Describe fundamental principles of machine learning on Azure

11. Optional - Lab - Creating an Azure Virtual Machine

So here I am back in is y'all. So just for the purpose of explaining to students what the entire purpose of it as your virtual machine is So this is a service as simple as spinning up a virtual machine in Azure. So let's say that you want to have a server in place, a server that's built on, let's say, Windows Server 2019,or let's say you want a Linux server in place. Normally, you would need to invest in hardware in your on-premises data centre in your company to go ahead and have a server in place. But the best thing about the cloud is that you can actually go out and spin up something known as a virtual machine on the cloud, and you can go ahead and host your applications on this virtual machine. So I'm going to just go ahead and show you the creation of a virtual machine and then log into the virtual machine so that you have an understanding of that compute infrastructure when it comes to your machine learning. So I could go on to all the resources here I can again click on "Add." Now over here, you can go ahead and create a virtual machine based on Windows Server 2016 Data Center. This is one way of doing it. You can also go ahead on to the virtual machine section over here, and you can go ahead and add a new virtual machine from here itself. So let me go ahead and just choose the virtual machine option. So over here as usual, you have to go ahead and choose your subscription and you need to go ahead and choose a resource group. You can go ahead and give a name to the virtual machine. You can leave the location as is in terms of availability. Also, you can go ahead and leave ithazardous now in terms of the image. So over here we have these prebuilt images in place. So let's say you want a server that is based on Windows Server 2019 Data Center. You can go ahead and choose that. Here I can go ahead and choose the size of the virtual machine. So if I go ahead and click on "See all sizes," So there are different size classes in place. Depending upon the size class that you choose, a large number of virtual CPUs will be assigned to the virtual machine. That amount of RAM will be assigned to the virtual machine. And you have other features as well that become available on the virtual machine depending upon the size that you choose. So over here, based on the size I'm choosing, it's going to go ahead and allocate two virtual CPUs and eight gigs of memory. Over here, you can see what the estimated cost per month is even if you click on See All Sizes over here and if you scroll on to the right over here. Also, you should be able to see the estimated cost per month. This is only when it comes to the compute of the Virtual Machine. Then you can go ahead and specify something known as the administrator account. This account will be used to log into the Azure Virtual Machine. Now, there are some password restrictions in place. Over here you can see the password restrictions. So make sure that the password that you specify here caters to these different restrictions. Please keep a note if you're following along on what the password you are specifying over here. So I'm creating a password that's based on characters, uppercase and lowercase, and numbers as well. Now go ahead and leave all the other settings as they are. So again, this is a wizard in place for creating the Azure Virtual Machine. This basically specifies the discs that are getallocated on the Azure Virtual Machine. This wizard will also go ahead and automatically create an Azure Virtual Network. This network is required for the hosting of Azure Virtual Machines. So, this wizard will create everything for you. I'll leave everything as it is. I'll go onto management, leave everything as it is, go onto advance, leave everything as it is, go ontags, go on to review and create. And I'm going to go ahead and create this Virtual Machine.So I'll go ahead and hit "Create." Please note that over here you can see the estimated cost of the virtual machine per hour. So I can go ahead and hit on Create. Now, this particular wizard is going to go ahead and create multiple resources. It's going to go ahead and create a virtual machine, a virtual network, etc. The entire purpose of this video is just to give you an idea of a virtual machine in Azure. This is just for those students who are new to Azure who are taking this course. For those who are very familiar with Azure, I said they can actually go ahead and just skip this and move onto the next chapter where we're going to go ahead and expand the pipeline on as your machine learning. Now, this will take around 5 minutes. Let's come back once we have the deployment in place. Now, once you have the deployment in place, you can go ahead with the resource. Now, we have different aspects of the resource itself. Now, you can actually go ahead and connect on to the Virtual Machine.So you can go ahead and hit on Connect and choose RDP. We can download the RDP file. I can go ahead and choose that file. I can click on it, hit on Connect.So over here, I'm going to go ahead and add those administrator account details. Remember, we used or specified this in the wizard in the creation of the Virtual Machine. Let me go ahead and hit on okay. Let me go ahead and hit on yes. And now you can see that we're actually connecting on to a server in Azure. So on the server in Azure, which we currently have,you can do anything on the server if you want. You can install Microsoft SQL Server, you can install Oracle, you can install your workloads, and you can store web applications. You now have a server running in the cloud. So, when it comes to the Azure machine learning service, it will use these compute virtual machines to run your machine learning workloads. Because remember, in the end, to build your machine learning model, you need to go ahead and take that data, take the algorithm, do some processing, and build that function, that machine learning model. And that needs to be done on some sort of compute infrastructure. All of this will be managed by the Azure Machine Learning service. You have to go ahead and specify what to create in terms of the underlying compute infrastructure. It will go ahead and create the virtual machines. It will deploy Python if required. It will go ahead and use the Python language to basically do whatever work you specify as your machine learning pipeline. As I said, the entire purpose of this particular chapter is just to explain that as your machine learning service. Now, just in case, if you're following along, in order to ensure that you don't bear any cost for the Asia virtual machine, So let me go ahead and close this first. So I'll close my remote session. Now, if I go on to all resources, so I have a lot of resources in place, if I just filter on the resource group, that's the learning group, and I hit on apply. So over here you can see we have a lot of resources in place. So if you want to go ahead and delete the machine learning resource, you can actually go out and choose the virtual machine, the public IP address, the network security group networkinterface, the disk, and the virtual network. You can go ahead and delete these resources. So remember, in order to go ahead and avoid incurring costs in Azure, if you don't need anything,just go ahead and delete those resources. The other resources over here are basically part of your machine learning workspace. So we should go ahead and delete those resources. We have to go ahead and delete the virtual machine resource that we just created because we don't require it. I just want to go out and show you your virtual machine service, so I can go ahead and choose these resources and hit on delete over here. I can go ahead and hit "yes," and confirm the deletion of the resources after the time you'll get a notification on the deletion, in case all the resources don't get deleted at one time. So this can happen again. You can actually go ahead and just select the resource and hit delete, right? So this marks the end of this chapter.

12. Lab - Building a Classification Machine Learning Pipeline - Compute Target

We left off in the previous chapter with the creation of a new compute target. Now instead of actually creating the compute target over here, let me go ahead and hit on Cancel and let me go on to the computers section, which is in the managed section. Let me right click and open this up in a new tab. So I want to go ahead and create that compute infrastructure from here. So over here you can actually go aheadand create different types of compute infrastructure. So you have compute instances, compute clusters, you have inference clusters, and you have attached compute. Now, for the purpose of training a machine learning model,we have to go on to compute clusters, and you have to go ahead and create a new compute cluster. So the reason I'm creating it from here is because I want you to understand what we are trying to create when it comes to the underlying compute infrastructure. So over here, let me go ahead and hit on Create. Now over here, we can go ahead and leave everything as it is if we go ahead and scroll down. So for any virtual machine type, you can choose whether you want CPU or if you're using or if you need graphics processing, you can go ahead and choose GPU. Then you can go ahead and scroll down. So over here you have the different instance sizes. So over here, for this particular size, you have four cores and you have 14 gigs of RAM. So let me go ahead and choose that and let me go on next. You can also see what the cost per hour is. So I'll go on to the next. Now over here, we have to go ahead and give a name for the compute infrastructure. So I'll just give it a simple name. Now, this will actually go ahead and create multiple nodes. So the advantage of having multiple nodes is that the machine learning process itself when you're training the module can actually be displayed across multiple nodes. It makes the process of machine learning much faster. Because you see, in order to develop that machine learning model, if you have a large dataset and if you are trying different types of algorithms, this can all take time. So the underlying compute infrastructure needs to be scalable. It needs to have a high number of CPUs in order to basically process your data sets and use your algorithms. Now over here, since we just have a very simple data set in terms of the minimum number of nodes,let me go ahead and increase it by one. Please note that in order to save oncost, because currently over here, if you're not going to be using this compute infrastructure for building your machine learning model, then you can actually go ahead and choose the minimum number of nodes as zero. If I go ahead and scale it up to one,this will mean that this compute cluster will always have one virtual machine node that will be running. Now the reason I'm specifying it as one is because I want to go ahead and run this workload right now. I want to go ahead and build my machine learning model right now. That's why I'm going and specifying the minimum number of nodes to one. Over here, you can actually go ahead and decide the maximum number of nodes that you want. For the moment, I'll also leave it as one. I'll leave all the other settings alone and let me go ahead and hit on create. So now it's going to go ahead and create that compute cluster. This might take around four to five minutes. Let's come back once you have the cluster in place. Now after some time, I can see that I have my compute cluster in place over here. So now let's go back to our pipeline. So I'll go back to our authoring pipeline. Now over here, let me go ahead and click on select Compute target. So now over here, we can see our compute target in place. Let me go ahead and choose that and hit on Save. Even though in this particular pipeline we are not training any machine learning model, I am still going to go ahead and process this particular step of splitting the data. So at any point in time now, you can actually go ahead and use an underlying compute infrastructure to run your pipeline. So this is good when you want to go ahead and see if certain modules are working as desired. So over here, I want to go ahead and make sure that the split data module is working as desired. So let me go ahead and hit submit. Now for this, you have to go ahead and create an experiment, so that basically creates something known as a pipeline run. So let me go ahead and actually create a new experiment. Then let me go ahead and hit submit. So now if I go ahead and just close this, it's actually going to go ahead and basically run this pipeline on that particular compute cluster. So over here you can actually see the pipeline is running and now you can see this module is also running. So on this designer now on LiveView, you're actually seeing the running different steps of your pipeline. Let's come back once this is completed. Once the operation is completed, So over here, if you just hover around the points here,you can see you have data set one as a result. And over here you have data set two. So there are two points as the output for this particular phase of your pipeline. You can now go ahead and visualise your data sets by right clicking on the split datamodule. So in data set one, you can see that you have roughly around 230 rows. And in data set two, you are roughly close to around 100 rows. So, over here you have 70% of your data and over here you have 30% of your data in place. right? Let's now mark it in on this chapter, and let's complete our entire pipeline in the next chapter itself.

13. Lab - Building a Classification Machine Learning Pipeline – Completion

Now here we are back on the canvas. So, the last time we went through this process, we split our data into training and test data. Now I'm going to go on to model training. I'm going to go in and drag the Train model module onto the canvas. So over here, at the moment, let me go ahead and just close the property screen. Two things are critical when training a model using the concepts we learned earlier. One is your algorithm and the other is your data. So, first, let me take data set one over here and drag it onto one of the Train model module's sections. Next, we have to go ahead and also map our algorithm. This is our machine learning algorithm. So over here in machine learning algorithms, So we have algorithms when it comes to regression, but what we want is basically classification. So over here I'm going to go on to the classification of two class boosted decision trees. So let me go ahead and drag this over here. Now I'll leave all the parameters as it is.Now, there is a lot that goes into deciding the right machine learning algorithm. And it's not necessary that if you go ahead and choose a machine learning algorithm to go ahead and train a model, it will give you the perfect accuracy or the best accuracy. Normally, in the field of data science, you try to use different machine learning algorithms. So you might use one algorithm, train a model, and see the results. Then you might go ahead and use another algorithm, see if it actually gives you better results. So you have to go ahead and try out different machine learning algorithms. I'm using a very simple algorithm when it comes to classes, when it comes to binary classification. So I'm going to go ahead and take that algorithm and drag it onto the model. So now our model has an algorithm and it has our training data. Now I'm going to go on to the Train model property and over here I do go ahead and click on the edit column. So over here, I go ahead and decide on what is a label column. So what is our model trying to predict? So remember, when it comes to our data, so over here we can see all the columns of data. So remember, we are trying to protect the income where the income is less than or equal to whether it's greater than 50K. So this model needs to go ahead and look at all the data in the features and then see what is the best way to predict the income. So let me go ahead and choose this, and then let me go ahead and click on save. So now our model knows what it needs to be traded for. Now, in case you're not getting the columns over here, So when you click on edit columns, if you don't get the columns, it's because you have not gone ahead and connected the data set onto the Train model module, right? So we've done our training in place. Now what about our test data? Well, after the training of the model, we have to go out and evaluate the model itself. So what? We can do that as the next step. We can go ahead and do a score of the module. So I can go ahead and search for a score. So I'll go ahead and drag the score module onto the canvas. I'll go ahead and leave it as it is. So let me go ahead and take the training model and put it on the scoring module. And then we have to go ahead and evaluate our model now for scoring the model itself. So one is the output of the training model, and then we have to go ahead and also use our test data in order to go ahead and do a score of the model itself. And then, finally, we can go ahead and evaluate the results. Let me go ahead and choose the Evaluate model. Let me bring it over here. Let me close this and let me go ahead and connect the scoring model or the evaluating model. And now we have our pipeline in place. So now in our pipeline, we have everything in place. We have our data set, we have the spread of our data into training and test data. We're using our machine learning algorithm to go ahead and train our model based on the training data. Then we're going ahead and scoring the model based on the training and based on the test data. And then we have to go ahead and evaluate the results to make sure that the model is the best model possible based on the data that we have and based on a machine learning algorithm. Now let me go ahead and hit submit. Now you can go ahead and choose a different experiment or choose an existing one. Let me go ahead and choose the existing one and go ahead and run everything on this compute target. Let me go ahead and hit submit. Right, so this will take some time. So remember, again, all of this is going to run on that compute cluster again over here. You will see that it has now gone ahead and put this in the running state. Now that we have our data split, it is time to start training our model. So after a few seconds, you can now see it's running the machine learning algorithm over here, right? So let's come back once this is completed.

14. Lab - Building a Classification Machine Learning Pipeline – Results

Once the pipeline is complete, So over here in the pipeline, you have gone ahead and trained your model. You've also gone ahead and scored and evaluated the model itself. Let me go on to evaluate the model. Let me right click and go on to visualise and go on to the evaluation results. Now over here, actually, you'll see a lot of information when it comes to the evaluation results. So let me go through some of the important aspects. The first important aspect I want to go through is something known as the confusion matrix. So this is important from an exam perspective. So over here, I want to go ahead and explain the confusion matrix. So over here we have four quadrants in place, right? So we have the first quadrant,second, third, and fourth. So, these quadrants actually map onto these blocks. So over here we have the true positives. You have the false positives. You have the false negatives. You have the true negatives. So what does this mean? So remember that when evaluating the training model, it's using the test data that you fed into the evaluation to go ahead and check if the model is accurate or not in terms of its prediction. So what is the meaning of "true positives"? So over here, this means that the model, when it took one row or the rows of test data, actually predicted that the income was less or equal to the actual value of the label itself. It was actually less than or equal to 50K. Remember, in your data, whether it's your training or test data, you still have the labels in place. The only difference is that, remember, you have taken 70% to go ahead and train the model, and now you're using the remaining 30% to go ahead and test the model. So over here, the model was able to accurately identify those rows wherein the income would be less than or equal to 7022 values in which the prediction was positive. At the same time, it also successfully predicted that if the income was greater than fifty thousand, And this also mapped on to the actual value. The places where it failed are the false positives and the false negatives. So over here, in terms of the false positive,it actually predicted for 828 values or 828 rows that it was less than or equal to 50K. But then the labels showed greater than fifty k. And when it came to the false negatives over here, it actually predicted that the income would be greater than 50K, but in fact it was less than or equal to. This confusion matrix actually gives you an idea of how the model has performed. Please note that you could go ahead and use another machine learning algorithm and see if it gives a better representation of the confusion matrix. Because, remember, ideally, you should not have or at least have a lower number of false negatives and false positives over here. So, returning to our machine learning algorithm, remember that we use a two-class boosted decision tree. So maybe you might go ahead and replace this algorithm with another algorithm that is available over here and see how that performs and then compare the results. So algorithms can actually give you different results when it comes to training your model. Please understand that it depends upon the data that you actually give to the model. But we are giving an example that also depends upon your machine learning algorithm. Also remember, over here,this was a binary classification. So here we only had to go ahead and decide on two values, whether they were less than or equal to 50K, but you can also have multi-class classification. So, for example, in the case which I mentioned earlier, wherein you want to go ahead and predict whether a customer would buy either an economy class fare ticket for an airline, business class ticket, or first class ticket, So, when it comes to the confusionmatrix, when it comes to the actual, you have three of them right over here. So this will be the actual and over here you'll have economy, you'll have business and you'll have first class. What is the actuality? And over here also, you'll have three. So, as predicted, you'll have the economy, business, and first class. So again, in terms of the confusion matrix,it's applicable for binary and multiclass classification, right? So this is important when it comes to the exam to understand this confusion matrix. Now, apart from that, what is also important from a machine learning perspective is your Roc curve. So this is known as the receiver operating characteristic curve. So this gives you the relationship between the true positive rate and the false positive rate. Again, the true and false positive rates are coming from the data that we have. So if I go ahead and scroll up the data that we have over here in the confusion matrix, right, the data over here, that's how you get the Roc curve. So it's usually a good thing to have a curve that looks like this. So if it's closer to one, that means you have a model that predicts values much better, which gives your prediction much better information, and you have other information that is given by the model result visualization. So going back over here, let's go ahead and again visualise the results. If you go and scroll down over here, you can also see other aspects such as the accuracy, the precision, etc. So as a data scientist, you'll go ahead and take this result visualization, and you'll try other machine learning algorithms and see which one gives you the best results. Remember, over here we are using the machine learning type of classification. Remember, we have the classification type, we have the regression type, and we have clustering. For the purpose of the exam,we have to focus on classification. That is, you know, classifying your data, whether it's yes or no, that's binary classification. And then we have regression. So we'll again do the similar exercise today and see how to build an apipeline when it comes to regression.

ExamSnap's Microsoft AI-900 Practice Test Questions and Exam Dumps, study guide, and video training course are complicated in premium bundle. The Exam Updated are monitored by Industry Leading IT Trainers with over 15 years of experience, Microsoft AI-900 Exam Dumps and Practice Test Questions cover all the Exam Objectives to make sure you pass your exam easily.