correct above bankage. That's also correct. We have algorithms neural networks. Yes. So in deep learning so as you can see in AI we have all the algorithms in machine learning we have datadriven algorithms. So in deep learning as well we must have datadriven algorithms right because machine learning datadriven algorithms in deep learning as well we will be having datadriven algorithm. That's that's pretty clear. But in deep learning we use a specific kind of algorithm. In machine learning we have different algorithms. In deep learning we have one specific kind of algorithm which is called as neural networks. Okay. So in deep learning whenever you see deep learning deep learning means neural networks. So whenever you say it basically in deep learning allow machines yes that that's the correct part of deep learning allow machine learning algorithms uh allow machines to learn and behave like human neurons. Okay. So in deep learning the way we learn. Now the question is how we learn. We have a brain right all of the human bar uh human beings comes with the brain. Okay. Uh now inside our brain uh we have neurons how many of them? Couple of billions of them. Okay. And whenever we do something, whenever we are learning about anything, whenever I'm reading a specific book or reading any another topic, whenever I'm moving my hand, whenever I'm even singing or listening or whatever I'm doing, a specific set of neurons in my brain get activated. Okay. And in our brain, we also have a very specific property which is called as neuroplasticity. That means the connections that we have in our brain of basically connections in the neurons that we have in our brain can be made and break as well. If you want if you are learning any specific topic or learning a lot about it then it will make new connection in our brain. If you left it for probably a month so connection will start getting weaker. This is how we learn right. We have our brain inside that brain whenever I'm learning speaking listening learning about a specific topic another topic whatever I'm doing specific set of neurons in my brain get activated okay and whenever they are activated it will make this uh connection stronger or weaker depend upon the use case okay and once that part is done this is how we learn kind of the same thing we also apply in deep learning algorithm so in deep learning as well Now if if I'm talking about human being we have neurons right which are biological in nature but in computers we'll try to make neurons as a mathematical function we'll try to make neurons and not just one or two we'll try to make like couple of hundreds or millions and if I'm talking about deep learning or probably generative algorithms we have geni models that have billions or trillion billions of neurons Right? So this is how we learn and this is how we try to mimic the same thing in computers. So in deep learning we have neural networks. In neural networks as well we have a lot of type. You might have heard about ANN which is artificial neural network which is the simplest kind of a neural network that we can have. You might have heard about the term RNN which is recurrent neural network which is again kind of a neural network only that works best on text data or sequential data. You might have heard about the term CNN which is convolutional neural network. Someone had mentioned as well but that's correct. We have feed forward neural networks right? So in convolutional neural network these are the networks that works best in images kind of data right and you might have heard about the term transformers. What are transformers? Okay just tell me one thing this is a very basic question. What does GPT stand for? What is the full form of GPT? very basic question. Okay. And before that I have asked you another question as well. What is transformer? So some of you might have even if they don't know they might have guessed as well. So GPT full form like the T part probably might be from transformers. Right? So a lot of you have mentioned it correctly. The GPT is nothing but generative pre-trained transformers. Right? generative and premium is something we are going to discuss as we go ahead in this lecture because al I'm also going to discuss about uh generative a in this particular lecture but if I talk about T part of it T or GPT so T is transformers transformer is nothing but a different kind of neural network architecture it's nothing but an architecture only that that works great with sequential data now you might have seen sequential data or text data we also RNN LSTM GRU and a lot of other architectures as well that comes under deep learning. But this is the state-of-the-art architecture transformer that works great and it's such a great architecture that in 2017 there is a famous research paper uh now because this lecture is being recorded so you can go back and watch about it again and again. The research paper is called as attention is all you need and inside that particular research paper uh they have introduced this architecture which is called as transformer. They have used a basically a layer which is called as attention layer which is all you need uh to train a model and based on that 2017 uh research paper only uh we have transformers that are so big so versatile so uh well managed and so so intelligent that they can they can like you can see uh probably example of charged okay attention is all you need is a research paper right attention is all you need you can go here this is the research paper I'm talking about okay this is the research paper where where like it's first coin the term attention is all you need and this is the sole architecture of it okay if you go back this is the the transform architecture those who already know machine learning and deep learning a little bit of it they might have seen this architecture earlier as well but if not That's totally fine. Uh but this is the groundbreaking and good part and and the funny part about it is it is from Google right 2017 paper by Google and first great company first very big company that has taken the advantage of transformer metal model and built uh the company around it is open AI. So open a is the most famous company when it comes to generate a model and working on the same architecture. But the soul is from Google. So yes these kind of thing happens right now if I'm going back to this part. So in deep learning we have algorithm that are great at that that use neural networks. Okay. So now so far going good. In AI we have all the algorithms but in ML we have datadriven algorithms. In deep learning we have datadriven algorithms but here we use a specific architecture which is called as neural networks. Neural networks burst a lot works a lot better than machine learning algorithms if we are giving enough data to it. Right now there is one more architecture or one more subset of it. Can anyone give it a guess which topic I'm talking about? Okay. Yes. So someone has mentioned NLP. Yes, that that's also something that that must be mentioned. So let me write here as well. Whenever I'm applying deep learning algorithms, Whenever applying deep learning algorithms on test data, sorry, whenever applying deep learning algorithms on text data, we call it as a new field which is called as a called as natural language processing. Whenever deep learning algorithms are applied on text data, we call it as NLP. Okay? And you might have heard about one more term and that's called as computer vision, right? And that's correct Sam. That's that's really that's really great that you have guessed it right. Whenever I'm applying machine learning or deep learning algorithms on images on videos, we call it as a new field which is called as computer vision. Okay. Now, now here comes the important part. Now where exactly do we have generative AI right let's talk about the elephant in the room okay because this is where most of the research nowadays going on this is the most fascinating application of generative uh this is the most fascinating application of deep learning in general okay or AI in general and the field is called as generative AI now What is generative AI? So here you can here you can see a mind map. Here you can see a vein diagram where you can see exactly what's happening. Okay. So what is generative AI? What are we going to study in these three days? So I mentioned earlier as well. So this is the general architecture we are discussing about what the different components do we have. In the next class we are going to discuss about machine learning models and how to do feature engineering on it. how to evaluate the models and then in day three we are going to build a machine learning machine learning model based project okay and we'll see how the deployment part is done on top of it as well okay today's proper theoretical lecture where we can have a detailed overview of what exactly do we have okay now what is generative AI now there are lot of you might have used tools like charg is an application of generative AI only right and if not charg you might have this what is AI agent that's also an application on top of generative AI right so what is generative AI generative AI is nothing but an algorithm only that comes that's nothing but a deep learning or that's nothing but a neural network only but generative AI algorithms are so great that they can generate new content right let's take an example if I go to Netflix. I'm not going on on Netflix right now. But let's take an example. Okay. If I open Netflix, you might have watched couple of movies, right? Based on the movies that you have who you have watched, Netflix will recommend you like these are the movies that you will most likely to watch next, right? How the Netflix algorithm knows that? Because you have watched couple of movies already. Same thing will happen if you are ordering something from a blanket, ordering something from uh Amazon, ordering something from Myntra, right? Same thing happens whenever you are are going through an mail or probably like that advertisement mail. Same thing happen whenever you watch any ad on Instagram. Okay, how these algorithms are doing it? Because these algorithms can predict your next move. Okay, now that's great. We have a lot of algorithms that can do that thing. But going to the same example if I'm talking if I'm going to Netflix recommendation system can recommend you the products can recommend you the movies can recommend you the series can recommend you documentaries but can it generate new documentaries the answer is no can it generate the script of new documentaries no that's the issue with deep learning algorithms in generative AI we have algorithms that can generate new content content. We have algorithms that can generate new images, that can generate new articles, that can write for you, that can draw for you, that can paint for you, that can act as a photographer for you and click pictures, that can generate videos for you, that can also generate audio for you. So, in generative AI, we have algorithms that can generate something new, that has the capability to generate something new. And sub is saying that's correct. Gamma is can help you create PPTs. That's correct. You you have mentioned one extra P in PPT. You have mentioned PPPT but that's totally fine. Yes, you can also write in this way. Okay. So that is correct. Clear. So in and and exactly in the same case you might have used enable labs. 11 Labs is again a platform where uh you can generate audio. You might have used Sora that can help you generate images. Not only images, it can also help you generate videos. You might have used Charged that can write articles, write post, summarize the content like you if you want to watch it to our podcast. So what you can do, you can go to YouTube. Uh download the transcribe of this video. There are a lot of YouTube extensions already available. Uh you can go there. Uh you can copy and paste the transcribe of the whole podcast of 2 hours. Copy and paste to JPT and ask it to summarize it. so that you can read the whole thing in just 2 minutes. You can have the crisp of the two-hour podcast in just 2 minutes. It will summarize it for you. Right? So, we have a lot of AI tools that can do these things for you. Before generative AI, we have some some things like that, but they are very specific. I don't know if I've used it. You might have seen or not, but probably four or five years ago or even more than that, uh Nvidia has launched an application or launched a model that can generate realistic faces. But those model that model that can generate faces of human beings cannot generate faces uh of dogs, cannot generate faces of a cat, cannot generate uh landscapes, cannot generate product images. So they are very specific products they are very specific algorithms but now generative algorithms can generate anything that you want right and you can take any vague thought of a choice for example let's let's go for a try okay I want you guys to give it a very vague thought very random thought okay defake is another one yes we have like we have gibble images yes you might have used it as well but I want you guys to write any vague thought or what I want you guys to uh do is uh just mention any random objects in the chats any random object of your choice so that I can show you that how intelligent these models are. You might have heard about these things. Okay, we have ball and uh we have Dorimmon, we have uh horse, we have apple, we have pen, we have ramen ball. Okay, we have books, we have crocs. Okay, we have clock, we have TCS. Let's not mention TCS, but let's Okay, let's add fighter jet and cat and Asher. Okay, Asher is something I don't know if it will take it or not, but let's mention a sir as well. Okay, so this is the basic that that's enough. Not RCB. That's that's totally fine. Okay, so this is what we have. Now what I'm doing now generate a story uh based on the uh things mentioned above. Make it maximum of one paragraph. Make it simple and show the point simple right very simple thought so you can see uh as I search notes for getting themes so here just I'm not going through it I'm just saying uh cat jump on his book knocking over the ramen ball and it's clear taking role Bob and this this is what we have right now not only that I can do one more thing now I'll ask it to general generate a realistic landscape image based on the scenario that you've generated. Okay. Now this is a very vague thought that we have and ask and I have asked you guys to give me all the topics and based on that here I have created a story because I'm also using my not using my brain to generate a story I'm asking Chad GBT to do so. So it has given me a story. Now all I need to do is wait until this series be generated. Okay let's go backless the story is generated and at the meantime now I can also do one thing I can give my own image to ch and I can mention okay this is how she s looks and uh on top of that you can see okay now now tell me what exactly that can be done. So in that case I can have a better image I can have a better Okay. So this is this is what we have. Okay. Now uh this is what generative AI is. Just just want to give you give you an overview of it. Okay. That can generate something new. There are a lot of applications of it. You might have used gamma. You might have used Gemini. You might have used uh deep uh uh deep sea carbon. You might have used uh perplexity. There are a lot of tools that we have. Okay. Once we have it you you might have also heard about a term which is called as agentic AI. Now what is agentic AI and how agentic is different? Now you know in generative AI we have some models that can generate new content, generate new videos uh and uh generate new audio, generate new files and things like that. What is Agentic AI? So TensorFlow Py which one to pick? Okay, for a beginner go for TensorFlow, for research purpose go for PyTorch. For a beginner go for TensorFlow or Gas. Okay, now the fun is what is agentic AI? Okay, what is agentic? Where agents would be working on parallel task. Okay, but before that it would be great if you can uh see what an agent is. Okay, now whenever I'm doing something like this, for example, here you can see this is what it has generated. Okay. And uh now because the model don't know uh how a sheer exactly looks. So it has generated probably a professor of of anything uh like that. So which one would you prefer? I don't know which one I should prefer. But I think this uh one like I don't know which one to use, which one to prefer. Okay. Let's let's take this one. It looks more realistic. Okay. Now what I can do uh I'm doing a very very basic uh stuff here. I'm going to my own images and I'm going to give one of the images of myself to the model and see if now my model this is a sir now regenerate the okay so rather than mentioning here I can go here and here I can mention like this is how a sheet looks now regenerate it. Okay. So now most probably it will regenerate uh aish in this particular way. Okay. But this is what uh and this is what AI does. Now what is agentic AI? Imagine if these AI tools can do a lot of things for you right for example if I'm opening JBT. Okay. So I'm opening it in uh in this particular mode. Imagine if I want to learn uh uh or if I want to uh if I'm a marketing person, okay, I want to generate some new images of my product. Now, one thing you can do is you can click one of the picture of your product, copy and paste it here and ask it to regenerate it. Okay? Or you can add different different kind of detailings into it. Okay? Or what it can do imagine now this is something you can do, right? For example, you can write uh generate the image of a candle for me that I can sell and you can mention okay this is exactly how I want my candle to look and once that is generated you can ask it to okay now write a description for me in the second step right so if I'm writing a step-by-step process what exactly I will I might be doing okay so first step would be to generate image of my product. Okay. Once the image is generated, now in the next step you will you will be uh write description because if your task is to sell your product, you need to probably these are this this particular agent you want to make to uh what I would say uh to post images or social media uh on your social media handles. So first of all you generate the images then you will write descriptions okay and these are all these things are the step by-step process that we I'm going to write okay so first of all I'll generate the images in the second step I'm going to write the description of the images and once the descriptions are written of the images now the next next task is that after writing the description I will uh probably go for also generate a BD okay because on Instagram I want to gen I want to post my video. Okay. So I'll generate all these different kind of things and once all these things are generated I want to go to my LinkedIn and on LinkedIn I will okay so this is something that now this is going to be a three-step process. Okay. So this is going to look a bit different now because first of all I'm going to generate the image. Okay, once my image is generated, I'm going to either write the description from it or generate a video from it as well. Okay, and once I have written the description and generated a video. Okay. Okay. Sorry I just Okay. So now once this part is done now the in the final part you can post the same thing on LinkedIn. Okay. You can post the same thing on Instagram. You can post kind of the same thing on any other social media handle as well. for example x right so the generated image with the written description I'll post on on LinkedIn generated video on Instagram so this is the whole pipeline you can make it by yourself means you can generate the images by going to sora once the images are generated you will go to chad gpt and ask you to write the description once you have written the description then you can go back and ask uh chat GPT or probably Sora to generate the videos once you have generated the videos then you are going to manually go to your LinkedIn and post the video, post the uh image with the text because LinkedIn will not take video, it will not boost video basically and Instagram whatever re that you have generated you can post it there as well and you can post it on different social media handles. This is a lot of manual work. Imagine if you can have an AI tool that can build pipeline for you that can take these steps for you. Not only that, it will automatically generate the images three times in a day. For example, you can set up at set it up uh that every uh like at 8:00 a.m. or at 2 p.m. and at 8:00 p.m. uh it will generate a new image. It will write the description, generate the video and post it on my LinkedIn, Instagram automatically. You can set this thing up. This is what Agentic AI is. Agentic AI can take decisions on top of you on in on behalf of you, right? For example, JGPT. If you go to JGBT and ask it to generate an itinary for you for Goa, it can gener it can make the itinary. But can it book the tickets for you? No. It can somehow tell you, okay, these are the flight that you this is the average price that you have. But can it book the tickets for you? No. Can it book the hotels for you? No. Right? But using aentic AI there are a lot of AI generative AI LLMs or basically because in generative AI whenever I'm talking about text we call it as LLM large language models. So open AAI's GPT 3.5 4 and 40 and these kind of things these are the LLMs only. Okay. So this is what agentic a can do. Okay. I hope you have a clear understanding what is agentic a is so a aentic chats. Yes, we have it. Okay. And this particular thing that I just mentioned, you can make your own system like this only uh like this as well using just a tool like NA10. I don't know if I have if you guys have used it or not. Anyone you who have used NA10 tool basic system it will choose TensorFlow Google. I'm a beginner only. Yes, TensorFlow with Google Collab will work. Okay. If I'm not used any tool, you can you can go go for it for sure. Okay, there are lot of lot of tools already available that will help you do make aentki applications. There's a tool called as uh crew.ai. Okay, there's tool called as NAN as I've just mentioned you. Okay, so Agent AI will help you do a lot of things on top of the existing generative A as well. Okay, so here you can see out of all these thing. Okay, agentic is again a subset of generative air. Okay, out of all these things, this is where most of the research nowadays going on. Okay, tool in the chat. Yes. So the tool name, let me write the tools here for you. So the tool is we have true.ai, we have uh and it we have a lot of other tools as well. So those like if I'm uh if I got if I forgot any of the tool make sure to share it in chats as well. Okay. What are libraries I'm going to study? We are going to discuss about pandas numpy uh skarn and uh we are going to do some visualization as well. So for that purpose uh I'm going to use uh like mattplot lab or c1 plot probably. So yes we are going to use these these all libraries and then we'll see what's our important functions you want to import. So I'll go to the specific library and import it then and there only. Okay. So this is what we have clear. So in agentic we have algorithms only. We have AI algorithms only. Uh means we have generative AI only. But these algorithms are so great that can they that they can take decisions on behalf of you. This is what agent is. So this is what we have in general. Okay. So, Olama, Grock, API, hugging faces, these are a lot of other tools as well that you can use to run like for example, is another one which will help you uh run the deep learning and those kind of models. But uh if you want more session on that detail on specific to generative AI because I can see there are a lot of people very interesting on generative AI. So do let me know. uh we can also have a discussion like we can also have a say same 3-day kind of a workshop on generative a as well or we can pick any of the library for example just hugging face and we can have a discussion 3-day workshop specifically on hugging face and then on generative a uh and then we can have a discussion about that as well so do let me know we can have a discussion on that as well okay so uh sir can we discuss about data pre-processing and feature engineering yes we are going to discuss about data prep-processing and feature engineering as well. Uh not in that much detail if you want to dig deeper into data analysis part. So for data analysis we already had done a workshop probably last week. So you can go to YouTube and check it out. Uh so okay we'll we'll plan it then okay we'll plan hugging face or probably I'll see what should be the flow. But we'll definitely have uh a session on generative I we'll we'll see how the generative I can work. Okay now once this part is done this is this is what we have in general. Now what I'm going to do I'm dig deep I'm going to dig deeper into just the machine learning part so that we can see what do we have in machine learning. Okay. Now we have seen that we have deep learning and all these things because this workshop is totally or basically mainly focused on deep mainly focused on machine learning. So let's see what do we have in machine learning apart from deep learning. So what do we have exactly in this particular region? Let's discuss about that. Okay. So in machine learning we have different kind of algorithms. Okay. Someone is being scraping to modeling not exactly starting from scraping. Uh we will uh not scrape the data from the internet. That's something we should not do uh when uh we are uh on live YouTube as well. But yes scraping is something that can be done. It will make the things uh more open for you. So that you can like from scraping the data to cleaning to pre-processing to analyze and to build a model and then uh storing it probably depend upon the use case that's definitely will add an extra layer of it. Okay, TensorFlow and KAS not in this particular workshop because this is more on the machine learning side. TensorFlow and KAS is more on deep learning generative AI side. Okay. So now let's see what do we have in machine learning. In machine learning we have different kind of algorithms. Okay. So some of you might have heard about these algorithms as well. In machine learning we have algorithm which is called as supervised learning algorithms. Okay. What else do we have? Let me know in the chat guys. We have supervised learning algorithm in machine learning. If I'm talking about we have supervised learning algorithm we have unsupervised learning that is that is correct okay and what else do we have okay sorry okay sorry again so we have supervised learning algorithm we have unsupervised learning algorithms and we have a reinforcement learning algorithm now tell me one very simple thing What is the difference between supervised, unsupervised and reinforcement learning algorithm? There must be some difference, right? You might have heard about this term that we have supervised, we have unsupervised and we have reinforcement learning algorithm. And in today's in tomorrow's class we'll discuss about what are the different algorithms do we have in all these parts as well because in supervised learning we have some set of algorithms that are different from unsupervised learning algorithm that are different from reinforcement learning algorithms and for now this particular diagram is not perfectly not perfect because you might be thinking algorithms no there are not any other algorithm here but I don't know how to properly display it here. So that's why I have made three different diagrams here, three different uh circles here. But right now in learning we have supervised unsupervised and reinforcement learning. Okay. So so in supervised learning a lot of people are mentioning in supervised learning we have labeled data. Yes, that is that is correct. In supervised learning we have labeled data. So in unsupervised learning we might have unlabelled data and that is correct as well. So supervised learning labelled data, unsupervised learning, unlabelled data. Okay, let's not talk about reinforcement learning right now. But yes, reinforcement learning definition is also correct. Reward and penalty based system that we have. Okay, what do we mean when you say about labeled data? Because everything comes under data only. The way you mention labeled and unlabelled data the same kind of algorithm we also have here. Okay. So now tell me what is what is labeled data. Okay. So in label data we have data that is already labeled. So you might be thinking okay wow this is very well defined this thing definition that I have given that in label data I have data that is labeled. What is label data exactly? Okay. So first thing is we have supervised learning algorithm means we are supervising the algorithms to learn. Okay. And whenever I'm saying supervising the algorithms to learn I'm saying this. So basically I'll do something like this. I'll give this particular image to my model. I'll click an image of this. This is a phone. Click the image and I will tell the algorithm that this is how a phone looks. This is a phone. And then it'll give another image that this is a remote. I'm not only giving the image, I'm also telling this is the image of a phone. This is the image. This is the image of a book. Right? So this is something that I'll I'll do. I'm supervising the algorithm that this is the input and this is the output. This is the input and the output is phone. This is the input. The output is that is the input. The output is a a book. Right? So this is what we have in label data. I have inputs as well as outputs. But in unsupervised learning, we just have inputs not the output. Okay, you can think of it as and as some other way. Imagine I have a folder. Okay, so let me make a folder for you. And inside this particular folder, okay, I have the images of cats and dogs. Okay, inside just one single folder, I have given the images of cats and dogs. That's fine. Right now, I I'm doing the exact same thing once more, but this time I have two separate folders. Okay? And inside one folder, I have the images of cats and inside another folder, I have images of dogs. So, I have images of cats and dogs separately. Okay, this one has the images of cats. This one has the images of dogs. Okay, so here are two ways. I can either have one single folder with the images of cat and dogs both and another option where I'm saving the images of just the cats in one folder and dogs in another folder. Okay. So whenever I'm going for this particular approach right so in this particular approach whenever this this data of two separate folders one having just cat another having just dogs given to the machine learning algorithm this is called as supervised learning approach because I'm teaching the algorithm that this is how a cat looks this is how a dog looks but in this particular way Because here both of them if I'm talking about cats or the dogs both of them are in one single folder so algorithm will not understand because for the algorithm I'm talking about the algorithm I'm training from scratch okay I'm not talking about because that is already trained but if the algorithm is not trained in the beginning we call it plus unlabelled data because we just have features. We just have images in the folder in one single folder or algorithm images for the algorithm. These are just two separate categories though I know that this is cat and dog because I know I've trained myself. Okay. But in this particular case and in this particular case as well the algorithm will not know this is category one algorithm go algorithm don't know that this is a cat for algorithm this is category 1 which is different from category 2 right so in this particular case if I'm talking about so whenever I have this kind of data Means I have inputs, I have images as well as their output as well as well as their labels. I call it as labeled data. Labeled data. Okay. And whenever I just have inputs not the output, I call it as unlabeled data. Okay. Most of the research or most of the projects you are going to make are going to be on supervised learning way are going to be on labelled data right because now we have enough data for a lot of use cases so most of the project you are going to make is going to be on supervised learning algorithms right so whenever I'm saying supervised learning this is what do I mean to say in supervised learning we have labeled the data in unsupervised learning we have unlabeled data. Okay. So let me mention them side by side so that it would be it would be easier for you to understand them as well. Right? So in supervised learning we have labeled data and in unsupervised learning we have unlabelled data. Okay. So far so good. This is going fine. Now anyone having any doubt in this do let me know. Okay. So in supervised learning we have labeled data. So this is we have inputs as well as outputs. And here we have unlabelled data. So this is this is how it looks. Now what do we have in reinforcement learning? Now which are the different kind of algorithms we have in supervised learning that you might have heard about linear regression, logistic regression. Uh we might have heard about SBM, decision tree, random forest, right? We are going to discuss about this okay in the next lecture but not in this lecture but sorry this is what we have supervised learning this is what we have unsupervised learning we have unsup uh unlabelled data okay what do we have in reinforcement learning okay what do we have exactly in reinforcement learning okay marks I I don't know like uh like how exactly the tendance thing will work. So Shiv Kotura is mentioning it's a feedback and reward kind of a system. Yeah, that is correct. Anyone want to give it give more brief about it? What is the reinforcement learning then? Right. So in reinforcement learning uh a use binary language label and cannot understand human language and label that can be translated. Reinforcement is binary for two binary languages. uh uh uh use binary language. Yeah, the definition is correct. But the definition of unlabeled and labeled is something uh something not correct puzzle. Okay. So now uh train and error uh learn through trial and error. Yeah, that's that's the perfect example. Uh the perfect example you can say is just how a a kid how uh how kids start walking. How a kid learns how to walk. So in the beginning the kid don't know how to walk right then by manually going through couple of not only just hundred but thousands of time uh it will start knowing how to set okay and after going for couple of thousand more iteration the kid will stand one day and then will take the first step then it will take the second step and this is how it start starts to walk and then uh after a couple of iterations uh the kid the or to walk and run, right? And then the kid can climb as well, right? So how the kid learn it's it's it's going for a lot and lot of iteration. Okay? Do we do the kid need to have any data like this is how we need to walk, this is how not you need to walk. The answer is no. Right? Because a lot of things are in our genes, right? Like who will teach a deer a just born deer that this is how you need to walk? A a deer can walk exactly immediately after uh the deer is born, right? Same goes for a lot of species unlike human beings because there's a lot of evolution that happen in between because of that like the kids are immaturely born but but that's a different story altogether. We can have a discussion on this on some other day. Uh but same for giraffe that that's also correct but for human beings it's not like that because yes that's a different set of a story. We can have a discussion on this on some other day. Okay. So in reinforcement learning we have the same kind of a system. A system that can play with itself will go for a reward policy like this is a positive step this is a negative step and it will try to maximize that positive step. This is how we started walking. Okay. But usually if I'm talking about uh a system so this is how it works. Uh reinforcement learning GFG. I'll get some article on GFG. So let me mention this particular diagram I was talking about. Okay. So I'll be having an agent which is probably uh which I call it as the kid itself. Okay. Then kid is playing with the environment. Environment is the environment itself. So on the kid uh so kid will make some random moves. Okay. And it seem like this is this is how we trained pets as well. Okay. This is how we train train pets. So basically uh whenever uh the pet is doing a correct move I'm going to give them a reward probably a cookie right and whenever the the the pet is not exactly doing whatever I want so what I can do I I will not provide them with the cookies or I'll not give them reward I can give them negative reward as well but for now let's treat it as I'm not going to give them reward okay so the task of this agent the task of the pet is to get the maximum reward. Right? So the pet will play in the environment and make some actions. If the actions are correct, the interpreter B for example is going to give them the positive reward otherwise negative reward. This is how it works. This is how this particular part works. Okay, this is how reinforcement learning works. So in reinforcement learning we have a environment agent loop that keep on happening. So first of all it will go to the first iteration then the second one then the third one and keep on doing it. Okay in Ashi Jang study was image created. Yes I'm going to show it show it to you. I'm going to wrap up the session probably two or three minutes. So yes after that I'm like at the end of the session we are going to have a discussion on this. Okay what exactly it is generated. So this is what reinforcement learning is. So in reinforcement learning we have an agent environment loop. Now the fun is in this agent environment loop. Do we need any data? In this agent environment loop, we don't have any data. We don't need any external data. We have sensory data but that's something that is generated throughout the process. Right? It is internal generated data. The agent will take random moves and whenever it's taking random moves if you you you are into a little bit of gaming or knows how game works how vector works so you know every action comes with a vector moving numbers moving okay so this is sensory data so this is the sensory data that the agent is generating okay a lot of game playing algorithm use this reinforcement learning approach now if I'm talking about deep sea carbon deep carbon used the same kind of approach Not exactly in this particular way but they have also used kind of the same thing. Okay. How this reward system works in these models. This interpreter is the reward system. The interpreter will give them positive or negative reward. Okay. So agent playing with this environment and this is called as the reward policy. Right? And the work of the agent is to maximize the rewards. Okay. So if I go back here. So we have supervised learning where we have labeled data. In unsupervised learning we have unlabeled data. In reinforcement learning we don't have external data. You might be thinking if you don't have the data machine learning is all about datadriven algorithms. Why don't we have reinforcement learning under AI? They must be in AI right? But because machine learning is all about datadriven algorithms. But here if I'm talking about this particular subset of machine learning, we have reinforcement learning here. Why we have reinforcement learning here? Because reinforcement learning algorithms are datadriven algorithms only. But the data is not external. It is internal data that the algorithm is generating by itself. The agent is generating by itself by playing with it by okay. So this is what we have. So just to make the life a little bit easier for you, your life a little bit easier for you. I'm copying and pasting the subage here. Allow this and this is what we have. Okay. So basically this is what we have discussed in this class. We started with introduction to machine learning. But to know machine learning first of all you need to know about all these different components that we have. So in AI we have all the algorithms that we have that can make decisions intelligent moves. In machine learning we have subset of AI. So I will having I will be having couple of algorithms but those algorithms are datadriven. Now what are the subset of machine learning? We have deep learning here. Okay. And in deep learning we use specific kind of algorithm which is called as neural network. This exactly how we learn. Not exactly like we learn because in our brain what exactly is happening no one exactly knows. There are lot of uh research being done. There are a lot of things that we have done as well. There are a lot of achievement we have also uh we have also got on top of it. But so far we have not make a sophisticated system as our human brain. We cannot save so far we have not have any system that can save all the data for our brain and save it somewhere right so that's a to that's a very sophisticated system but we'll try to mimic the same kind of behavior that we have brain we have neurons millions or billions of them so we'll take the same kind of architecture pass the data through them and uh see what happens and that work great so in deep learning we use neural networks neural networks if you are applying on NLP We have specific kind of neural networks comes under natural language processing. Applying deep learning algorithm architectures on vision kind of a data images, videos, I call it as computer vision. And if I'm training enough big data sets uh that can generate something new, we have generate API. And to generate something new, we need a lot of data. Imagine Chad GPT has the knowledge of everything. A lot of the things means almost all the things that that we have. So to have the knowledge of almost all the thing it must be trained on all the data of the internet. Okay. And usually in a company whenever you are making a machine learning or deep learning project you will be having just couple of thousands or sometime couple of millions of rows to train on. But here we have billions or trillions of rows because the data is in general very vast. Right? So this is what generative AI is. And now we have agents that can make decision on behalf of you. Okay. And there are a couple of tools as well. For example, uh like those who are into development, web development, they they might have used a tool called as uh cursor AI. Okay. There is another tool uh that you can use which is called as lovable AI. There is another tool uh that you might have used, some of you might have used uh which is called as Sir studio sir. Yes sir. Multinomial logistic regression is a supervised or unsupervised which one sir uh which one you're talking about? Multinnomial log logistic regression. Uh multinnomial logistic regression is uh so your question is it supervised or unsupervised? Uh it is a part of machine learning algorithm. No it is a part of machine learning only. So as you discuss what is supervised, unsupervised or reinforcement learning. So it is uh it is in which part supervised, unsupervised. Okay. So uh in tomorrow's class we are going to discuss about what are the different algorithms we have in supervised unsupervised and reinforcement learning. But if I'm talking about linear regression and logistic regression, these are the algorithm that comes under supervised algorithms. Supervised learning algorithms. Okay sir. Okay sir. Thank you. Okay. Because in supervised as well we have two more components classification and regression. That's not something we have uh discussed about it. Uh but yes we are going to discuss in the next class. So in the next class we our main task is to discuss about uh supervised learning algorithm. What are the different algorithms do we have? Uh we'll take some of them. We'll dig deeper into the maths behind it and we'll write the code on top of it as well. and the class a day tomorrow uh day after tomorrow we'll be building a one full-fledged project uh on this okay so this is what we have and uh to show you this particular to share this particular whole content that we have discussed so far so I'm copying this whole thing copy it and I'm pasting in the chats here okay so this is nothing but a whole uh what I would say lecture notes for you so this This is what we have discussed. Okay. Will be 1 hour long. Okay. It will be a little bit more longer than 1 hour. It will be 1.5 to 2 hours. Okay. So that's it. And uh if not right now if you have any discussion uh probably in future. So I'm also sharing my LinkedIn with you guys as well. So you guys can go to Google search Ashish Changra and this is the link. Uh sorry this is the GitHub link this is where you can go okay Ashi Janga geeks for geeks and you guys can connect with me over there as well okay and let me share the chat link here as well so that would be easier for you okay so that is it and let's go to this particular part and let's see what it has generated image how are it going to generate update the scene because it violate the content policy I don't know why it is violating ing the policy. Okay. Uh go ahead. I don't know why it is saying that it is violating the content policy or some something like that. But it must not do it. And here you can see now it has started generating. Okay. So let's wait for like half a minute more and let's see what exactly the output is because also I'm very curious to see as well. So at the meantime if anyone is having any question, any doubt do let me know. Okay. Uh uh hello sir. Yes. Uh sir actually I need a one question it sir in this three days of online this ML program can we get the certificate also after this course after this completion of workshop okay we get yes yes yes uh like I'm not pretty sure but you will get I'm I'm pretty I'm not pretty sure but please go to the website and check there there must be some certificate like as well the completion one okay please go to the website and check once okay or otherwise in tomorrow's class I'll go through the website and and let you know about because there are a lot of people uh telling me about uh the class attendance and things like that so I will come up with come up prepared with whatever next coming so I'll go through the non- tech question and doubts here as well okay yes sir okay sir thank you these classes will theory only. No, this particular class is theoretical. Uh tomorrow's class and day after tomorrow's class. Tomorrow's class as well we will be having a bit of theory. Uh because obviously we will go through all the different kind of algorithms. For that purpose I need to know first of all what are the different algorithms do we have and after that we'll be discussing about uh the hands-on part as well the coding part. Okay sir. Yes. Uh sir actually I am from CSC blockchain specialization. I have I have updated my first year. So sir can you please tell me about what should I learn on AI so that I can do something in my specialization especially in Ethereum blockchain crypto currency. So can you please just tell okay I want to know about this. Oh it's a very fast field first of all right? uh because uh in web development there are a lot of technologies coming in uh same goes for AI same goes for app development there are a lot of things that are emerging nowadays so first of all like uh because you are in just first year explore all the fields explore what exactly do we have in web development try do all the things okay but make sure that also focus on AI okay how you can do all these things but using AI you can do the exact same thing in a lot lesser time okay because in first year you have enough time right if you're in third or a fourth year I might have like recommend you in a different way so in third and a fourth year your main task would be go for one specialization and dig deeper into it okay probably take if if your your task is to go for AI so go for AI go for machine learning algorithm deep learning algorithm but don't spend too much time dust on this because we have a lot of a tools uh generative a tool that can write the things uh for you. Okay. So your task is not to do not to do the hard work but to the smart work right. So those who are uh and in the last year probably are starting their career into it then there might be a different things but rest of the way like you like I I hope I answer your question okay required for tomorrow's session just Google collab will work that's it you don't need anything else thank you sir thank you sir thank you for explaining required for machine learning engineering yes a little bit of DSA is required less dictionaries data prep-processing data cleaning and So yeah that much is required rest of the thing can be can be easily handled. So these are the images that is generated. I'm not pretty uh much sure if they are correct or not but this is something that I generated. So it is all the things that I wanted. Okay if this person looks exactly like me or not. I am not pretty sure but I'm pretty sure it's not looking like me but okay this something that is generated right and uh some web frameworks. What can you suggest some web frameworks sir? Frameworks. Uh which frameworks like in Python like web framework. Okay. Web frameworks. So Python web frameworks you can go for fast AP something you can go with you can go for Django. There are a lot of frameworks that we have it or you need to uh you need to be be updated about it like what are the latest and the fastest tools that we have in market if right now uh if you are going back to flask or things like uh flask or basically streamllet it's very easy to do but they are not something that I use in the industry so search on the internet what are the latest tools that we are coming up what are the latest frameworks that are coming up now that that you can to fast the process So but one I mentioned is you can use fast API. Okay. So flask for prototyping Django is there. Yes, there are a lot of them. Okay. So thank you so much everyone for joining in and uh I hope uh you guys have enjoyed this lecture. Make sure to give your your feedbacks here as well. And uh in tomorrow's session and day after tomorrow, tomorrow's session as I we have mentioned we are going to discuss about what are the different machine learning algorithms we have. We'll dig deeper into them the maths behind it and we'll see if we will take an algorithm and build it from scratch so that you can have a detailed understanding on of you might have heard about like open air has released a model that has a billion parameter, five billion parameter, 10 billion parameter. what are these parameters exactly and we will try to make an algorithm that has just one single parameter okay and that too by very basic maths once we have it then we'll dig deeper into like 5 10 and different on top of it but we'll see what exactly has to be done and the day after tomorrow we'll be having a full-fledged uh project that we are going to take hello sir yes uh so I want to like start learning about data science So where can I start from like uh can you provide any road map like which the current industries are looking for technology? So uh technology wise if I'm talking about data science so it is a three three-step process okay first step is data analysis okay because if I'm talking about data science it's all about building algorithms on top of uh or building predictions on top of existing data. So first of all you need to know data analysis uh which includes SQL which includes Python which include data analysis of Python and not only data analysis it comes with a lot of things as well because data exploration uh feature engineering and data analysis data visualization and on top of that you need to know machine learning algorithms how algorithms can be applied on these uh on these things. Okay. On top of that, I'll go companies will also ask for deep learning. Uh and on top of it that depends upon company to company. If they want uh the generative AI part as well, if they want deployment as well, if they want like MLOps, if what are their requirement best thing you can do is uh and that's the industry ready thing that will make industry ready as well. Go to any of the job providing platform be it uh glassd door be it no.com be geeks for geeks beat any platform or you can ask well go to the company's profile go to the description what are their requirements okay are they want like do they want just machine learning or big data or data analysis or python or libraries or transformers or hugging face or or crew.ai AI what exactly do they want okay go to 10 to 20 different profiles and you will get the common points what exactly do they want that is exactly what you need to know okay and once you have all these points just make them in a flow and start learning and start learning by making projects okay don't just keep on watching the videos start making it this is how you learn okay okay thank you sir okay so thank you so much everyone for joining I have one question Okay. Okay. Okay. Let's Okay. I'll take one more and then we'll we'll take rest of the question tomorrow session. Okay. Yes. Go ahead. Uh so uh as we seen the machine learning we have a lot of theoretical concepts and algorithms. So I basically have uh knowledge and currently learn in a front end development. So how front end developer uh helps to machine learning? front end developer machine learning need. Okay. So first thing is front end developer machine learning right because uh basic application with the back end uh and if you are just the front end developers uh so back because front end is something uh uh that there are a lot of AI tools uh that can make the front end that that can make great front by the way. Okay. So This is a personal thought and there are a lot of AI tools that are already available in front end as well. So the but back end is something that will take some time to embed some some great tools that can replicate that part that can overcome that part as well. So uh so this is this is one part okay machine learning deep learning projects they're basically models that has to be deployed somewhere finally okay and whenever these models are deployed on the servers eventually okay so machine learning you want your career to be made in web development only then only start taking some pre-built models up model train fetch some models that are already trained and try to build an application on top of it okay that is the deployment part of that so make sure that you know the deployment part of machine learning or deep learning and that's not only machine learning deep learning in generative way as well generative deployment because now we have LLM tools this is just the API calls that we are doing from the back end. So that is something that uh that you should be work on which project uh deployment how you can integrate everything. Yes. Where to deploy ML model? There are lot of uh uh platforms available. Uh we'll try to do it using streamllet but you can also host it on AWS uh which is a more scalable part. Okay. How to get projects from GitHub GitHub AB projects like whatever projects we have we are going to deploy it on GitHub as well. So remaining questions I'm going to take uh in the next class. So and if not that you guys can connect with me on LinkedIn as well. So thank you so much everyone for joining in and I'll see you in next class then take care everyone. Bye-bye. Have a nice not day but okay sir sir. Yes sir. Uh sir can you just reshare the notes please? like the link. Just give me a second. Okay. Thank you, [Music] sir. And these notes are pretty pretty much straight part. These are nothing but a vein diagram only that we have. Okay. So, I have shared it with you again. Thank you so much everyone and I'll see you tomorrow same time. Bye-bye. Bye, son.