When you talk about the production of the site So there might be separate civil which is holding with the data We had a member of house management There will be a separate data we are talking about Deliveries will be a separate several Julie holding the lead We’re talking about customer satisfaction Lord of Data has come from different sites So this is how we should planning store particularly that into one day Yep Post that what we do is okay Put this into a major cluster which we call it as data cleaning or data preparation Basically so here supposed to process first of all validate the reader.
Before brothers this inerrancy valid See not all data is supposed to go and sit on a machine learning models Not possible So initially what we do is we do radar allegations like we check if the data is scale and scale If the data has normality nor normality is are the different columns for most fantastic in the sense if you talk about linear regression Okay so yes I’m gonna represent the line How many of the points are not going to come on the line So if I put a line and 99% of the points are outside the line it makes no sense for me to implement this kind off models So that is what So I will pick up topics one by one in each your procession Such ability covered everything Post validations.
Once we have confirmed the locator looks kind of interesting and could be used into one of the machinery in courts Ah boast that Okay um hind fish Can you hear me, Yes Well it is What I’m doing is I’m just giving an overview of what happens under industry side off data science Okay so there’s nothing really the topic today so I’ll spend around interesting meets every session on covering of whatever you miss so far Something in my sight, Okay And for today’s session I have you know complete maybes, Ah the theory and practical bold part of it Then we haven’t touched upon Jokinen will complete it because it’s a very simple topic and pose that I will also try to complete Eskimos Okay so that you guys were prepared for the videos And even I know extreme is going to release next week.
It’s good to do it together so that you come to know what each other guard amiss Andi when the reduced will go very easy in extremis leading a little troublesome That’s what I feel Sometimes he doesn’t get the concepts if it if you can redo quickly Astrium half an hour If it gives people needed else we’ll cover the different topics Okay off So post valuation What we do is we do data cleaning Okay So industry-leading part of a different case you have missing values everything case you feel some of the values that wrong because not always will get a perfect valued nor data free So in this case what do we do is we do all these techniques to make the data suitable due to being used for the implementation like the most common problem in the areas of the number of zeros or missing lands.
For example, if we’re talking about a deal asset and let’s say you have got a majority of the columns as eels what Begin toe either begin to remove them Okay All we can impute some other value here Now the challenge is if you removed we might lose X percentage of getting out off originally because we’re to keep this in mind Secondly if we try to imputed value on that we are imputing experts and age off unknown valued will It might seem to know what it does it itself We’re not sure what we’re including you And if we have multiple techniques it will never be an accurate value Definitely Okay so this is a very important topic and I’m gonna touch up Ah in a very detailed way because not many many people you know have a good exposure under data feeding part of especially missing value replacement and all the stuff Course that you have to organize your readers.
Let’s say if you have some rows and columns to be up and down this card, So I called it did organize So this is a step where you are going to prepare your data for machine learning Okay post this thing What you do is we put the push today trying to machine gun on me you know to get Puts out of the machine learning stuff After this we do a very important step So usually in those offer in the city collar testing Okay well I will not call it purely testing but I will say I will take this into away there I will try to do some testing on this machine learning model that we have done So, for example, we’ll do performance testing stress testing a lot of testings we do So we’ll take this model and put it in the very stressed and moment Where will come to know what is the minimum accuracy that the model can do before it is invalid Or what is the maximum accuracy the model can do before it breaks something.
We call business feature engineer wherein we’ll pick up one particular lemon which is the best over here and we try to you know improvise accuracy So we tried to squeeze the maximum accuracy possible the, for example, the first run We would have got an accuracy of since 78 person for example, Okay Now for future engineering what we’ll do is we’ll try to change the variables in and out such a real lack At least we would be able to push this toe towards reports and more So let us say we put this to 81 Plus this is the goal of feature engineering and also in the steps we try to understand if in case just in case there are some features from the data which are not good for us we try to remove them and try to get a better back to this.
This is like the last-minute brush upon on accuracy good and post act We do a visualization reporting production delivery So are a lot of stuff over there which happens after I’m gonna let us say we do this type of stuff at the end of our modern so that I will show you how to package You’re fighting Accord How to adapt it on when I know where Blake If I guess it is a Web interface how do you do A lot of things we can do over here So far for my other batches wherever I’ve taken So far as I have covered all these topics one by one except the last one which I’m supposed to show you in the last session So what I’ll do now is trying toe every week will try to cover some topics and then three minutes.
I’ll try to give you an overview of what happened so that at least you know the complete picture What was that Um because no hero this DP world here Shona there’s only one part it would happen on a school date Oh no it has multiple interfaces again So SQL stuff ends over here So once we collect the data and put it into a finder’s date, Africa, in data gathering step Okay Yeah sort of Where you are getting a daytime They’re putting into a new Uh-huh Yeah And this is the second part the DP World which I sure am nothing but he a pine box Data from its were playing around with primal zero That’s it.
We put on the other Things are fight only living Is Crichton from here Ok but where is Phil Comes It still comes here Why Before taking from a survivor or once we take it from server See SQL is basically equity which we which fight and will give it to the servant for fetching the later that So it comes as a part of it, Okay It’s been getting the data from the server It s good that Yes Okay Okay But not all businesses Very interesting If you have a static data.