Moving on to credit cards. First up, whether a cardholder has purchased these items at our store before, and again a reminder, this is a credit card transaction whether or not it's fraudulent or not. We don't have enough examples of cardholders who have purchased and cardholders who haven't purchased. Yeah. I mean hopefully if your business is any good, you'll have a ton of transactions in your history. And it doesn't again, it doesn't matter which item, or which store, because we're defining in such a way we'll have enough customers who've purchased it, and enough customers who have not purchased it. But, suppose we got a hyper specific, and we define this as whether or not a cardholder has purchased a bag of diapers between 8:30 PM and 9 PM at a specific store number one, two, three. Now, you can see this is way too specific. So, it really depends on how we define the problem. If you define it general enough such that you have enough examples for the good value, then you're in good shape. Next up, the distance between the cardholder's address and your physical storefront. We have enough examples of customers who say live 10 miles away, yeah sure why not. More 50 miles away, 60 miles away, 100 miles away, may be starting to become a problem, you're dealing with scarcity. So, this is basically where we start grouping things together. You can't use your value as it is. So, say you're going to take all the customers that live more than 50 miles away, and then treat them all together as one group. You're not actually going to take a specific customer who lives 912 miles away, and use that number in your training data set. Because I don't know how neural network will happily know that anytime somebody who comes in from 912 miles away, it will include them in that larger 50 plus group, because that one time that this person came in who live far away, and then used their card, they didn't commit a fraud. So, that's basically what you want to avoid. We're talking about the values of features and not the values of labels. So, how do you make sure that you do this? How do you make sure that you have enough examples of a particular value? Well, the easiest way to do it what a lot of data scientists do, is you actually plot histograms of your input features. And that will give you the frequency of occurrences for each the different values in your data sets. You'll practice that a little bit more in your next lab. Okay. Next up, consider the category of the item being purchased. Absolutely, well, I'm hoping you'll have more than five examples for each category that you're going to choose. Last up, an online purchase, or an in-person purchase. Again, you'll definitely have more examples of these, hopefully more than five, that shouldn't be a problem at all.