NOVA: This is an active learning dataset. The goal is the predict the values of a particular target variable (labels). There are 16970 observable variables and NO actionable variable. There are no (known) unobservable or hidden variables. There might be missing values (coded as NaN) or infinite values (coded as -Inf or Inf). Data initially come unlabeled. Unlabeled data are "free". You may query for labels by providing a list of samples in a SAMPLE file. Note: only the first 9733 examples may be queried for labels, the remaining 9733 examples are used for test purpose only. To get labels, you must also provide predicted values for all the samples in a PREDICT file that you package with your SAMPLE file in a zip archive. You will get in return a LABEL file, with the labels of the samples you requested. You have an **initial budget of 9733 experimental cash units (ECU)**. This allows you to purchase all the training labels. You will be charged 1 ECU per label. The goal is to optimize your queries to climb the learning curve as fast as possible. Good luck!