What type of meta feature is extracted from a model using the training data, such as properties of decision tree models?