AWS sagemaker introduction | AWS sagemaker tutorial | AWS sagemaker tutorial for beginners

so guys finally we are starting the core machine learning and data science components of AWS OK and we are starting the first tutorial on Sage maker let us go and see what is this stage maker all about if you go inside the AWS console and if you click on services you will see something like machine learning inside machine learning you will see many components and most important component that we are going to learn today is Amazon Sage maker okay so here you can see build train and deploy machine learning models okay this is one of the oldest and most used component for data science point of view so let's see what we are going to learn in this session guys first of all we are going to learn what is Sage maker basics of that then what are the various features that sagemaker offers okay not from only sagemaker point of view but in AWS from data science and machine learning point of view what all important things and most used things are being offered okay as you can see here there are a lot of features that is being offered Amazon poly and Panorama and omics kind of thing but not everything is being used that much okay so we will see what important features sagemaker offers and we use then there are two things in sagemaker one is called notebook instance another is called Studio okay so we will see what is the difference and how they are you know used for different purposes after that we will see how to launch a studio demo and how to launch a notebook instance okay and then I'm going to show you how you can be careful about the charges so that you don't end up paying any money to the Amazon unnecessarily okay so guys what what I want to show you here is what is Sage maker first of all okay so as you could see in the console itself it says that it is you know fully managed it is for building training and deploying machine learning models right built and deploy machine learning models but as you know right this is a cloud offering so it will be giving lot of things to us I mean it will make our life easy so how sagemaker makes a data scientist's life easy let's try to see which some of the features I have listed here features not everything but some of the very important features and that we use most of the times okay so as I told you prepare build train and deploy four important stages if we take of a machine learning pipeline right let's see some of the features that you will use mostly and it's important to know first feature is ground truth so what ground truth does is I will give you one simple example here suppose you want to do a sentiment analysis on Twitter okay so you collect data this is your Tweet one this is your Tweet two this is your Tweet three okay but you do not have normally a Target column you want to you know do a classification where you want to say this is positive sentiment this is negative sentiment but normally you don't have this in real world so it's a big challenge for data scientists to create these levels okay and this is nothing but ground truth so what ground truth will do is it will enable you to create labels for your data that is one feature of it okay this one example I gave for test it can be image based levels it can be different kind of labels okay so that is how ground truth help you to prepare data for your machine learning now what clarified us clarify is a tool that enables user to uh you know see how their data how their model is getting biased or how the data is behaving okay so it will detect bias and understand model predictions in a much better way that you and me cannot do you know that systematic level of analysis okay I will show you these things in upcoming sessions I am just trying to tell you what are important features you should be aware of okay data Wrangler makes your life easy in aggregation uh kind of sorting kind of filtering kind of joining so it it makes your life very easy in all those data preparation kind of activities okay now what is jump start in build phase coming to the build phase there is something called as a jump start jump start enables user who are not too technical to build their machine learning model suppose somebody who does not want to write code who is not too technical but they want to build their machine learning model by GUI kind of interface okay so this jump start will enable them for Discovery training and deployment of model using solution example some graphical user interface kind of thing okay what is this autopilot does autopilot suppose you do not want to do lot of hard work in building your model okay so autopilot will do automatically create models and it will give you visibility what your model is doing what your data is saying so make your model you can say building life easy okay and then AWS gives you lot of built-in algorithms built-in notebooks built-in algorithms which you can use to train your model to build your model for example if you say I need XT boost so XG boost built-in algorithm with some good features is already there that you can use okay so you can also bring your own Trend model your own Docker container everything you can use so many NLP and vision algorithms you will find here NLP kind of algorithm or computer vision latest algorithm you will find here so that way it makes your life easy now let's see some things in the train services so in train Services there is something known as experiments guys what experiments we'll do is it will enable you to track visualize and share model artifacts across team so what this means is you for example you can create a uh you know experiment here you can say this is my data import next you can say this is my data cleaning next you can say this is my you know level labeling so this way you can create an experiment and whatever your model artifacts gets generated it can be shared experiments can be shared among your teams okay then comes manage training what manage training does is AWS will take care of your training fully managed training it will do it will allocate the resources it will it will ensure that you know cheap resources are are being used so mostly this say that you know ninety percent of training cost they reduce if they fully manage your training services okay then comes the distributed training suppose you have very very large deep learning model very very large data set so distributed training is one area which will enable you to train your model in a distributed kind of environment all these things I am giving you one two liners knowingly guys because you must understand what are the basics of these things okay there are many things but I'm not covering I'm just covering few things to give you an idea now what is in deploy in deploy you can do various kinds of monitoring okay so model monitoring will give you access to maintain accuracy of the deploy models if your model is underperforming so it will enable you to have a look at that and take action on that then in in deploy various endpoints are facilitated by AWS okay so endpoints can be like multi-model in points or endpoints can be like a multi-container endpoints so all of it are for different different purposes for reducing cost by multiple models at per instance or however you want to do it okay so when we deploy the model we can cover these things then the last inference so you can have multiple kinds of inference you can have real time inference if you have a STD traffic patterns you can have a serverless inference for intermittent traffic patterns so depending on what kind of traffic patterns you have for your model right every time you don't want to use your model right maybe somebody wants to use once in a week somebody wants to use daily somebody wants to keep it running 24 cross seven based on that you can have different kind of inferences okay so AWS supports that aspect as well so as I told you in prepare build train and deploy these are some of the things that is most used and you must be aware of that now let's go ahead and see some basic differences between notebook institution notebook Studio I am sorry sagemaker Studio okay so here what you have to understand is notebook instance is something which is very easy to launch okay easy to build and launch on the other hand if you see Studio it is also easy to launch but there has to be some configuration done okay some configs needed to launch okay some configs needed now this studio you can treat this as complete IDE that you can use integrated development environment and this notebook instance you can you can just treat it as it will run inside your AWS console inside console a Jupiter notebook will open and you can work on that notebook okay one important thing to note here guys your resources to notebook instance will get allocated by AWS itself at the moment you are launching or running the notebook okay so resource allocation will be taken care by AWS so I can say here fully managed okay this is fully managed by AWS you no need to do anything that is where it is easy to launch and easy to work with but most of the time we generally work in a team in a collaborative environment and we want a ID kind of setup with gitlab integration and many other things right GitHub integration many other things and here what will happen is you may need to manage few things by your own okay so some of the things you need to do self-management fine and that is where uh though it provides kind of more features Studio it is uh heavy for the management purpose notebook instance is quick and on the go it will allocate resource and it will start working let me show you how these two things are different in the uh Management console okay a quick demo for these two things so if you open Amazon sagemaker right so I have clicked on Sage Maker Now and it is opening so if you can see here in Sage maker on the left hand side you will see studio and studio lab okay two things options are there and then here if you come see in ground truth I was showing you guys how to how you can label the jobs in ground truth labeling jobs leveling data set so this will help you to create you know more better data for your model synthetic data Etc so first of all notebook instance I want to show you let's go to notebook instance this is an instance I launched but let me let me launch one more instance for demo purpose okay create notebook instance you can give any name My Demo instance okay it will ask you what kind of instance you want to create so what kind of server it will attach in the background elastic inference and what kind of platform you want to create okay so there are various options you can go with lab 3 Lab 3 the latest let's go with lab 3 okay and then you can simply come here keep all the default option and say create notebook instance okay so if you see here notebook instance is pending so it will start in some time but let me let me start a previously built notebook okay so let me start this the first one which I created now will start in some time but this my knot I had created some time back so let me create it now let me try to start it now so both are bending status so before that we can open it and see few things here this is your AR and AR and my Arn means your Amazon resource name or unique name with which Amazon identifies this okay at the moment I created it addressed 5 GB of EVS volume with my notebook okay and then it it attached this platform okay if this pending becomes running which is which is happening now then you can start either Jupiter or Jupiter lab and you can work with that like you work in your local Jupiter notebook okay but there are few limitations with notebook here it's not a fully managed you know full integrated environment kind of thing okay so there are many things which you need to be careful when you are sharing between your teams so suppose you are working on one notebook instance somebody else in your team is working on other notebook instance so when you share your work right then there has to be a few things to be taken care which I will discuss later both these notebooks instances are starting let it start meanwhile I will show you uh what setting you need to do to launch a studio okay so I will click on Studio and if you can see here if you are doing it for the first time it will it will show you here create domain domain is nothing you can give any username there okay so I gave one full data science and once you create domain it will tell you to create a user so just you can say default user I have created two default users okay this is my one default user this is my other default user so what two things you need to do guys you need to create a domain and you need to just go ahead with the default user okay and here this studio is launching this is Amazon sagemaker Studio that I was telling you this is a IDE kind of environment okay on the other hand if you see notebook instances this is not ID this is a simple Jupiter notebook that you can work like you work in your local machine okay so it takes some time to start the first one I created now so it is taking I understand but second one should not take this much of time anyway let's wait and see so as you can see both the notebook instance have been started now and once it starts it will show like this in service and the meaning of in Services if I let's say train a model here now and if I use it for some purpose there is a ec2 instance that is supporting me in the background and I will be charged for that based on whatever service I am using okay so be very careful on what you are doing when you have started a notebook instance okay I am I have launched the Jupiter lab from there so a Jupiter lab will open I mean Jupiter notebook kind of environment and here you can see I had launched the studio itself right so Studio has also started so as I was telling you studio will look like ID this is the look of the studio and notebook instance will look like a normal Jupiter notebook so if I go here and I can launch here so this is my first you know Jupiter notebook I can write my python code here okay and then there are many things we can do but just to show you how to create and how to launch a Lab Studio and how to launch a Jupiter notebook instance remember guys always if you launch this and keep it open Studio then there is no problem because nothing is running in background as of now but if you keep a notebook instance open and leave it like that that's a problem okay so go here and say stop and go here and say stop always make it a habit of stopping if you are not using it otherwise you may be unnecessarily charged for that so in our next video we will see how we can use Jupiter instance and Studio left for doing python experimentation and whatever features or whatever things we discussed here in various aspects some of these things are very very important and I will try to cover some of these things in upcoming lectures okay and I am also going to show you how you can deploy your model and create endpoints in the sagemaker kind of environment same code will run in studio and notebook instance both there will not be any difference in the code so please go ahead and press the like button guys if you're liking AWS series and please don't forget to subscribe to the channel I'll see you all in the next video wherever you are stay safe and take care

Channel: Unfold Data Science

Views: 19,973

Keywords: AWS sagemaker introduction, AWS sagemaker tutorial, AWS sagemaker tutorial for beginners, AWS sagemaker studio lab, AWS sagemaker notebook, AWS sagemaker demo, AWS sagemaker machine learning, AWS sagemaker important components, AWS sagemaker, aws sagemaker deploy model, sagemaker, sagemaker tutorial, sagemaker studio, sagemaker pipeline, training jobs, aws sagemaker training jobs, aws sagemaker demo, aws sagemaker in hindi, aws sagemaker telugu, aws sagemaker tamil

Id: agq-C4XyL3E

Length: 16min 19sec (979 seconds)

Published: Thu Dec 15 2022