In doing that I ran into one more avoidable but strange issue in using xgboost: when run for a small number of rounds it at first appears that xgboost doesn’t get the unconditional average or grand average right (let alone the conditional averages Nina was working with)!
Let’s take a look at that by running a trivial example in R.
In our previous article , we showed that generalized linear models are unbiased, or calibrated: they preserve the conditional expectations and rollups of the training data. A calibrated model is important in many applications, particularly when financial data is involved.
However, when making predictions on individuals, a biased model may be preferable; biased models may be more accurate, or make predictions with lower relative error than an unbiased model. For example, tree-based ensemble models tend to be highly accurate, and are often the modeling approach of choice for many machine learning applications. In this note, we will show that tree-based models are biased, or uncalibrated. This means they may not always represent the best bias/variance trade-off.
In our previous post in this series, we introduced sessionization, or converting log data into a form that’s suitable for analysis. We looked at basic considerations, like dealing with time, choosing an appropriate dataset for training models, and choosing appropriate (and achievable) business goals. In that previous example, we sessionized the data by considering all possible aggregations (window widths) of the data as features. Such naive sessionization can quickly lead to very wide data sets, with potentially more features than you have datums (and collinear features, as well). In this post, we will use the same example, but try to select our features more intelligently.
Recall that you have a mobile app with both free (A) and paid (B) actions; if a customer’s tasks involve too many paid actions, they will abandon the app. Your goal is to detect when a customer is in a state when they are likely to abandon, and offer them (perhaps through an in-app ad) a more economical alternative, for example a “Pro User” subscription that allows them to do what they are currently doing at a lower rate. You don’t want to be too aggressive about showing customers this ad, because showing it to someone who doesn’t need the subscription service is likely to antagonize them (and convince them to stop using your app).
You want to build a model that predicts whether a customer will abandon the app (“exit”) within seven days. Your training set is a set of 648 customers who were present on a specific reference day (“day 0”); their activity on day 0 and the ten days previous to that (days 1 through 10), and how many days until each customer exited (Inf for customers who never exit), counting from day 0. For each day, you constructed all possible windows within those ten days, and counted the relative rates of A events and B events in each window. This gives you 132 features per row. You also have a hold-out set of 660 customers, with the same structure. You can download the wide data set used for these examples as an .rData file here. The explanation of the variable names is in the previous post in this series.
In the previous installment, we built a regularized (ridge) logistic regression model over all 132 features. This model didn’t perform too badly, but in general there is more danger of overfitting when working with very wide data sets; in addition, it is quite expensive to analyze a large number of variables with standard implementations of logistic regression. In this installment, we will look for potentially more robust and less expensive ways of analyzing this data.