The discretized intervals should not hide patterns. We must carefully choose our
intervals or we may miss potential discoveries. For example, if the intervals are
too big we may miss important discoveries that occur at a smaller resolution,
but if the intervals are too small we may not have enough data to infer
patterns. We refer to this as the resolution problem. Additionally, if the intervals
are determined by examining features in isolation then with discretization we
may destroy interactions that occur between several features.