stanford maxent classification prediction

Question

I am trying to use stanford maxent classification to classify data using 4 features and the 5th one is class. When I am supplying test data then the accuracy is 99% which seems next to impossible.

When I check the labels predicted, they were exactly same as in the data itself. But then I changed labels in data and the predicted labels got changed. I suppose the prediction using a classifier model should not consider the labels given in data. Please suggest how to use the maxent model on unseen data.

Answer 1

Your classifier should not use the label of the datum as a feature. At training time, pass in a datum with the label set; at test time, pass in a datum featurized in the same way but without the label set.