mlpccg/Clustering.py

   1 from sklearn.cluster import AffinityPropagation
   2 from sklearn.feature_extraction import DictVectorizer
   3 from collections import defaultdict
   4
   5 from mlpccg.CardDb import CARDDB
   6 from mlpccg.DeckList import DeckList
   7
   8 class Clustering:
   9     def __init__(self, records):
  10         self.decks = []
  11         self.features = []
  12         self.placements = {}
  13         self.clusters = defaultdict(list)
  14         self.all_ids = CARDDB._by_id.keys()
  15         self.labels = []
  16         self.records = records
  17         self.vectorizer = DictVectorizer()
  18
  19         for record in self.records:
  20             self.features += [self.extract_features_alt(record['decklist'])]
  21             self.decks += [record['decklist']]
  22             self.placements[record['decklist']] = int(record['placement'])
  23
  24         X = self.vectorizer.fit_transform(self.features).toarray()
  25         self.af = AffinityPropagation().fit(X)
  26         self.labels = self.af.labels_
  27
  28         self.n_clusters = len(set(self.labels)) - (1 if -1 in self.labels else 0)
  29
  30         for i, deck in enumerate(self.decks):
  31             self.clusters[self.labels[i]] += [deck]
  32
  33     def extract_features_alt(self, deck):
  34         features = dict(zip(self.all_ids, [0] * len(self.all_ids)))
  35         for card in deck.cards:
  36             features[card['id'].lower()] += 1
  37
  38         return features
  39
  40     def extract_features(self, deck):
  41         features = {
  42             'Blue': 0, 'Yellow': 0, 'Purple': 0, 'White': 0, 'Orange': 0, 'Pink': 0,
  43             'Friend': 0, 'Event': 0, 'Resource': 0, 'Troublemaker': 0, 'Mane': 0, 'Problem': 0,
  44             'Power': 0.0, 'Control': 0}
  45
  46         for card in deck.cards:
  47             if card['type'] == 'Friend' or card['type'] == 'Mane':
  48                 features[card['color']] += 1
  49
  50             features[card['type']] += 1
  51
  52             if card.has_key('power'):
  53                 features['Power'] += card['power']
  54
  55             if card.has_key('control'):
  56                 features['Control'] += card['control']
  57
  58             features['Power'] = features['Power'] / len(deck.cards)
  59
  60         return features
  61
  62     def predict(self, decklist):
  63         X = self.vectorizer.fit_transform([self.extract_features_alt(decklist)]).toarray()
  64         return self.af.predict(X)[0]
  65
  66     def ranking(self):
  67         ranking = []
  68         cluster_placements = defaultdict(dict)
  69         for i, deck in enumerate(self.decks):
  70             stats = cluster_placements[self.labels[i]]
  71             try:
  72                 stats['sum'] += self.placements[deck]
  73                 stats['num'] += 1
  74             except KeyError:
  75                 stats['label'] = self.labels[i]
  76                 stats['sum'] = self.placements[deck]
  77                 stats['num'] = 1
  78
  79             stats['avg'] = stats['sum'] / float(stats['num'])
  80
  81         for stats in sorted(cluster_placements.values(), key=lambda x: x['avg']):
  82             ranking += [(stats['avg'], stats['label'])]
  83
  84         return ranking