INSURANCE: This is a discrete Bayesian network. There is no particular target variable. There are 27 observable variables (observable=[1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27]) and 27 actionable variables (actionable=[1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27]). There are no unobservable or hidden variables. The goal is to predict non-manipulated values in test data. For each line of the test data, a single variable is not manipulated (moving target) -- value NaN=missing. Return that value in the PREDICT file. No training and test data are initially provided: you need to request data and pay for them in virtual cash. You have an **initial budget of 756000 experimental cash units (ECU)**. This allows you to purchase all the default training set, which costs 126000 ECU, and the test set, which costs 630000 ECU. The goal is to optimize your spending to do better than you can do by just purchasing the default training set by planning your experiments. You need to purchase the test set to make your final predictions, but you can save by requesting only a subset of the variables. But, be careful: keep enough cash to purchase the test set, otherwise you will have to make predictions on unknown samples. There are 2000 training examples (in the default training set), and 10000 example in each test set. For training data, you may request a subset of variables, e.g. exclude the target, which is expensive. For test data, you cannot choose the manipulations: we provide you with a query including the manipulation to be performed on the test set. Hence, the test data will consist of pre-manipulation observations and manipulations. You will be charged 9 ECU for samples, 2 ECU per variable observation (in addition to the cost per sample) -- you pay both to observe variables before and after manipulation --, and4 ECU per variable manipulation (in addition to the cost per sample). Good luck!