Quantifying Outlierness of Funds from their Categories using Supervised Similarity