dmt4sp

The dmt4sp prototype is a command line tool to extract episodes and episode rules under the minimal occurrence semantics as defined in [1], supporting various constraints, over a single sequence or several sequences of events.

Contact: Christophe Rigotti

Three kinds of patterns can be extracted

• serial episodes

• serial episode rules having a single event type in the consequent

• quantitative episodes: grouping “homogeneous” occurrences of the serial episodes with respect to the time elapsed between the event types (patterns called quantitative episodes and defined in [2])

Support constraints (minimal occurrence semantics)

• minimum number of occurrences

• minimum number of sequences in which the pattern must occur

Temporal constraints

• maximum window size

• minimum gap

• a kind of maximum gap (not the standard max gap constraint used for sequential patterns)

Syntactic constraints

• minimum pattern length

• maximum pattern length

• last event type of the pattern (for episode rules this allows to set the event type of the consequent, to discard the other rules)

• prefix of the pattern (with wildcard place holder)

Some other options

• a threshold to discard event types that are too frequent

• for rules: minimum confidence

• for groups of occurrences (quantitative episodes): parameters to define the homogeneity of the groups and their minimum size

• several input and output formats (the input can be a single long sequence or a set of sequences)

• output of occurrence locations

• and more …

[1]

Discovery of frequent episodes in event sequences. Mannila, H.,Toivonen, H. and Verkamo, A.I. DMKD Journal, volume 1, pp. 259-289, 1997.

[2]

Extracting Trees of Quantitative Serial Episodes. M. Nanni and C. Rigotti. Knowledge Discovery in Inductive Databases 5th International Workshop KDID‘06 Revised Selected and Invited Papers. LNCS 4747, pp. 170-188, 2007.