episode_mining

An inplementation of sequential patterm mining method [1].

How to use

1. Set event sequence and episodes

   sequence = sorted([
       (31, 'E'), (32, 'D'), (33, 'F'), (35, 'A'), (37, 'B'), (38, 'C'), (39, 'E'),
       (40, 'F'), (42, 'C'), (44, 'D'), (46, 'B'), (47, 'A'), (48, 'D'), (50, 'C'),
       (53, 'E'), (54, 'F'), (55, 'C'), (57, 'B'), (58, 'E'), (59, 'A'), (60, 'E'),
       (61, 'C'), (62, 'F'), (65, 'A'), (67, 'D'),
   ], key=lambda x:x[0])
   
   episodes = sorted(['A', 'B', 'C', 'D', 'E', 'F', 'AA', 'AB', 'EF', 'CD',])

2. Initialize WINEPI class

   >>> from episode_mining.winepi import WINEPI
   >>> w = WINEPI(sequence, episodes, 'parallel')
   # to mine serial episodes, set 'serial' insted of 'parallel'

3. Discover frequent (parallel) episodes

   # discover_frequent_episodes(t_s, t_e, win, min_fr):
   #    t_s    : start time of target sequence
   #    t_e    : end time of target sequence
   #    win    : window size
   #    min_fr : threshold of frequency of episodes
   >>> w.discover_frequent_episodes(29, 68, 5, 0.1)
   [<ParallelEpisode: A / 0.46511627907>,
    <ParallelEpisode: B / 0.348837209302>,
    <ParallelEpisode: C / 0.558139534884>,
    <ParallelEpisode: D / 0.441860465116>,
    <ParallelEpisode: E / 0.511627906977>,
    <ParallelEpisode: F / 0.46511627907>,
    <ParallelEpisode: A B / 0.232558139535>,
    <ParallelEpisode: C D / 0.139534883721>,
    <ParallelEpisode: E F / 0.348837209302>]

4. Generate rules

   # generate_rules(t_s, t_e, win, min_fr, min_conf)
   #    t_s      : start time of target sequence
   #    t_e      : end time of target sequence
   #    win      : window size
   #    min_fr   : threshold of frequency of episodes
   #    min_conf : threshold of confidence of rules
   >>> w.generate_rules(29, 68, 5, 0.1, 0.1)
   [<Rule: A -> A B / 0.5>,
    <Rule: B -> A B / 0.666666666667>,
    <Rule: C -> C D / 0.25>,
    <Rule: D -> C D / 0.315789473684>,
    <Rule: E -> E F / 0.681818181818>,
    <Rule: F -> E F / 0.75>]

TODO

• Implement MINEPI method

Reference

1. H. Mannila, H. Toivonen, and A. I. Verkamo, “Discovery of Frequent Episodes in Event Sequences,” Data Min. Knowl. Discov., vol. 1, no. 3, pp. 259–289, 1997.