Fig. 2
From: Dynamic order Markov model for categorical sequence clustering
To compare with the PTS above, we use the same dataset (250 sequences over \(\Sigma = \lbrace { a, b, c \rbrace }\) L = 3 and \(\tau\) = 250) to demonstrate the pattern detection process of SPD: a Initial PST constructed from dataset without pruning. b Find merge candidates with threshold \(\tau\), the subtrees share the same color will be merged. c Final SPD, the node below red dotted line will be pruned. d Final PST with threshold \(\tau =250\) for comparison with SPD. There is an extra sparse pattern \(\mathbf {a*b}\) in (c) compared with (d)