Moving on to dataset shape and skew. So two quick slides here on dataset shape and skew. So your dataset, it can come in a variety of different forms. When it's given to you, when you're accessing it, you'd have a ton of columns but not that many rows. We call that a wide but short dataset, which is probably honestly the worst case scenario. It means you have a ton of things that you want to collect data for but you just don't have that many instances or occurrences or records. Maybe not as bad but still pretty terrible as in the lower left now. We have a dataset that's taller than wide. I mean, you have a lot of observations, but say you only have two columns or three columns and you want to learn a ton more. And you just can't do that much great analysis when you can only select three columns from your dataset. And again, it depends on your data-set or maybe in the first one, the upper left maybe serve a small dataset. Your average number of columns and an average number of rows and that's all you've got. Our perfect scenario, is you have whatever the right amount of columns is for you and enough records to make judgments and inferences from your data and insights. And again, there's no magic number. I'm not going to say like a thousand columns wide in two million rows long or whatever the perfect ratio is between both of those. But just keep that point in mind. This is especially true when you start to get into machine learning concepts, when you're trying to make different training and test datasets. You can make sure that they're actually have enough observations to make a justifiable conclusion about your records. The same is true for data analysis. Speaking of data analysis, one of the amazing things you can do when you're exploring your data is look at, all right for this one given column take first names. What's the frequency of occurrence? How many people are named James or something like that in my population or in my organization? I'm just curious to know immediately. All right. Well, there is a lot of James' in the organization or the most common name for this U.S. charity that is present across multiple of these charities is blank. And we'll go through that example in the cloud data prep demo in just a few. And the skew of the data. So the distribution if it's not uniform, it could be really, really skewed in one direction. So, I would say for example, if we just looked at the nonprofit data that we pulled in and we saw extremely heavy skewed towards charities in California. This is something that immediately can jump out at us, that could say a couple of things. One, it can question the validity of the data. Did we collect everything or was there just California charities that were collected as well? So skew, is something that you want to watch out for from an exploration standpoint. It also has performance implications. If your dataset is heavily skewed, it could affect your partitioning of your data, which we'll talk a little bit more about in the performance section and the third part of this course with advanced insights.
