The goal of targeted marketing to identify the people who are most likely to be buyers. In this game, as marketing consultant M, you will try to find the best 20 prospects out of 100 about whom you have some demographic attributes. You will be given a training set of 500 previous people, their demographic attributes and whether they bought or not. The demographic attributes are things such as age, wealth, sex etc though they will come in as just A1 through Ak. The value of k is under 50, but will be determined at game time.
You may find it useful to consider a Naive Bayesian classifier because you want a binary value (will buy or won't) based on real-valued attributes.
You will also play the role of spoiler S in that you will determine the weight (positive or negative, in the range -1 to 1) for each attribute (e.g. wealth might have a positive weight but age might be negative). You will provide these to the architect.
Receive weights from S. You will generate, for each attribute for each of the 600 prospects, a value from a uniform distribution between 0 and 1. Then you will perform the dot product of the spoiler weights with each prospect's values.
Next the architect will order these dot products. Rank these in ascending order, so least dot product has a rank of 1 and the greatest has a rank of 600. Divide each rank by 600 to get a normalized rank. Assign each prospect a value of boughtsomething with probability 0.3 * its normalized rank. and boughtnothing otherwise. Permute the order of the prospects. Make all of their values for each attribute visible to M. For the first 500, also give the prospects' bought status.
M will then give you 20 predictions for the best prospects of the unlabeled 100. You will give M a score equal to the number of those 20 who have a hidden value of boughtsomething.