There are two datasets, Musk1 and Musk2. Both are about predicting whether a molecule has a musky smell or not. A molecule is described by the different shapes it can fold into (conformers), each bag corresponds to a molecule and each instance to one of its conformers. Conformers are responsible for the properties of a molecule, i.e. its smell. If at least one of the conformers can cause a molecule to smell musky, the molecule is positive for the musky class. If none of the conformers have this property, the molecule is negative.

Original source


  title={Solving the multiple instance problem with axis-parallel rectangles},
  author={Dietterich, Thomas G and Lathrop, Richard H and Lozano-P{\'e}rez, Tom{\'a}s},
  journal={Artificial Intelligence},


Files – This file contains two different .MAT files for the Musk1 and Musk2 problems. You need the MIL toolbox to load this version of the dataset correctly.

3 thoughts on “Musk

  1. Gökhan Çoban

    reallifemil function of mil_tools is unable to load the given .mat files.

    It tries to import musk/ and musk/ instead of musk/musk1.mat and musk/musk2.mat.

    If we change the lines, it gives the following error:

    Warning: Error occurred while trying to call loadobj on a dataset object:
    Reference to non-existent field ‘props’.
    > In importdata (line 197)
    In gendatmusk (line 41)
    In reallifemil (line 139)
    In Datasets (line 1)
    Subscript indices must either be real positive integers or logicals.

      1. Veronika Post author

        Sorry for my late reply Gökhan, I didn’t receive any updates from WordPress that somebody had left a comment.

        You are totally right about this! All files provided on this website should be loaded with prload.m.


Leave a Reply

Your email address will not be published. Required fields are marked *