Exercise on learning preferences over consumption bundles