Okay proceed we have completed I did understand him there we completed our preparation. So now we will look into modeling evaluation I will say maybe deployment also okay. So the SDK allows you to do the modeling quite easily okay. So these are modeling and evaluation. So this treatise I will be using for these are modeling evaluation and maybe deployment okay. So acgme changed CSE CSV.

So this data is try some of the From a CSV, CSV and then give some label Reynoso and then this is a test data Yvonne a labor, this one with labor the one with labor we will be using it for the evaluation Okay. Why do we need the labor for the test data when we train when we model or classifier using the training data. So we want to advise how a grazed maga hiker is a pacifier. So the classic area put in the CME test with a v cross v over here So, Daddy pacify Armando cross see whether sigma metric the given crossfitter given class label here given criteria given class data okay so it does not match, then we won't count correctly classify if the match is correctly classified. So, based on this we can calculate out the error error of the classifier models okay.

So, to evaluate the model, we just need a row, one common for the neural network okay so is now They're complicated. So, we go back to these the SDK screenwriter kvk remove everything okay now we come to the end each neuron network okay. So the train our b D drive a CME a CSV dandy as far as I say we need one with the labels. Now if the labels with a cross so it should be a CME underscore estore e d drive a see me underscore as to CSV the y variable here the y variable here is the ah colon number or the column number 40 okay there we go. So if you see a Y variable here and no s variable that means up we indicate the y variable, the rest of the variables will be the input variables. So, Y variable is the target variable.

So, after I put in the variable number or the column number, the rest of the columns will be the input okay So I need to see the data the DRI a see me train asymmetry Okay, this is a first column design is the second column This is the column. So variable zero variable one variable two okay so the grass variable will be the target variable, so there will be over two okay so this one is wider base we'll go to number of perceptrons This one is a neural network, so is a multi layer perceptron order by in this one you only have the one hidden layer. So, this number of perceptrons is the number of perceptrons inside the hidden layer. So, either poor or 10 or 20 learning rate usually or be 4.01 day durations is the number of times you want to train these neural network.

So, I will putting 10 or maybe 100 rows, okay. Okay. Oh, okay. That will run right. Okay, so this is the evaluation. As you can see, this is the site All squared error is a row one.

So sum of squared error is taking the predicted outcome minus away the minus the way the origin actual outcome and then based on Debbie's square square the result and then some are everything together. So the sum of square where the smaller the number the better the more accurate model ease Okay, so from here we have done the model evaluation. Okay, the model and the model evaluation case okay. Okay from here he just created the MA MLP or neural network model and then evaluate the error is around one okay. So, if you are not happy with accuracy, then you can go back to well to the business understanding stage and try to understand your data more and then prepare the tomorrow and then continue to train a new neural network monitor and do the evaluation until you Gaddy accuracy or the smaller or the More of sum of square you want to achieve okay?

So this is Robbie the model evaluation then this one will be more on the deployment okay. N de pie man you can guess who in the Mada neuron network then he's fine is the AC T. CSD then that's why you'll be the acgme CME underscore s dot CSV. Number two perceptron Okay, learning rate 0.01 iterations 100. Okay, one way to improve the model in terms of the precision, you can also play around with para adisa settings. So let's say the number of perceptrons and number more than in gray and the iterations, okay. So one way to improve the model is to go back to the pre previous data mining stage.

Okay, and then carry on with all the data understanding data preparation and modeling. Now as you can change the settings for the models, okay okay for a more increase the number of perceptrons increase the learning rate increase the iterations is okay. So now, we are logged in deployment stage. So we create a new neural network model based on the previous settings perceptrons 10 learning rate 0.01 iteration 100 The train is the same power the SPI is different. This one is ah that's why we father the boss So, you can RTS this thus far is a new set of data to predict okay. So we are using this neuron our mod same as the settings in a valuation stage to predict a new asset and then get a result.

So you can say this model is actually the station is actually the deployment okay so Eva VA photo doobie without further ado we will run the script okay so the Generate Addy dry acgme underscore train dot CSV underscore pre data dot CSV okay So let's see the data. Okay, so Rafi de sada prediction from this neuron ever model. Okay, baby there. We finish this chapter cavia finish this chapter. Okay, in this chapter we have used the SDK screwed right okay to go Go through the Chris Yamada. So we have go through the data understanding the script writer.

Data Preparation, we've done screenwriter modeling evaluation and deployment with this screenwriter. Okay, so you can see is that the SDK engine. So Dean so the SDK engine is ah, the SDK engine is more difficult to use is a interpreter. The var any Kyle help to the User So, you can only just type in the command here import CSV by separator equal is co equal Okay, in this D the SDK and G ah we do not need to put in a semicolon, okay because he just would run one command. Okay. Then the SDK script writer provide a better interface to help the user to Riah Ryan's scrape a list of commands to run in the SDK engine.

So via this the SDK screenwriter. We do not have to clear memory I saw the command we just about created, we can easily modify the properties. So these d d SDK screenwriter help us to write the the SDK script more easily. And they asked us the IDS SDK and G more easily. Okay, so the SDK and the SDK script writer is free. Open Source.

Ah, the SDK studio is the next chapter we will be looking into. The SDK studio provides a more user interface, user friendly interface. So the SDK studio do not need a need a user to write any script. He just generate a scrape using the interface So yeah, we are we come to the end of the chapter tree on using the SP DK script writer and the D SDK engine.