Usage



In [1]:

    
from episode_miner import EventText, EventSequences, Episode, Episodes
from estnltk.taggers import EventTagger
from IPython.display import HTML, FileLink

Find from the texts



In [2]:

    
event_vocabulary = [{'term': 'üks'}, 
                    {'term': 'kaks'}]    
event_tagger = EventTagger(event_vocabulary, case_sensitive=False, return_layer=True)

event_text1 = EventText('Üks kaks kolm neli kolm. Kaks üks kaks kolm neli kolm üks kaks.', event_tagger=event_tagger)
event_text2 = EventText('Kaks üks kaks kolm neli kolm üks kaks.', event_tagger=event_tagger)
texts = [event_text1, event_text2]
event_sequences = EventSequences(event_texts=texts, classificator='term', time_scale='start')
html = event_sequences.pretty_print()
HTML(html)









    Out[2]:







    
        
        
        PrettyPrinter
    
    
    

		Üks kaks kolm neli kolm. Kaks üks kaks kolm neli kolm üks kaks.
		Kaks üks kaks kolm neli kolm üks kaks.

frequent serial episodes which consist of words üks and kaks.

Find frequent episodes

Let the width of the Winepi window be 31 characters and the minimal relative support of serial episodes be 30%.

Find frequent serial episodes.



In [3]:

    
frequent_episodes = event_sequences.find_serial_episodes(window_width=31, 
                                                         min_frequency=0.3, 
                                                         only_full_windows=False, 
                                                         allow_intermediate_events=True)
list(zip(frequent_episodes, frequent_episodes.abs_support(), frequent_episodes.rel_support()))









    Out[3]:





[(('üks'), 140, 0.8695652173913043),
 (('kaks'), 149, 0.9254658385093167),
 (('üks', 'kaks'), 129, 0.8012422360248447),
 (('kaks', 'üks'), 70, 0.43478260869565216),
 (('kaks', 'kaks'), 67, 0.4161490683229814),
 (('kaks', 'üks', 'kaks'), 58, 0.36024844720496896)]

It turns out that the episode ('kaks', 'üks', 'kaks') appears in 58 Winepi windows. Since the length of the first and the second text is 63 and 38 characters, respctively, the total number of winepi windows is 63+31-1+38+31-1=161. Therefore the relative frequency of this episode is 58 / 161 = 36%.

Write the results to file.



In [4]:

    
frequent_episodes.to_json(file='data/episodes.txt')
FileLink('data/episodes.txt')









    Out[4]:




data/episodes.txt

If file==None (the default), then to_json returns the corresponding string.

Find episode examples

Find examples for the previously found frequent episodes.



In [5]:

    
event_sequences.find_episode_examples(frequent_episodes, 
                                      window_width=31,
                                      allow_intermediate_events=True,
                                      number_of_examples='ALL')

Write the results to file.



In [6]:

    
frequent_episodes.examples_to_json(file='data/episode_examples.txt')
FileLink('data/episode_examples.txt')









    Out[6]:




data/episode_examples.txt

The lines of the file of examples correspond to the lines of the file of episodes.

If file==None (the default), then examples_to_json returns the corresponding string.

Choose the frequent episode ('kaks', 'üks', 'kaks') and pretty print the examples:



In [7]:

    
HTML(frequent_episodes[5].examples_pretty_print())









    Out[7]:







    
        
        
        PrettyPrinter
    
    
    

		Üks kaks kolm neli kolm. Kaks üks kaks kolm neli kolm üks kaks.
		Üks kaks kolm neli kolm. Kaks üks kaks kolm neli kolm üks kaks.
		Üks kaks kolm neli kolm. Kaks üks kaks kolm neli kolm üks kaks.
		Kaks üks kaks kolm neli kolm üks kaks.
		Kaks üks kaks kolm neli kolm üks kaks.

Find the support

Find the absolute and the relative support (frequency) for the episodes (üks, kaks, üks) and (kaks, kaks, üks). It is more efficient to find the support for a list of episodes, than for each episode separately.



In [8]:

    
episode1 = Episode(('üks', 'kaks', 'üks'))
episode2 = Episode(('kaks', 'kaks', 'üks'))
episodes = Episodes([episode1, episode2])
event_sequences.support(episodes=episodes,
                        window_width=31,
                        only_full_windows=False, 
                        allow_intermediate_events=True)
episodes.abs_support(), episodes.rel_support()









    Out[8]:





([15, 9], [0.09316770186335403, 0.055900621118012424])