I'm moving on to point number four. Need to have enough examples of your future value inside of your data set. And a good rule of thumb that we use and again is just a rule of thumb is that I like to have at least five examples of any particular value before I use it in my model. That's five examples of the value inside your training data set even before it touches training. So, what do you mean by that? Well, let's take an example. If you have a purchasing category equal to automobile then you have to have enough transactions of fraud or no fraud auto purchases so you can take a look at fraudulent auto transactions for your model. If you only have three auto transactions in your data set, and all three of them are not fraud, and essentially the model is going to learn that nobody can ever commit fraud on auto transactions. It makes sense, right? Because you have no auto transactions that are marked as fraud, and that's ultimately going to be a huge problem. So, to avoid having issues like this where you don't have enough examples, notice that I'm not saying that you have to have at least five categories, I'm saying that you have to have at least five samples. So, for every value of a particular column you need to have those five examples. So, back to our cracked driveway example from our housing model earlier. If you believe the photo showing a cracked driveway should be a good indicator of a housing price, be sure you have enough samples of cracked driveway photos for your model to learn and train off of.