Python Tutorial : Exploratory analysis of KPIs

Great work on the exercises. Let's pick up where we left off. We want to determine which conversion rate metric is the most appropriate. Note that most companies will have many KPIs each serving a different purpose, and that here we are only working through one of these cases. To calculate our potential KPIs and to see its performance among various groups, we will use the groupby and Ag Pandas methods. This lesson will focus on these methods and the next lesson will more fully explore applying them in practice. We can call the groupby method on a dataframe to specify groups to aggregate over. Here we will use it on our combined demographics and purchase data dataset. The primary argument is by to which we want to provide a list of dataframe fields that we want to group on. Here the potentially relevant fields are country, device, gender and age. Here we will be grouping by country and device. The next relevant argument is axis, which specifies whether we are grouping by row or column values. The default value zero groups by columns, which is what we will do here and for the remainder of the course. The other argument of interest is as index. By default this argument is true, which means that the grouped by fields become indices. We want to set this to false so that this does not happen. This returns a data frame groupedby object. The next step is to aggregate over these groups. The easiest way to do this is to call an aggregation method on the dataframe groupedby object. Let's call mean on the price value of our dataframe. The output is the mean amount paid per subscription across all purchasing users. In this case, rather than being calculated over the entire set of data, it is calculated over each device country combination. Any built in function similar to mean can be called on a dataframe grouped by object. However, more flexible options exist through use of the AG method. The easiest way to use this method is to pass a function like mean to it. As we can see, this has the same impact as when we called mean directly. It can be further expanded by passing in a list of functions like mean and median and calculating both. Its true flexibility comes from a third type of argument. We can pass AG a dictionary where the keys are column names within our data set like purchase or age and the values are a list of functions to be applied over those columns still broken out by groups. Let's find the mean, minimum and maximum value of each of our purchase and age as an example. Another great flexibility of the AG method is that we can also pass our own functions into aggregateover not only built in ones. Here is a function that finds the truncated mean value, I.e. it removes the top and bottom 10% before averaging. We can aggregate our age over the country and device groupings with this function. The only distinction is that when inputting this function, we do not want its name in quotations as we did for the built in functions. It is important to cover the techniques before proceeding to applying them. In the next video, we will look at how to use these to examine KPIs across cohorts and discuss why this is valuable. Let's practice these tools before.