Diversifying Design of Nucleic Acid Aptamers Using Unsupervised Machine Learning