Welcome back to my third class of Microsoft Azure Machine Learning Studio is just last time a little bit. I just made some mistakes. So we just went to fourth class. Sorry, certain class. Yeah, that's good. Sorry.

So wanted to make sure we are in the same inner loop so that you don't want to miss. So here in this, we know now that if you got familiar with this interface, so we know just those 232 classes who was for familiar relation for your user interface now, it becomes a little complex here. Let's Let's see here. Okay, so first thing we created a project we have this project Udemy Okay, story of our data. It's empty right now. Everything is empty at this point of time because we are done.

Nothing, right? No. All right, so let's go to this section experiment number probably I like touching a dial our let's see our yodas introduction in the last class. Because no coach, I'm not touching the code that requires coding. But we look at it actually, if I have time, so I mean, I have time but I you know, we see that if if it works, because I have not tried to be honest, I didn't practice that. So that's the reason.

So experiments, let's go with experiments. So that's what we are working on. That's what there you will be creating an algorithm. Let's see here, what we can do here. Let's go ahead and click on you. We're going to do first thing okay.

First, we'll just try to understand this tutorial or not that is really available, but from what we have learned from this tutorial. So here this It shows an experiment in phase two sharp steps. So here it says let's build a machine learning experiment to predict income level based on demographic information. Okay, that's great. So we will predict an income level based on democrat demographic information. So let's get started.

So seeing that new Okay, it's creating by itself. So, here if you see this is a flow chart. Now, if you see this, it has done automatically for you. I have not done anything. So yeah, I wanted to explain this data sets actually, but since we have that, okay, so step one, add data set from example, the sample contains census and income. So what this will do it for you, so I'm not doing anything just let let let system do it for you.

So this is Watch you know how we can do we visualize it and see this this is good. I mean, this should be good for you actually for for understanding. So the so in this part, what we have learned is what we can. So it's all about data, it's all about data. So we are predicting here in a see this income prediction, we are doing an income prediction here. And this is the data which is not it's kind of database but we are not looking to the database technically we are just using this as a UI find w so now assuming that we are just you know, drag and drop a data for income prediction.

So what we are doing is we are predicting income. Okay, so this income is there. So what let's split randomly, the data set into training for 70% test set from 30%. Now we have pretty tame for the 70% of data, how much 70% look like and the test set for other 30%. Okay, let's, let's, let's show it. How does it do?

So we're using split data, okay? It's 70% it's good. Total using a random seed is 12345 Okay, that's good. Now it's showing in select and machine learning or our job. So here is where this algorithm starts from. You have the data, right?

You have the data from adult census income binary, okay, you have the data, what we are doing is income prediction. Now you are working on data, you split the data into two parts one is 70%. The other two is 30%. That's good. Now you're to use that machine learning algorithm let's let's see this. How does it looks like the two the two class boosted decision tree algorithm and train it with the training data sets.

Let's find out that this is okay. So I have not used this one. So this is using two class boosted ation with this split data and using the train model. Okay. You have to specify this column which column Do you want income? So I yeah, that's what I thought.

So, because this is income prediction is pretty simple right? This was simple very simple thing actually. So you are actually predicting income with this is the data which you have visualized already. So, of course, I understand you don't get it in the first attempt. So, that is the reason we will attempt it more than once. Make prediction.

So let's use a we're using two class boosted decision. What does that mean? That's a really good question. So here, there is that's a model That's, that's, you know that's gonna come in redial actually you can you manipulate it as per your requirement, but that same build algorithm. So predict let's use trained model to predict on the test data set. So we have this test data set, right let's use core model of course, I'll be using that too.

And then evaluate this evaluate to show view the graph actually. Right. That's good. Now we got to run this experiment. This was this was very simple. This is you can run it, but only so this is a tutorial so you know, we will just I don't want it to skip this one.

Because I wanted to show you how does it work? And, okay, successfully ran this and we visualize it. Let's see what happens. So this spring, taking With scored levels and probabilities, and you want to see this evaluation model do so let's visualize that to see what it has. So this is how this graph shows up. And it shows this really good numbers.

But to be honest, no, you don't see much of what what did we do? To be honest, we, you know, we just predicted this for machine learning. We didn't do this anything for this to this is for machine learning to predict the data that means 70% of data use, I'm sorry, 70 or 75 75% of data we know try to predict unless 30% is to train the data, right? train it to into see how does it look like Okay, so that's pretty much that's it. This is very simple one and it was very easy. And everything was done by this system.

We've done nothing. Okay, let's No Buddhist web services, we deploy this, you know, we create this. This automatically does this web services for you. And this predicts and that's it. This prediction experiment was created and showed that the free modeler is created for you. Web Service morning kit, the data flow application program interface you about the Genet, okay, can you show me how?

So that's web service import, and, in fact, the web service input schema and put the schema exploding income label. Okay, great. So we do we do that. So here, actually, they're trying to do this a little extra step. Probably. This is really a bit complex, but the whoosh we'll see here when we go through this one, actually So here you're using income from of course, because you're predicting data just from the income.

So you are create successfully created predictive experiments in the web services. So now you'll help the system to infer web service output schema by not only including the score table and of core probabilities. After that, we'll run this predictive swim. It's okay, that's great. Show me how this mini map is actually making it. Ugly, this code levels this actually is supposed to have this column called snow column for those good reasons, no columns, but once you run that, now this goes to run.

It takes some time. Not a lot, but take some time and once it runs You see how it's finished running in now this predictive experiment this income what we predicted is ready to deploy as a web service. That's great. Show me how Oh, this is simple click just click on deploy web services and it gets so Oh yeah, so this is what you get it you know, let's test it. Okay. Good.

Okay. Testing predictive. Now we congratulations you didn't know I didn't do anything wrong, everything was done by the system. So. So we have created API is application programming interface to predict income levels based on demographic information. You can hook it up to your mobile desktop now.

Yeah. So you can use this, this key is generated with this key you can close this. So here, you know, you can download this one, okay? And this is you can use as a third party for no for this one and you can configure you can configure here in the amount of information here so we don't want to mess up at this point. So now what you have this is know if you go here this DMT because we do experiments if you see this there's two in Canada two I don't know why got add to. One is the for web services, the other is for experiment.

So that's the reason it got saved into 4k. So that was just an example for you to show you how in but in the next class will be enough In real time example, and I will be running that. And we will be testing that. Because we want to model a bit mathematical operations to actually I wanted to show you those options. Actually, this showed nothing for you actually, this was very simple. And you can go to the experimental view this, close this mini map, minimize it.

Okay, let's look into this pretty quick. This is not a code. This is a flow chart, right? And this was pretty easy. Now let's look into it out here, you have this option. So let me show you.

So now this allows me a lot of options to explain you a lot of things here. Okay, I'm not going to expand on optics. Okay. Here, let's see what what we have done. It was just more of a win rate than know what happened. So what we actually did was there was a data here, let's use that.

Data. So there are the data of, of adult census income binary classification where you have this age and you know, there's a lot of data here. Okay, there's a lot of information here. This table should be there in the database. This is a similar thing what you did. So we've predicted the income actually.

Okay, so this income is the last one right? we predicted the how the income will be. So what we did is first thing is we split the data. Let's click on this split the data into 70% that means this will be 70% this will be 30% I'm assuming and then this random seed is kind of randomly they have added more information on this one is this. So here Oh, we have this Oh, good luck. Let's see this output luck.

Wow. This has nothing so Suppose you show up here, but it cannot match. So here you have this a book. Let's visualize it. And so this has a 70% of the data. And this has probably this will be I'm saying the both of those 15 columns, and that's fine.

So we have speed. And then today you have to split the data right now to split the data and train the data. Train that model. Okay? model into income. So here, here's a senior income, you can train I think you can train one column at a time, one column at a time.

I tried for multiple columns didn't work, but probably you might try it for the probably want to use another flowchart and it becomes complex so so here what kind of algorithm we use This is not an ad. So this is this is the maze. So here is a binary classifier using a boosted decision tree algorithm. Okay, this is a, this is an algorithm and using this algorithm, you're training a model, you're training a model, and you're actually getting, you know, training to that model and feeding this 70% of data and developing a scorecard. Let's visualize this data. Here.

What do you actually see yet shows what happened when you train that model? Using that? I'm sorry, when was that two class boosted decision tree to class boosted decision tree? What do you have done is your train date through the income, right, just put in income and probability God generated a new column got generated It says, and the probability got generated. So the income wages income and scored labels based on this code labels, the probability got generated How does it do? So there's a actually there's there's no data here because there is some data loss.

But we will take care of this in the next class actually. So what it did was it actually built in that algorithm that I got what it let's find out let's see what it does. So it you know, you can specify this how many trees you need, so it's kind of tree right you know, tree and how many instances you want based on what the learning rate. So how do you want the machine to learn right spine to pi 3.4 point five. The short goes 100% thousand percent. Or how long it goes.

Because this is still in the new stage, so probably might go high, it's probably not so good in like getting good data and good numbers, I mean, okay, and the number of trees to constructed is hundred. We didn't put any seats here so so this is the A Guide to boosted decision trees and I got them where you train the model using this constraints parameter Actually, this these, these are the parameters in the two two class boosted decision tree and you turn the monitor and you've generated a graph let's let's see here, too, this was actually and based on that, we generated this graph on the false positive rate to the x axis and true pass rate on the y axis. Here we generate an income prediction. So what we understand by this Of course I mean to be honest, yeah, it's not easy to what we're trying to do here.

That's a good thing. I'm sorry. Actually, that's not a good thing. Um, we have three options actually our OC and our MCs Okay, yeah oppression, because how the oppression the system is and when you do this, so recall and oppression on the y axis and the lift, okay? And this is how it gets. And he okay here are the numbers.

So this is the threshold and we use the accuracy and used an accuracy of point 865 and the pressure of point seven, five. So what it did was it really predicted probably, I would imagine it predicted the future income based on that data. Assume at least that it has predicted the future income based on what data you have a lot of prediction. Right? So we have we have been breached. So most probably has predicted the future income.

Base based on that model based on the Colorado two booster and remember the two based on this you can use there's a lot of algorithms you can go to in a moment. But when this data was generated, and this data what happens is this data is not useful for you, but it's for machine for the machine because it's machine learning, right? Are for the machine to predict the future data. Okay, so this is what the machine actually not for you. So you don't do it in this data. But this data actually, this.

This is the real data We know this data. All right? Let's see here. We know it should work class, education, education numbers, module status, occupation, we know this right? So this is a normal, our personal information that is stored, but there is done that personal information, we are predicted the future income, right, based on this based on this arrival. Okay.

So that's pretty much and we deployed it into predicting experiment. Let's see. That's how we did it in the web services. And we add that tutorial, I actually added a new data set and we know that web services output showed up so this is more for it's good for machine to her this, you know, put the feed the data into this machine. And that machine uses this model and based on that model, predicts the data that's about but the whole point of this Microsoft Azure Machine Learning Studio is to come up with new algorithms to predict a new a new types and new new columns you know you have this income not just income this prize you know if you are in our automobiles our will do this for automobiles actually this was for this we go to the Emirates and we do that manually actually this was done during this was kind of doing it so, we were back to this training experiment and see here So, let's go through here machine learning right.

So you hear you have a lot of options will this can be tracked this year. Good. So this is a Gotham's, you know multiple addition forest multiplication jungle Adams wanted me okay. So, I'm sorry this is this models we have this anomaly detection, classification clustering and regression. So, we had about this layer regression we go to this linear regression okay. And this anomaly is this for a different kind of you know things and classification I think this is just classified right to based to class to class boosted yeah it uses classification method, how do you classify the income kind of thing okay.

And clustering, you can use clustering there is only one k means clustering and then you have regulation. So, these are what are these, these are the various types and methods of in algorithms. Okay. We'll do one more linear regression, and we'll see how this thing works. Okay. neural net integration, let's say you have neural network in one frame that one, you can use that do.

So we work on regulation thing, and score. That's not machine learning, actually. It's part of machine learning. So, here the score model is to see how this leads basically to what we did in the school money, let's visualize score model is to predict the data. What happened when we did this probability generated this probability labels, you know, scored labels and scored probabilities generated, that's good. So that's what this score is all about.

You can use these options here, various options, cream, he of course, you know, take this to the supervisor supervised learning and unsupervised learning. So I'm not sure this is They're they're here, but we'll use most of time use train model, or you can do your own experiments during the of the interview here a lot more than what I actually did. So you can do train anomaly detection model. I'll work on train models. So this will be working on train model, which already key created, you know, which is already created, which we already have the data, we are training the data, but I'm sorry, we are training it based on that data 70% point seven right. So that's about training.

Now we have Iran which was never shut up the reality evaluate is to get the graphs okay. So, we Oh, you iterate, you can cross evaluate, you can evaluate. So these are the various options for your machine learning. Then you Want to call the if you want to call? Do you have this Python language, our language, statistical functions? Oh, these are very important.

This is what I wanted to talk about. Let's talk about this for now. Okay? Although it's too early to talk about this, but I want to know who even finished this action. So here, what you can do is you can apply mathematical operation, let's say, your score model. Let's let's visualize this.

Let's visualize this. And once you have predicted this, it has generated some values, right point 038015. You know what these numbers are, I mean, not big numbers smaller massively, let's say three digits 0.038. So you want to round off. So in that case, you want to use this mathematical, I'm sorry, math operation, not just math operation, you can use debug multiplication, subtraction, addition, all those things. So that's month statistical functions, you want to summarize it, you can use discrete I mean, use a lot more activities, okay?

That is statistical functions. Then you have text analysis, if you have any model, which does the text analysis, if you are, this is much more complex actually, you use I'm not sure if you have this in a sample, but you use or if you have this, but it becomes more complex will use more complex than this one. We'll see how we can use in another class the next class we'll see that but we'll have probably will not be using this text analysis because text analysis if you have data, which is in the text and you want to analyze that based on this model, so you can also do that. And then you have time series time series with anomaly restriction Web Services input and output we have just done as web services and deca create your best type of clusters and analysis. So these are the options for machine learning.

And oh, and actually are you have lots of options here. Yeah, so one of the streamers it should have okay and you have various samples. I'm using one of them samples for you to understand I Oh formations. data format conversions. You can convert the data convert to CSV convert two data sets you know, if you have data set you want to convert to CSV means that we're not using that in this class but just good enough we can use this export data transformation. Oh yeah, this is really important.

So if your ETL work, so to be, this is what we'll be using it actually mode, you can filter a data, you know, you can, let's say you want some data. You know, let's visualize this again. Let's say you want to use filter data, let's say you want income based on let's say, not sure like education, let's say your bachelor's and master data, so you can filter it. So only the data which is in masters and doctorate and Bachelor for short, and the rest of data, you can remove it so that you can use the filter. So you have a fire filter, IR filter, okay, median filter, and average, through these are the various filters you can apply for this. I've just given an example.

You don't have to be that way. Or this lot more than 10. You can do this You can do transformations to not. So these are the transformations of data how you can transform the data in machine learning. Right? Then you can count, you can modify the table parameters, or you have one of the sample sizes, you can also edit the metadata to.

So I wanted to show you that one that was very important. But unfortunately, that was taking too long. That's why I want to skip that one. Because I leave it as it is the homework. You have this data, and I show you actually, but that won't run because it takes too much time. probably want to use this and forget about it.

After probably one or two I want to see that it's ran successfully or not. This takes too much time. And then your split data. That's what we did split data Scale energy as you can scale it, okay, so these are the radius options you can perform. So you have this, a machine learning human does not just do machine learning, but also transformation of the data is also important. So in that case you use that too.

So, I hope that gives some information about this way. So, I wanted to explain this one before so that you have some time but we did this interior. Okay. So, we'll stop here at this you know, and we continue the next class, this will for income prediction, will do for automobile price prediction, that will not be This tutorial will do it. Probably had been looking at this tutorial and showing it how and we will do it manually. Okay, so we'll do it manually.

And we'll actually go to this section and you know, created You experiment here you have this two experiments, right? We create a new experiment. And from here using blank, and we will create an auto buy price prediction and see how that works. So I'll be using more more fluid flow chart robot is there here, but that will give you some information about how to predict. We have predicted the income right it's in you've seen the graph will predict price and also see the graph. And we'll see how that thing works out.

I hope I didn't make mistake because that's a little more complex than this. This was very simple because it was in baked by Microsoft Azure Machine studio and it showed for you but that is manually I will be doing it. So so I hope We don't make any mistakes. Okay, so that's pretty much no for this tutorial. This was for income prediction. This tutorial It was a tutorial from automated tutorial.

That was machine learning machine learning. So machine learning already told you how to learn it. So, okay, so that you know that's pretty much and thanks for watching. I hope you liked this class and I'll see you in the next class. Thanks for watching