Sorry if this is obvious but I haven't been able to find it anywhere else. I'm using xgboost's importance() function to generate a matrix of information about the variables in a multinomial classification model. I know the "Gain" column indicates roughly how much each variable contributed to the model's predictions. But I don't have a sense for which direction that contribution was in.
For example, in glmnet's version of this function, there are positive and negative values for gain, so for a binary classification you can see what correlated strongly with each outcome, both 1 and 0. I'm sorting my input into five groups. Should I assume that a higher gain value is always correlated with a higher group number? What is correlated with low group numbers?