Hello and welcome back. In this video we will perform cross validation on our model. So, first of all we need to import a wrapper that will allow us to use Keras models in scikit learn. We need to do this because the cross validation function is actually in scikit_learn. So lets import the KerasClassifier wrapper. And then let's define a helper function that is called build-logistic-regression-model and this just builds the model like we did before. So defines the sequential model, add the dense layer with one input, one input, one output and the sigmoid activation function and then compiles the model using a certain laws the binary_crossentropy and also requesting the accuracy. And this is the optimizer we'll se later what it is. Then we return the compile model from our function. We've defined this function because the KerasClassifier wrapper needs a build function to work, so we actually need it to do this helper function for KerasClassifier, notice that in KerasClassifier we also define the number of epoches and whether or not it needs to be verbosed. So basically now this model which is KerasClassifier wrapper will behave not just like a Keras model but like a Scikit_learn model. Which means it exposes all the methods required by the cross_val_score function from Scikit_learn to perform. So, let's do cross validation, we import the cross_val_score and we also import this class which is Kfold cross validation which by now you are familiar with. 

And we define a three fold cross validation, so let's check the signature, this is the number of splits so fold is three and just explicitated so we are splitting our data in three equal sides sub-sets and resetting shuffle equals true which means we will take random shuffling of the data before we do our splits. Okay, so these are cross validations. And then we pass the model which is our KerasClassifier the x and the y, the features and the labels and the cross validation iterator to the cross_val_score function. Okay, so a cross_val_score function needs an estimator which we've created here, features, labels and bunch of other things that we're not setting and those that has those parameters we can set other things. but most importantly it has a cv module, cv argument sorry. Okay, so we execute this line, it will take a little bit of time, okay great. And we have obtained three scores for our three fold cross validation, notice how they're similar, which means the shuffling did its job. Helper didn't end up with really biased samples. And now we can print the mean and the standard deviation of the scores, so the cross validation accuracy is 82 percent plus or minus 0.25 percent. This is actually good accuracy. the data set is small, but still. So, pretty satisfied with this result. And see you in the next video