Problems with Brazilian Coffee

[mhaut]

Dear authors.
I am trying to replicate your results with Brazilian Coffee dataset, shown in table 2.
When you talk about accuracy, are you considering the training and test or only the test?
If you are only considering the test data, how do you do it? Because I am trying to reach the same results with this code, without changes, but I can't. However, with UC Merced dataset is no problem.

Thank you for your help.